brain_3_3_ok


25 

 
 

AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of  
Brain-Like Artificial Intelligence 

 

Rosemarie Velik 

CTR Carinthian Tech Research, 
Europastraße 4/1, A-9524 Villach/St. Magdalen, Austria 

velik.rosi@gmx.at 
 

Abstract 
The general objective of Artificial Intelligence (AI) is to make machines – particularly 

computers – do things that require intelligence when done by humans. In the last 60 years, AI has 
significantly progressed and today forms an important part of industry and technology. However, 
despite the many successes, fundamental questions concerning the creation of human-level 
intelligence in machines still remain open and will probably not be answerable when continuing on 
the current, mainly mathematic-algorithmically-guided path of AI. With the novel discipline of 
Brain-Like Artificial Intelligence, one potential way out of this dilemma has been suggested. Brain-
Like AI aims at analyzing and deciphering the working mechanisms of the brain and translating this 
knowledge into implementable AI architectures with the objective to develop in this way more 
efficient, flexible, and capable technical systems This article aims at giving a review about this 
young and still heterogeneous and dynamic research field.  
 

Keywords: Brain-Like Artificial Intelligence, Artificial Brains, Human-Level Machine 
Intelligence 

 
1. Introduction 
The research field of Artificial Intelligence (AI) is concerned with making machines – 

particularly computers – do things that require intelligence when done by humans. In the 60 years 
of its existence, it has celebrated dramatic successes and equally dramatic failures. Today, AI has 
become an important and essential part of technology and industry and provides solutions to some 
of the most complex problems in computer science. Nevertheless, in terms of its original goal – to 
create true human-level intelligence in machines – Strong AI has not succeeded yet and perhaps 
never will. Today, AI researchers are able to create computers that can perform jobs that are 
difficult for persons like logic, algebra problem solving, path planning, or playing chess. However, 
they are still struggling with developing a computer that is capable of carrying out tasks that are 
simple to do for humans like perceiving their environment, assessing complex situations, and taking 
everyday life decisions. Approaches in the past have mainly concentrated on creating intelligence in 
computational devices by developing programs that exhibit some kind of “behavior” or “skill” that 
resembles specific facets of human (or animal) behavior or skills. Investigating the structures, 
information processing principles, and functions in the brain that lead to the emergence of such 
behavior and skills was so far generally out of scope of AI technology. For this reason, today's 
computers and AI programs have simply very little in common with actual brains and minds. 
Unfortunately, if continuing this mainly mathematically-algorithmically oriented path, any major 
breakthrough in AI is rather improbable. A paradigm shift might be necessary. With the novel 
research field of Brain-Like Artificial Intelligence, one potential way out of this dilemma has been 
suggested [111, 137]. However, the discipline of Brain-Like Artificial Intelligence is still not 
broadly recognized and so far a common research community has not established. Researchers 
influencing this domain are in fact mostly situated on the borders of other research fields. So far, no 
standard textbooks or overview articles exist concisely defining the field and explaining the 
objectives, methodologies, difficulties, etc. Those who want to gain insight into this domain have to 
laboriously gather up and put together information themselves from bits widely scattered through 

BRAINStorming 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 26 

literature. These circumstances of course do not contribute to an acceleration of the developments in 
this field. To overcome this problem, this article aims to provide a first comprehensive review on 
this young domain of science. Although it might happen that certain other researchers of this so far 
disparate and scattered field will disagree on particular points described in this general overview or 
claim additional material, such a first draft (or chapter) of a “manifesto” for Brain-Like AI can 
certainly constitute a basis for fruitful discussions and contribute to further development. 

The review starts with a brief discussion of the field of Artificial Intelligence in general and 
of important current sub-disciplines and their flaws in Chapter 2 in order to better understand the 
backgrounds that led to the emergence of the domain of Brain-Like Artificial Intelligence. 
Afterwards, Chapters 3 to 8 focus on the description of the field of Brain-Like Artificial 
Intelligence. Chapter 3 outlines the basic concept and the general aims of Brain-Like AI. Chapter 4 
lists examples of conferences, journals, individual researchers, and research groups that have started 
to promote and support the idea of brain-inspired AI design. In Chapter 5, a demarcation of the 
current concepts to artificial neural networks – one of the most prominent ancestors of Brain-Like 
AI – is made. Afterwards, a classification scheme for approaches so far presented is proposed in 
Chapter 6. Next, current problems, stumbling blocks, and challenges of the field are summarized in 
Chapter 7. Last but not least, a route for how to overcome these problems in the future is suggested 
in Chapter 8. 

 
2. The field of artificial intelligence today and relevant sub-disciplines 
„Artificial Intelligence (AI) is the science of making machines do things that would require 

intelligence if done by humans” [83]. The basic claim of AI is that the central property of (human) 
intelligence can be precisely described and thus simulated by a machine [78]. Over the last 60 years, 
AI has changed from a computational discipline into a highly interdisciplinary field incorporating 
many areas [128]. Artificial Intelligence has given rise to optimism [65, 115], but has also suffered 
severe setbacks [4, 137]. Today, it constitutes an essential part of technology and industry and is 
providing solutions to some of the most complex problems in computer science [65]. Although there 
does not seem to exist a formal, generally accepted definition for Artificial Intelligence, various 
suggestions for specifying the field of AI (or rather particular sub-domains) have been made over 
the years. John McCarthy, who introduced the term in 1956 [29], defines it as „the science and 
engineering of making intelligent machines” [78]. More recent AI textbooks describe the domain 
amongst others as „the study and design of intelligent agents” [98] where an intelligent agent is a 
system that „perceives its environment and takes actions that maximize its chances of success” 
[108]. Today, AI research is highly specialized and deeply divided into sub-domains that have little 
in common with each other and therefore often fail to communicate [79]. Sub-branches have grown 
up around the work of individual researchers, particular institutions, the solution of specific 
problems, the application of widely differing tools, and longstanding differences of opinion about by 
what methodology AI should or can be achieved. In the following, a suggestion for a classification 
of these approaches into five different groups is given (Section 2.1) followed by a discussion of the 
strengths and weaknesses of each group (Section 2.2). 

 
2.1. A possible classification of existing approaches into sub-disciplines 
Providing a generally accepted taxonomy for classifying existing approaches into sub-

disciplines of AI is not an easy task. Due to the heterogeneity of the field and the overall lack of 
interchange between groups with different ideologies, there seems to exist more confusion than 
consensus about this topic. Part of the researchers propose a distinction according to either the 
particular method used (rule-based systems, neural networks, statistics, fuzzy logics, genetic 
algorithms, ontologies, etc.) or the application (pattern recognition, optimization, planning, theorem 
proving, machine vision, language processing, reasoning, etc.). However, in practice, this leads to 
lists of AI classes easily containing several hundred items. If a researcher does not find an 
appropriate entry for his/her research, he/she is tempted to introduce a new term, which does not 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   27 

really contribute to a resolution of the categorization problem. Therefore, it is suggested here to 
better distinguish approaches based to their ideology, methodology, and goal. Accordingly, a 
classification of the five AI sub-disciplines as described below is proposed, including a description 
of the core essence of each of these domains. The first three sub-disciplines (Applied Artificial 
Intelligence, Artificial General Intelligence, and Embodied Artificial Intelligence) are the ones that 
are the most broadly known today. The further two sub-disciplines (Bio-Inspired Artificial 
Intelligence and Brain-Like Artificial Intelligence) have been less often explicitly mentioned in 
literature, are however relevant for this article.  

When reading these descriptions, one has to be aware that for Artificial Intelligence in 
general as well as for the existing sub-disciplines, different, partly not complementary definitions 
can be found in literature for one and the same term. One example of this are the terms Strong AI 
and Weak AI, which were originally introduced by J. Searle [110] as part of his Chinese room 
argument to distinguish between two different hypotheses about Artificial Intelligence. According to 
the Strong AI hypothesis, a suitably designed computer system can actually replicate aspects of 
human intelligence and thus have a mind. According to the Weak AI hypothesis, a computer system 
can (just) simulate aspects of human intelligence and thus (only) act as if it had a mind. This initial 
usage of those terms is fundamentally different from the way they are widely used today in 
academic AI research and textbooks. Nowadays, these terms are often rather used to distinguish two 
particular AI sub-domains (see below) instead of two hypotheses about the achievable goal of AI. 
Furthermore, one has to be aware that boundaries between the different suggested sub-disciplines 
can in some cases be a bit blurry. Being able to provide a perfect, unambiguous taxonomy in a field 
like AI is probably an illusion. It can thus happen that a particular approach could theoretically be 
assigned to more than one sub-discipline. For instance, a neural network approach could principally 
be regarded as a bio-inspired (maybe even slightly brain-inspired) approach. If the focus is however 
rather on employing a “pre-fabricated” network type to solve a very particular task (e.g., the 
classification of metallic parts of different shapes [133]), it should better be considered as a method 
of Applied AI. Thus, looking at the objectives, methodology, and followed ideology of a research 
approach is important for achieving the most adequate categorization. 

 
• Applied Artificial Intelligence: One potent sub-field of AI, also referred to as mainstream 

AI by some authors, is Applied Artificial Intelligence [27] (also called Narrow Artificial 
Intelligence [46] or Weak Artificial Intelligence [17]). Applied AI is concerned with the use 
of hardware and software to study or accomplish specific tasks of problem solving or 
reasoning that do not encompass the complete range of human cognitive abilities or are in 
most cases even completely outside their focus. Applied AI researchers aim at creating 
programs that demonstrate “intelligence” in a specialized area such as medical diagnosis, 
chess playing, algebraic calculation, mathematical theorem-proving, automobile-driving, etc. 

• Artificial General Intelligence: A second sub-domain of AI is Artificial General 
Intelligence (AGI) [46]. AGI is not a precisely defined term. P. Wang and B. Goertzel [146] 
report that in contrast to Applied AI, which aims at particular aspects of applications of 
intelligence, AGI aims at “intelligence” as a whole, having many aspects and being applied 
in many situations. R. Koene [61] points out that “some AGI researchers are explicitly 
pursuing forms of (general) intelligence designed from first principles and without a desire 

for comparability or compatibility with human intelligence”. However, he also states that 
“many of the underlying objectives that drive the search for AGI also involve an interest in 

anthropomorphic interpretations of intelligent behavior”. In the proceedings of the third 
International Conference on Artificial General Intelligence in 2010, AGI is described as the 
research field focusing on the original and ultimate goal of AI – to create broad human-like 
and trans-human intelligence by exploring all available paths, including theoretical and 
experimental computer science, cognitive science, neuroscience, and innovative 
interdisciplinary methodologies [12]. Some researchers place the term AGI today more or 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 28 

less on a level with the term Strong Artificial Intelligence [27]. In contrast to its original use, 
Strong AI now refers to the design of machines of human intelligence level, i.e., machines 
that can successfully perform any intellectual task a human being can do [65]. C. Pennachin 
and B. Goertzel [94] point out that the term Artificial General Intelligence (AGI) was 
introduced in order to avoid the connotations “human-level” or “human-like”' of the term 
Strong AI, which were widely used before, and to further allow the inclusion of non-human 
models. 

• Embodied Artificial Intelligence: A third important sub-discipline of AI is Embodied 
Artificial Intelligence [97] (also called Embodied Intelligence [18], New Artificial 
Intelligence [87], Behavior-Based Artificial Intelligence [76], or Nouvelle Artificial 
Intelligence [96]). The goal of Embodied Artificial Intelligence is to study “how intelligence 
emerges as a result of sensorimotor activity, constrained by the physical body and mental 

developmental program” [127]. Accordingly, the field of embodied intelligence is strongly 
connected to the design of “intelligent” agents that perceive their environment and take 
actions that maximize their chances of success [108].  

• Bio-Inspired Artificial Intelligence: Another sub-discipline of AI is the field of Bio-
Inspired Artificial Intelligence [39] (also mentioned in correlation with Biologically-Inspired 
Artificial Intelligence [25], Naturally-Inspired Artificial Intelligence [13], Bionics [132], and 
Artificial Life [14]). Bio-Inspired Artificial Intelligence focuses on the design of intelligent 
machines by taking inspiration from biology. J. Chappell et al. [24] point out that this 
includes “(1) attempts to mimic details of the morphology and behavior of humans or other 
animals; (2) attempts to apply theories about how neural or evolutionary mechanisms work, 

to solve engineering problems; (3) applying techniques inspired by social behaviors of 

organisms, including swarming, flocking, use of pheromone trails, etc.; (4) attempts to 

understand the problems solved by biological evolution in order to clarify goals and 

requirements for AI/robotics.” 

• Brain-Like Artificial Intelligence: A further emerging sub-domain of AI holding great 
promises and being subject of this article is the field of Brain-Like Artificial Intelligence [41, 
111] (also called Brain-Inspired Artificial Intelligence [137]). Brain-Like Artificial 
Intelligence – a term that is currently being established [111] – is concerned with 
investigating how the human (or animal) brain is structured and how it functions with the 
aim to emulate these concepts for the design of more efficient, effective, and “intelligent” 
technical systems [137, 73]. 
 
2.2. Weak points of current sub-disciplines 
In the approximately 60 years since its development and implementation, research in 

Artificial Intelligence has produced many useful outputs in terms of solved technical problems and 
has also contributed insights to our understanding of the “nature of intelligence”. Nevertheless, the 
overall challenge of developing “truly intelligent machines” applicable for all kinds of tasks that can 
be solved by humans is enormous and much remains to be done to overcome it. Different sub-
disciplines of AI have targeted this challenge from different perspectives. Brilliant researchers can 
be found in each of them. Nevertheless, each field today still shows certain “intrinsic” weak points. 
An attempt to identify and describe them is made in the following. 

 
Applied Artificial Intelligence 
Although there exist initiatives to encourage people to focus on the advancement of the 

scientific understanding of cognitive systems1, researchers today choose to focus for the most part 
on Applied AI [94], i.e., specific sub-problems such as data mining, computer vision, logistics, 
speech recognition, medical diagnostics, banking software, industrial robotics, search engines, etc., 

                                                 
1
 http://www.socsci.ru.nl/CogSys2/PDFs/Cogsys2\_Opening\_by\_Colette\_Maloney.pdf 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   29 

where they can produce commercial applications and verifiable results [90, 108]. Applied AI can 
therefore be considered as AI mainstream research direction. Some of the methods and programs 
developed in this context are extremely successful in what they do [94]. Some AI researchers today 
still hope that more general purpose AI systems can emerge by combining the programs that solve 
various sub-problems using an integrated subsumption architecture, agent architecture, or cognitive 
architecture [44, 86]. However, the success of achieving a general level of intelligence by such 
“superposition approaches” is heavily doubted by other AI researchers. In an interview in Hal's 
Legacy in 2000, M. Minsky, one of the pioneers and co-founders of AI, pointed out that “we really 
haven't progressed too far toward a truly intelligent machine. We have collections of dumb 

specialists in small domains; the true majesty of general intelligence still awaits our attack. We 

have got to get back to the deepest questions of AI and general intelligence and quit wasting time on 

little projects that don't contribute to the main goal.” [45] 
 
Artificial General Intelligence 
Excessive optimism in the 1950s and 1960s concerning Artificial General Intelligence – 

although at this time this name was not used – has given way to the recognition of the extreme 
difficulty of creating human-level intelligence, which is today considered as the possibly hardest 
problem that science has ever undertaken. Thus, in the following decades, most AI researchers have 
devoted little attention to Artificial General Intelligence. Nevertheless, today, a small number of 
scientists are active in AGI research. The research is extremely diverse and frequently pioneering in 
nature. Estimates of the time needed before a truly flexible AGI system could be built, if at all, vary 
from a decade to over a century [94]. R. Kurzweil and colleagues point out that a timeframe 
between 2015 and 2056 is plausible [65]. However, many AI researchers of other sub-disciplines are 
not at all convinced that progress will be that fast. Critics doubt whether research in the next 
decades will even produce a system with the overall intellectual ability of an ant [27]. 

 
Embodied Artificial Intelligence 
While Applied AI and Artificial General Intelligence have been centered round the idea of 

the computational representation of the world, the approach of Embodied AI is centered around 
action in the world. Embodied Intelligence suggests that the basic assumption of traditional AI, 
according to which cognition is equal to computation, is not adequate and needs to be reconsidered. 
When the digital computer was invented, many felt that the core and essence of thinking and 
intelligence had been found. Computers were considered “electronic brains” in those days and the 
human mind was viewed as a computer [96]. In the 1980s and 1990s, a subset of researchers from 
Artificial Intelligence, psychology, cognitive science, and brain science eventually realized that the 
idea of computers as intelligent machines, at least in the form they exist today, was misguided. They 
concluded that the brain does not run “programs” and has not evolved to perform mathematical 
proofs2 but to control behavior that ensures the organism’s survival in the real world [96]. 
Accordingly, they further concluded that intelligence needs a body equipped with sensors and 
actuators and an environment on which to act. Based on this claim, a new sub-discipline of AI 
evolved: the field of Embodied Artificial Intelligence [18]. Embodied Intelligence has moved its 
focus away from computers and towards more interactive systems such as robots or ubiquitous 
environments equipped with many sensors and actuators. It focuses not necessarily on human 
intelligence but rather on non-human biological systems such as insects [97]. Important research 
areas in Embodied AI today include walking, locomotion, low-level sensory-motor intelligence, 
orientation behavior, path-finding, elementary behaviors such as following and obstacle avoidance, 
etc. [97]. 

                                                 
2
 In this context it should be mentioned that other researchers partly disagree, as certain animals and particularly 

humans need cognitive capabilities that allow for planning (acting-as-if), for instance, in order to cope with dangerous 
situations where a solution cannot be worked out by trial and error as this would most likely result in getting injured or 
killed [28,125]. 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 30 

Although usually not directly related to each other in scientific literature [92], Embodied 
Artificial Intelligence seems to have much in common with the domain of Cybernetics [57], which 
already emerged approximately a decade before the field of AI was founded [149]. Cybernetics was 
particularly interested in the question of how “animals and humans maintained equilibrium within, 
and responded appropriately to, their ever-changing environment” [70]. 

Today, the field of Embodied Artificial Intelligence is still highly diverse and far from being 
theoretically stable [97]. Interestingly, many researchers of Embodied Intelligence do not refer to 
themselves as working in Artificial Intelligence but rather in robotics, engineering of adaptive 
systems, artificial life, adaptive locomotion, bio-inspired systems, and neuroinformatics [97]. In 
Embodied AI, the identification of the fact that natural intelligent systems are all biological and 
have to perceive and interact with their environment was without a doubt an important step. 
However, unfortunately, this approach offered new pitfalls. One essential criticism of Embodied AI 
is that the slogans of embodiment encouraged researchers to focus exclusively on “on-line” 
interaction of robots with the environment (see for example [101]) without “reflecting” about their 
activities (e.g., to consider what they did before, why they did it, what would have happened if they 
had not done it, whether they should try something different next time, etc.). R. Pfeifer and F. Iida 
[97] point out that “to date, most robots are specialized, either for walking or other kinds of 
locomotion purposes, or for sensory-motor manipulation, but rarely are they skilled at performing a 

wide spectrum of tasks. One of the fundamental unresolved problems has been and still is how 

thinking emerges from an embodied system. Proactively speaking, the central issue could be 

captured by the question: How does walking relate to thinking?” Thus, what happened in the field 
of Embodied AI was that most researchers failed to consider “disconnected” cognition not directly 
involving sensorimotor coordination as much as researchers in earlier AI failed to consider 
“connected” (online) cognition, though there was more justification for the omissions in the early 
days of AI due to still inadequate technology. In this context, A. Sloman [124] mentions that “the 
tautology that a robot that acts and perceives in the world must be embodied is often combined with 

false premises, such as the premise that a particular type of body is a requirement for intelligence, 

or for human intelligence, or the premise that all cognition is concerned with sensorimotor 

interactions, or the premises that all cognition is implemented in a dynamical system closely 

coupled with sensors and effectors”. He further criticizes that “robots so far produced by „nouvelle 
AI“ are pre-constructed and typically move around rather pointlessly (possibly doing something 

useful as a side-effect), or achieve arbitrary goals imposed from outside, as opposed to doing what 

is required to grow, maintain, or reproduce themselves”. J. Starzyk [128] points out that 
“Embodied Intelligence revived the field of autonomous robots, but as robotics thrived, research on 

Embodied Intelligence started to concentrate on the commercial aspects of robots with a lot of effort 

spent on embodiment and a little on intelligence. Researchers today concentrate on advancements 

of robotic technology like building artificial skin and muscles. While this might be relevant for the 

development of robots, it diverts attention from developing intelligence. Furthermore, currently, 

process efficiency dominates over generality. It is more cost efficient and cheaper to design a robot 

for a specific task than one that can perform various tasks, and most certainly even more than one 

that can actually make its own decision”.  
 
Bio-Inspired Artificial Intelligence 
Parallel to the emergence of Embodied Intelligence, more attention also started to be directed 

towards the derivation of biological concepts [97]. While Artificial Intelligence has traditionally 
focused on the computational reproduction of the behavior and abilities of humans, Bio-Inspired AI 
takes inspirations from a wider range of biological structures that show the capability of adaptation 
and autonomous self-organization. N. Forbes [38] describes the focus of the field to be on “the use 
of biology in algorithm construction, the study of how biological systems communicate and process 

information, and the development of information processing systems that use biological materials, 

are based on biological models, or both”. As outlined in [39, 75, 11], Bio-Inspired AI envisions 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   31 

amongst others evolutionary computation, evolutionary electronics, biologically inspired hardware, 
bioelectronics, amorphous computing, artificial immune systems, genetic algorithms, DNA 
computation, cellular automata, biomolecular self-assembly, artificial neural networks, biorobotics, 
and swarm intelligence. In the book “Bio-Inspired Artificial Intelligence – Theories, Methods, and 
Technologies”, D. Charles et al. [25] categorize Biologically-Inspired approaches to AI into 7 
groups: evolutionary systems, cellular systems, neural systems, developmental systems, immune 
systems, behavioral systems, and collective systems. Like Embodied AI, Bio-Inspired AI also has 
some common roots with the field of Cybernetics, which is interested in human and animal nervous 
systems [70]. However, the central aim of Cybernetics is the understanding with less emphasis on 
the construction of useful artifacts. Without doubt, Bio-Inspired AI and bio-inspired technology 
have great potential. However, approaches suggested in this field are of great diversity, which 
makes it difficult to provide an analysis of the strengths and flaws of this domain with general 
validity. While some approaches like artificial neural networks and genetic algorithms are already 
well established methods that find application in various areas, many others are still on the level of 
very basic research and concept clarification. Bio-Inspired AI is not yet widely considered as an 
own, separate research field, but rather it summarizes biologically-inspired approaches introduced in 
diverse other disciplines. In general, the term “bio-inspired” implies that inspiration for technical 
development is taken from nature. However, the level of abstraction at which nature serves as 
archetype varies greatly. Part of the approaches currently consider and investigate information 
processing and computational principles occurring in nature while others analyze biological 
organisms on a behavioral level or take nature more as a metaphor than as a model [38]. In the latter 
context, B. Sendhoff et al. [111] criticize that so far, the contribution of so called “Bio-Inspired” AI 
approaches to a better understanding of “biological intelligence” and “brain intelligence” has been 
very limited. Similar as in classical AI, there exist a number of approaches that have resulted from a 
symbiotic cooperation between engineers, computer scientists, and researchers from life sciences. 
However, a significant number of models are still developed by researchers from the technical 
science discipline only. A. Schierwagen [109] comments that in practice, this has sometimes led to 
so-called bio-inspired technologies that take advantages of fancy biological names as selling 
arguments without an in-depth exploitation of biological principles. 

 
Artificial Intelligence as a Whole 
In summary, it can be concluded that throughout its history, Artificial Intelligence has 

always been a topic of much controversy [29, 90]. Taking inventory today, Artificial Intelligence 
can achieve good solutions that can compete with or exceed human (or at least higher developed 
animal) cognitive abilities and skills for problems that are sufficiently well structured and whose 
complexity can be controlled. Today's AI systems particularly outperform humans in jobs that are of 
an “abstract nature” like logic, solving puzzles and algebra problems, path planning, playing chess, 
etc. and that can be well described in a mathematic or algorithmic form and thus be relatively easily 
implemented computationally (see also [97]). Certain problems have these characteristics. However, 
natural environments do not [111]. AI researchers have so far struggled to develop computational 
systems that are capable of carrying out tasks that are simple and “natural” for humans, e.g., 
perception in the real world, object manipulation, sensory-motor coordination, locomotion, common 
sense, evaluating complex situations, making decisions in real world situations, everyday natural 
language, etc. [96, 137]. J. Starzyk [128] criticizes that Artificial Intelligence often applies the 
principle of chip design to create a system with no intelligence and call it intelligent as long as it 
does its job. During its history, Artificial Intelligence challenges have experienced a revival in 
various disciplines related to AI, e.g., computational intelligence [98], cognitive automation [129], 
cognitive computation [55], and cognitive robotics [26]. However, the basic aims and difficulties 
have remained mainly intact.  

Despite the many successes of Artificial Intelligence, fundamental problems of intelligence 
are still unanswered today [128]. As B. Sendhoff et al. [111, 112] point out: “Superior functionality 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 32 

in a single problem domain can often already be achieved by specialized technological systems 

(thinking about a simple calculator or a chess program), however, no technological system can 

robustly cope with the plethora of tasks encountered by any (higher developed) animal during its 
life time. We have to acknowledge that in spite of its tremendous advances in areas like neural 

networks, computational and artificial intelligence, and fuzzy systems, existing approaches are still 

not able to mimic even for instance the lower-level sensory capabilities of humans or animals.”  
Thus, the open question remains: How to continue on the path to machine intelligence for the 

problem domains that cannot yet be handled with existing approaches? An emerging path that – if 
followed wisely – could hold a solution to this question is the field of Brain-Like Artificial 
Intelligence. 

 
3. The basic concept and aims of Brain-Like Artificial Intelligence 
 
3.1. Basic concept 
As outlined in Section 2.2, Artificial Intelligence today mainly provides methods and tools to 

automate specific, well-structured tasks that take place in environments of limited complexity. For 
tasks that show these characteristics, technical systems can compete with or even surpass human 
performance. However, there does not currently exist a technical system that can even come close to 
competing with the overall capacity and the capabilities of the human mind. The same holds true for 
perceptual, motor control, and other cognitive abilities of higher developed animals. In order to 
substantially progress further a paradigm shift might be necessary in this field. With Brain-Like AI, 
a potential way out of the current AI dilemma has been proposed. Its basic idea is highly intuitive 
[35, 95, 99, 135]: use the brain as archetype for AI model development! However – considering the 
last 60 years of AI history – it is quite clear that it has so far not been that easy to implement this 
idea properly (see also Chapter 7).  

One difficulty with current approaches in Brain-Like Artificial Intelligence is, as it can also 
be observed in other sub-disciplines of AI, that different researchers have partly divergent opinions 
on what Brain-Like Artificial Intelligence actually is and how it should be done (see also Chapter 7). 
In the book Creating Brain-Like Intelligence: From Basic Principles to Complex Intelligent 
Systems, B. Sendhoff et al. [111] point out: “As the name suggests, we would like to achieve 
intelligence as demonstrated by brains preferably those of highly evolved creatures.” Although part 
of the contributions presented in this book actually aim at investigating and emulating the structures, 
information processing principles, and functioning of the brain that lead to the emergence of certain 
skills and abilities, it is left open if an analogy to the brain is sufficient in terms of externally emitted 
functions or if it should also be in terms of internal processes and principles. However, if limiting 
the similarity to emitted functions, no clear demarcation line to the other fields of AI discussed in 
Section 2 – particularly to AGI – can be drawn. Brain-Like Artificial Intelligence as understood in 
this article therefore refers to both similarity in internal brain processes and principles and their 
emitted function. While principally the brain of different higher developed organisms could be 
analyzed for this purpose (see also Section 6.1), the focus is on the human brain in this article. 
Accordingly, the “basic dogma” of Brain-Like Artificial Intelligence as used here is the following:  

 
“Basic Dogma” of Brain-Like Artificial Intelligence as Used Here 
It is well appreciated that the human brain is the most sophisticated, powerful, efficient, effective, 

flexible and intelligent information processing system known. Therefore, the functioning of the 

human brain, its structural organization, and information processing principles should be used as 

archetype for designing artificial intelligent systems instead of just emulating its behavior in a black 

box manner. To achieve this, approaches should not build on work from engineers only but on a 

close cooperation between engineers and brain scientists. 

 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   33 

Acknowledging the simplicity, clarity, and intuitive logic of this basic dogma, the question 
arises why such a brain-like approach to AI has not already been followed more intensively from the 
very beginning of the field of AI. One main reason for this is that 60 years ago, when the domain of 
AI started to evolve, the available body of knowledge about the functioning of the brain was still 
very limited. Apart from simple nerve cell models, the brain could merely be considered as a black 
box system of which only the external behavior could be observed and studied. The development of 
new measuring and imaging techniques in the last decades has shed new light on the internal 
structures, information processing principles, and functioning of the brain. In this context, B. 
Goertzel [45] states: “Prior to the last few decades, the traditional disciplines of psychology and 
neuroscience offered extremely little to AI designers. Since the emergence of cognitive science and 

cognitive neuroscience, things have gotten a little better.” It might now be time to start using and 
extending this acquired knowledge by translating it to functioning and technically implementable 
models. 

 
3.2. Basic aims 
In simple words, the research field of Brain-Like AI is concerned with the development and 

implementation of concepts and models of the human (or animal) brain and their application in 
various kinds of technical systems and domains. The motivation for research in Brain-Like AI is 
twofold and outlined in the following [41, 48, 99]:  

 
• On the one hand, these models and their implementation shall in the future lead to better 

information processing, computation, and automation systems. Domains which are eagerly 
waiting for such novel “intelligent” approaches in order to deal with their upcoming needs 
are, amongst others, the field of machine perception [130, 138], sensor fusion [135], ambient 
intelligence [19], interactive systems [20, 140], smart buildings [67, 136, 139], automation 
[143], robotics [22, 32, 33], agent technologies [68], information processing [50, 131], 
computation [6, 142], and web intelligence [151].  

• On the other hand, the models shall contribute to a better understanding of the human (or 
animal) brain. When trying to technically implement models of the brain, unknown points 
cannot just be left open without even mentioning them as the developed machine has to be 
able to perform certain functions. Therefore, this emulation process is an important means to 
detect weak points and inconsistencies in current brain theories and to contribute to the 
development of new brain theories [82, 138, 141, 144]. 
 
Of course, each researcher will place their particular focus in terms of these two basic 

motivations. While some concentrate on topics of the first point, others may emphasize the second. 
 

4. Support from the scientific community 
Although still being a heterogeneous and emerging discipline, the principal idea of brain-

inspired approaches to solutions in AI is today well supported by numerous researchers. N. Zhoung 
[151] points out that “AI research has not produced major breakthrough recently due to a lack of 
understanding of human brains and natural intelligence.” He predicts that the next major 
advancements in AI will most likely result “by an in-depth understanding of human intelligence and 
its application in the design and implementation of systems with human-level intelligence”. M. 
Minsky suggests that findings about how natural intelligence (particularly the brain) works have to 
be the basis for developing concepts for technical approaches trying to achieve intelligence [134]. J. 
Kirchmar and G. Edelman [63] state that “the analogy to the natural brain must be taken serious” 
[112]. C. Pennachin and B. Goertzel [94] comment that “one almost sure way to create Artificial 
General Intelligence would be to exactly copy the human brain”. C. Goerick [42] discusses: “We 
would like to create brain-like intelligent systems, hence we have to understand how the brain 

works. The artifacts we are creating should show what we have understood from the brain so far, 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 34 

and should help formulating the next questions aiming for further understanding the brain.” R. 
Koene [61] indicates that “the brain's implementation is not necessarily the best one according to 
criteria used to measure performance at solving a particular problem, but at least it is an existing 

implementation, and we have some idea of the specifications that it meets”. D. Dietrich and his team 
emphasize that one way to solve the dilemma of AI and automation in general is the application of 
bionic approaches, particularly of approaches considering the human brain and mind. They 
recommend a close cooperation between engineers and brain scientists for this purpose [21, 35, 68, 
91, 100, 104, 134, 150]. J. Hawkins [51] states that “brains are totally unlike the relatively simple 
structures of even the most complex computer chip” and that therefore “the way we build computers 
today won't take us down the path to create truly intelligent machines.” S. Potter [99] points out that 
“yet, little attention in the AI field has been directed toward actual brains.” He illustrates that “the 
human brain is the best example of intelligence known, with unsurpassed ability for complex, real-

time interaction with a dynamic world” and concludes that “if AI were to become less artificial, 
more brain-like, it might come closer to accomplishing the feats routinely carried out by natural 

intelligence.” S. Amari [6] indicates that “the more is known about the functionality of the brain the 
more intelligent the information systems will become”. B. Goertzel [45] points out that “we have a 
great deal of respect for the fact that the human brain/mind actually exists and functions, so it 

would be foolish not to learn from it all that we can.” 
Within the last few years, the first conferences and workshops promoting the topic of Brain-

Like AI have taken place, e.g., the Engineering-Neuro-Psychoanalysis Forum (2007)3, the Brain-
Inspired Cognitive Systems Conference (2004–2012)4, the Conference on Brain-Inspired 
Information Technology (2006)5, the International Symposium on Creating Brain-Like Intelligence 
(2007)6, the International Conference on Brain Informatics (2009–2012)7, the AAAI Spring 
Symposium (2006)8 with the title Between a Rock and a Hard Place: Cognitive Science Principles 
Meet AI-Hard Problems, and the International Conference on Artificial Neural Networks (2011)9 
with focus on Machine Learning re-inspired by Brain and Cognition. Other conferences have 
dedicated tracks and special sessions to this topic, e.g., the INDIN (2008–2010)10 on Cognitive and 
Computational Intelligence in Industrial Informatics, the HIS (2010) on Modeling the Mind11, and 
the Africon (2011)12 on Challenges in Autonomic Systems looking for New Bionic Approaches for 
Modeling the Mind. 

Similarly, some pioneer journals have started to deal with topics of Brain-Like AI like the 
journal of Broad Research on Artificial Intelligence and Neuroscience13 or the special issue on 
Brain-Inspired Cognitive Agents of the ACM Transactions on Intelligent Systems and Technology14. 

Moreover, the first books have been published promoting the topic of “brain-inspiredness”. 
Among these are the books Simulating the Mind: A Technical Neuropsychoanalytical Approach 
[35], Creating Brain-Like Intelligence: From Basic Principles to Complex Intelligent Systems [111], 
and Brain-Like Computing and Intelligent Information Systems [6]. 

                                                 
3
 http://www.indin2007.org/enf/index.php 

4
 http://www.conscious-robots.com/en/publications/conferences/bics-2010.-brain-inspired-cognitive-systems-

confe.html 
5
 http://www.elsevier.com/wps/find/bookdescription.authors/712521/description 

6
 http://wiredshelf.com/book/creating-brain-like-intelligence-276694 

7
 http://wi-consortium.org/conferences/amtbi11/amtbi.php?conf=bi\&here=cfpart 

8
 http://www.aaai.org/Press/Reports/Symposia/Spring/ss-06-02.php 

9
 http://www.aaai.org/Symposia/Fall/fss08symposia.php\#fs04 

10
 http://indin2010.ist.osaka-u.ac.jp/ 

11
 http://hsi.wsiz.rzeszow.pl/ 

12
 http://www.africon2011.co.za/ 

13
 http://www.edusoft.ro/brain/index.php/brain 

14
 http://tist.acm.org/CFPs/TIST-SI-BICA-10.pdf 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   35 

 
5. Artificial neural networks – the ancestors of Brain-like AI? 
When claiming that Brain-Like Artificial Intelligence is a novel and recent research field, 

one question that will most certainly pop up is: “Is the concept really that new?” The next question 
that is most likely to follow is: “What about artificial neural networks?” These are justified 
questions, which will be considered in more detail in the following. 

In principle, the drive to study and emulate the functions of the (human) brain might date 
back thousands of years. However, only with the development of modern day electronics have 
attempts to emulate the brain and thinking processes become more tangible. In the 1940s, the area of 
neural network research began to emerge [52, 80]. Since then, artificial neural network (ANN) 
research has gone through various highs and lows. Today, artificial neural networks are used for 
pattern recognition, classification, and prediction. Without doubt, ANNs are useful tools and have 
brought solutions to various problems that are difficult to solve with conventional computers. 
However, they have yet failed to produce any kind of behavior that could be considered truly 
intelligent. Several researchers, for instance H. De Garis [31], claimed that this is due to the fact that 
current networks were just too small in terms of number of used neurons. However, this explanation 
is most likely insufficient. I. Aleksander [3] indicates: “The point is that these puzzles are not 
puzzles because our neural models are not large enough.” Although ANNs have often been viewed 
as simplified models of neural processing in the brain, today their relation to biological brain 
architectures is questionable. In fact, artificial neural networks do contain some neuron-like 
attributes (connection strengths, inhibition/excitation, etc.) but they overlook many other factors, 
which may be significant to the brain's functioning [103]. J. Hawkins argues that artificial neural 
network research ignores essential properties of the (human) cortex and instead prefers to focus on 
simple models that have been successful at solving simple problems [51]. More specifically, up to 
today, artificial neural networks have mainly been an attempt to emulate processing aspects of 
individual biological neurons, not of whole neural brain networks and not of learning principles in 
such biological networks. Unlike brains that seem to have already a great deal of organization 
present even during fetal development, ANNs generally start from a jumbled, randomized set of 
connection strengths [103]. The methods for achieving networking through learning in ANNs are 
based mainly on mathematical models, not on biological ones.  

Accordingly, the question to what extent ANNs are actually brain-inspired can be answered 
as follows. Artificial neural networks, as they exist today, are systems that are loosely patterned on 
the nervous system. Their basic processing elements can be considered to be brain-inspired (or 
better neuron-inspired) and the principal idea to put them together to form networks. However, the 
rest of the framework, i.e., the way how artificial neurons are structured into networks, how their 
interconnections are determined through learning processes, etc. is based on mere mathematical 
assumptions. Therefore, it is suggested here that artificial neural networks should rather be classified 
as an approach of Bio-Inspired AI or even Applied AI than one of Brain-Like AI. Without question, 
the initial idea behind ANNs is brilliant and it is astonishing how many tasks can today be solved 
with them. Unfortunately, these promising results in some areas were probably also the reason that 
have led a number of researchers to come to the inappropriate conclusion that ANNs as they exist 
today have much in common with brains. In fact, their success has to be mainly attributed to the 
mathematical theory behind them and not to biological similarities with the brain. Nevertheless, it is 
not impossible that in the future, further developments in this field based on novel and extended 
insights from brain research will lead to neural network structures that are actually worthy to be 
called “brain-like”. 

 
6. Possible classifications of Brain-like AI approaches 
As outlined before, the principal idea behind Brain-Like Artificial Intelligence has existed 

for several decades. However, so far, many stumbling blocks and detours had to be taken 
concerning its implementation. Thus, Brain-Inspired Artificial Intelligence can still be considered a 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 36 

quite novel, dynamic, and heterogeneous field. Different researchers have different opinions 
regarding by what means and how closely one needs to simulate the brain in order to achieve brain-
like behavior [73].  

On the one extreme, there are AI approaches that are based purely on abstract mathematical 
theories of intelligence and do not draw any mapping between natural and artificial cognition at any 
level. Such approaches, which rest on the assumption of vast levels of computing power greater than 
anything that will be physically possible in a foreseeable future, have clearly nothing to do with 
Brain-Inspired Artificial Intelligence as described and defined in this article.  

On the other extreme, there are claims, e.g., by R. Kurzweil [64], that the most effective path 
to Strong AI is to reverse-engineer the human brain to a level as precise as necessary to replicate its 
function. In practice, this has led to “brain simulations” aiming at a level of precision as detailed as 
the cellular level – e.g., in the Blue Brain project [77] – or even the molecular level – as in the Cajal 
Blue Brain project15. 

In sum, regardless of the level of detail of simulation, most approaches suggested so far 
should be considered as rather loosely brain-like models [41]. Looking at the great variety of 
existing approaches, finding a good classification scheme for them is not an easy task and category 
boundaries generally remain vague. Nevertheless, in the following sub-sections, several possible 
ways for a classification are suggested. 
 

6.1. Biological organism used as blueprint 
Traditionally, Artificial Intelligence has focused its attention mainly on the cognitive 

abilities of humans. The roots for this go back as far as to R. Descartes who stated that only humans 
are capable of thought. He considered animal behavior as a series of unthinking mechanical 
responses to stimuli that originate from the animal's internal or external environment [59]. In 
contrast, many researchers today disagree with this view – at least for higher developed animals. For 
instance H. Terrace [15] claims that “animals do think but cannot master language”. A. Sloman 
goes even further and attributes animals with some kind of mathematical competences. Specifically, 
his research interests are related to the biological basis of mathematical competence in humans and 
creative problem solving in other animals [126]. S. Kak [59] points out that “as other animals learn 
concepts nonverbally, it is hard for humans, as verbal animals, to determine their concepts”. 

Based on these facts, Brain-Like Artificial Intelligence aiming at the emulation of brain 
functions cannot only take inspiration from the human brain but principally also from brain 
architectures of all kinds of higher developed animals. Thus, the biological species which is taken as 
inspiration for the blueprint of a brain-like architecture constitutes one possible way of 
classification. Prominent species for such brain and behavior models have so far ranged from 
elephants, cats, cheetahs, deer, primates, and squirrels over crows, pigeons, and parrots to octopuses 
and insects. Nevertheless, as A. Sloman pointed out in a private discussion with him, one has to be 
aware that no matter which organism's brain we take as archetype, we cannot avoid explicitly or 
implicitly focusing on human brains, or at least their competences. After all, we have (at least 
initially) to use human ways of thinking when we ask what other organisms can and cannot do.  

 
6.2. Bottom-up versus top-down design 
Another possibility of categorization is – similar as also in mainstream AI – a division into 

bottom-up (neural) models and top-down (symbolic) models [27]. Generally, in top-down AI, 
cognition is treated as a high-level phenomenon that is independent of the low-level details of the 
implementing mechanism. Researchers in bottom-up AI take an opposite approach and simulate 
networks of neurons. They then investigate what aspects of cognition can be recreated in these 
networks.  

 

                                                 
15

 http://cajalbbp.cesvima.upm.es/ 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   37 

6.3. Involved discipline of brain science 
A further method of classification would be the discipline of brain science that served as the 

inspiration for the technical model. Disciplines today involved in this task range from neuroscience 
over psychology and pedagogics to psychoanalysis. While neuroscience can be considered as a 
“bottom-level” approach, the further disciplines target the problem of creating Artificial Intelligence 
from a “top-level” perspective. Furthermore, in the last decades, disciplines started to emerge, which 
aim at investigating both top and bottom-level aspects of cognition and particularly their 
correlations. Amongst these are neuro-psychology, neuro-psychoanalysis, and cognitive 
neuroscience.  

Although the methodology of classification based on the underlying brain theory would 
make sense, it does not seem to have yet found its way into scientific literature. One problem with 
this kind of classification might certainly be that today, the content and borders of each of these 
disciplines of brain sciences and cognitive sciences are not clearly defined. Each discipline is again 
split into further sub-disciplines, which show partial overlap but partial contradiction with other sub-
areas. None of the disciplines can so far provide a satisfactory explanation for how the brain works 
and accordingly cannot yet give a detailed prescription for the construction of truly “intelligent” 
systems [45] (see also Chapter 7).  

 
6.4. Emulated brain function 
Another possible way of classification can be done according to the brain function that is 

emulated. This can range from vision, audition, multimodal perception, and motor control over 
language processing, decision-making, reasoning, and planning, to emotions, learning, and 
consciousness. In sum, the effort spent so far in truly brain-like approaches to these challenges is 
still quite limited. The field which has probably received the most attention is computer and 
machine vision, particularly for the lower processing levels. Interestingly, also in brain sciences, the 
lower levels of vision are one of the best explored systems of the brain [Nishimoto2011]. In this 
work, the focus is particularly on the development and implementation of models for multimodal 
perception, “emotional” situation assessment, and decision-making and their interplay. 

 
6.5. Large-scale brain simulations versus (brain-inspired) cognitive architectures 
A methodology related to the bottom-up and top-down categorization of Section 6.2 but 

perhaps more contemporary for a distinction of brain-inspired approaches is suggested by H. Garis 
et al. [41] and B. Goertzel et al. [48]. This taxonomy will be described more in detail in the 
following. The authors distinguish between large-scale brain simulations and (biologically inspired) 
cognitive architectures

16. In the following, a description of these two classes of approaches is given 
including a number of examples of so far proposed models. For this purpose, the review articles [48] 
and [41] will serve as the main references. This description does not at this point include a judgment 
to which extent the individual mentioned models are actually brain-inspired. A critical reflection 
about the usage of the term “brain-inspired” and the problems coming along with it is given in 
Chapter 7. 

 
Large-scale brain simulations 
Large-scale brain simulations describe systems that attempt to simulate in software the 

structure and dynamics of particular subsystems of the brain. When looking at existing approaches 
to large-scale brain simulations, it becomes clear that the scope of interpretation of the notion “brain 
simulation” is broad. Different researches are approaching this task with very different objectives in 
mind. H. Garis et al. [41] distinguish between five groups ordering them according to decreasing 
neurobiological fidelity: 

                                                 
16

 The term “biologically inspired” as used by H. Garis and B. Goertzel mainly refers to architectures inspired by the 
brain and shall therefore be described under the term brain-inspired cognitive architectures in the following. 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 38 

 
1. Models, which can actually be connected to parts of the human brain or body and that can serve 

the same role as the brain system they emulate. An example is Boahen's artificial retina [16]. 
2. Precise functional simulations of subsystems of the brain, their internal dynamics, and their 

mapping of inputs to outputs that explain exactly what the brain sub-system does to control the 
organism.  

3. Models, which quantitatively simulate the generic behavior and internal dynamics of certain 
brain sub-systems, however, without precisely functionally simulating these sub-systems. The 
best known examples of this type are Markram's cortical models promoted within the Blue 
Brain Project [77] and its continuation, the Human Brain Project17. Furthermore, such 
approaches are amongst others targeted by Izhikevich's simulations [56] and Edelman’s brain-
based devices [62]. 

4. Models, which quantitatively simulate subsystems of the brain or whole brains at a high level, 
not including a simulation of the particular details of dynamics or inputs/outputs. Such 
approaches have the aim to explore some of the overall properties of the given system. 
Examples are Just's CAPS work [122] or Horwitz's cognitive simulations [54]. 

5. Models, which demonstrate the capacity of hardware to simulate large neural models. These 
models are based on particular classes of equations. However, there is no claim about the match 
of the models with empirical neuroscience data. Examples are Boahen's Neurogrid work [16], 
Modha's cat brain simulation [7], or S. Furber's SpiNNaker architecture [60]. 

 
All of the abovementioned approaches are validly called large-scale brain simulations but 

constitute very different forms of research of which not all might be suitable for Brain-Like AI 
design. According to H. Garis, the first two categories are adequate to serve as components of brain-
inspired cognitive architectures or other AI systems. Simulations in the third and fourth category are 
useful for guiding neuroscience or hardware development but are not directly useful for AI. Finally, 
the fifth category is not directly useful, neither for neuroscience nor AI, but it serves as “proof of 
principle” with the motivation to lead on to other, more directly useful work in later steps. 

H. Garis points out that at present state of technology, large-scale brain simulations have 
proven to be useful mainly for refining equations used for neural and synaptic modeling and for 
helping to substantiate conceptual models of brain structures and functions by connecting these 
models with detailed electrophysiological data in working simulations. They inform us about the 
dynamics of brain regions and allow us to probe the complexity that underlies structures such as 
cortical columns and the emergent nonlinear coherence that arises when large numbers of neurons 
are appropriately coupled. Nevertheless, detailed brain simulations do not yet yield intelligent 
behaviors not only due to a lack of processing power of (still inadequate) hardware for simulation 
but more importantly due to a lack of knowledge of the intermediate-scale structure of the brain, so 
as to be able to encode it into simulations. Furthermore, an embodiment of brain signals seems to be 
important. However, this aspect has been seldom considered so far in brain simulations. 

H. Garis et al. [41] indicate that in case large-scale brain simulation research programs were 
successful, the biggest benefits are still lying ahead including the followings:  

 
1. Gathering and testing neuroscience data  
2. Cracking the neural code  
3. Understanding neocortical information processing  
4. A novel tool for drug discovery for brain disorders 
5. A foundation for whole-brain simulations  
6. A foundation for human-like Artificial General Intelligence 

 

                                                 
17

 http://www.humanbrainproject.eu/ 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   39 

Garis at al. [41] point out that it is not yet clear whether large-scale brain simulations will be 
the first approach to lead to success in human-like intelligence, but it is certainly a plausible and 
promising approach and different researchers in this field have set their long-term sights specifically 
in this direction. 
 
 

(Brain-inspired) cognitive architectures 
(Brain-inspired) cognitive architectures attempt to achieve brain-like functionalities by 

emulating the brain's high-level architecture without necessarily simulating its lower-level specifics 
[48]. The term brain-inspired (or biologically-inspired) cognitive architectures became common 
only within the last years through the DARPA funding program administered by the Information 
Processing Technology Office. Nevertheless, the basic concept of cognitive architectures itself is as 
old as the field of Artificial Intelligence. While there is no rigid demarcation that separates brain-
inspired cognitive architectures from cognitive architectures in general, the term is intended to 
distinguish cognitive architectures drawing significant direct inspiration from the brain and those 
based exclusively (or nearly exclusively) on models of the mind [48]. While brain simulations are 
intended to display not only similar functions to a brain, but also closely similar internal structures 
and dynamics, brain-inspired cognitive architectures are mainly intended to display (still loosely) 
similar functions to a brain. They furthermore aim at displaying internal structures that are 
conceptually inspired by the brain (and not just the mind) but not necessarily extremely similar to it. 
In practice, drawing a connection between the organization and the activation patterns of the brain 
and functions of the mind that emerge from them is certainly still a highly challenging task. Thus, 
the “loosely similar functions to the brain” are in many cases still extremely loose and it is doubtful 
if these approaches should actually be labeled as “brain-inspired cognitive architectures” or rather 
simply as “cognitive architectures”. In their survey of cognitive architectures, W. Duch at al. [36, 
48] divide existing cognitive architectures into three paradigms:  

 
1. Symbolic Architectures: Symbolic paradigms follow a long tradition in AI and are based on 

the assumption that the mind exists mainly to manipulate symbols that represent aspects of the 
world or themselves [93]. Examples for symbolic architectures are amongst many others SOAR 
[66], EPIC [105], ICARUS [69], NARS [145], or SNePS [113]. In [48], they are considered to 
be psychology-inspired cognitive architectures instead of brain-inspired approaches. While 
symbolic architectures contain many valuable ideas and have yielded some interesting results, it 
is doubtful whether they will give rise to the emergent structures and dynamics required for 
human-like general intelligence.  

2. Emergentist Architectures: Emergentist cognitive architectures consider abstract symbolic 
processing to result and emerge from lower-level subsymbolic dynamics, which are in most 
cases biologically inspired up to a certain extent. They are designed to simulate neural networks 
or other aspects of human brain function [48]. A few examples for emergentist architectures are 
HTM [51], DeSTIN [9], IBCA [102], Cortronics [117], or NOMAND [37]. Furthermore, there 
exists a set of emergentist architectures focused specifically on developmental robotics [74] to 
control robots without significant „hard-wiring“ of knowledge and capabilities. Here, robots 
learn via their engagement with the world [48]. Examples for existing architectures are DAV 
[49], SAIL [148], FLOWERS [11], and IM-CLEVER [81]. Similar to classical AI connectionist 
systems, emergentist cognitive architectures have proven to be strong in recognizing patterns in 
high-dimensional data, reinforcement learning, and associative memory. However, probably 
due to their so far only quite superficial analogy to brain circuits, they have not yet led to the 
emergence of any brain-like functions [48].  

3. Hybrid Architectures: Due to the complementary strengths and weaknesses of symbolic and 
emergentist approaches, researchers have started recently to focus on integrative, hybrid 
architectures. These combine subsystems operating according to the two different paradigms 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 40 

described above [48]. Combinations have been suggested in many different ways. Examples are 
the connection of a large symbolic system with a large subsymbolic system or the creation for 
populations of small agents, each of which is both symbolic and subsymbolic in nature [48]. 
Examples for hybrid cognitive architectures are CLARION [120], DUAL [88], LIDA [40], 
ACT-R [71], MicroPsi [10], Polyscheme [23], 4CAPS [58], Shruti [114], the Novamente AI 
Engine [43], 4D/RCS [2], or OpenCogPrime [47]. In [89], N. Nilsson claims that “AI systems 
that achieve human-level intelligence will involve a combination of symbolic and non-symbolic 

processing.”  
 
Résumé 
In summary, as outlined by B. Goertzel at al. in [47], today, “brain simulations tell us about 

cortical columns and about the way collections of neurons “spontaneously” organize into 

collectives, but they do not yet tell us anything specific about how brains achieve goals, select 

actions, or process information. On the other hand, (brain-inspired) cognitive architectures tell us 

how brains may do things, but so far their intelligent behaviors are quite simplistic compared to 

real brains.” They consider three possible future directions that could lead to the emergence of 
human-like Artificial Intelligence:  

 
1. Large-scale brain simulations, which simulate multiple brain regions and their 

interconnections, thus verging on being brain-inspired cognitive architectures.  
2. Brain-inspired cognitive architectures, which integrate more detailed neural dynamics into 

their processing, enabling greater creativity and flexibility of response.  
3. Hybrid architectures that link brain-inspired cognitive architectures elements with brain 

simulation elements. 
 
B. Goertzel at al. give preference to the third option and argue that this will be the most rapid 

approach towards the twofold goal of understanding the brain and emulating human-like 
intelligence in computers. In principle, this approach can be considered as an addressing of the 
problem of emulating the brain by a combined bottom-up neuroscientific and top-down cognitive 
approach. With this, it is hoped to learn more about the crucial intermediate levels of cognition, 
about which the least body of knowledge is available currently [45]. 

 
7. General challenges and stumbling blocks 
Considering the potential benefits of using the brain as an archetype for intelligent system 

design, it is at first glance surprising that so far only relatively limited attention has been paid to this 
approach in the field of Artificial Intelligence in comparison to merely mathematically and 
algorithmically guided concepts. Of course, attempts for the design of Brain-Like AI have existed 
since the very beginning of AI – with its most popular example being artificial neural networks (see 
Chapter 5). However, up to now, inspiration from the brain has mainly been drawn just on an 
abstract level not really justifying the label “brain-like” or “brain-inspired” intelligence [137]. As 
discussed in the following sub-sections, the reasons for this are multiple. While part of them are 
purely scientific challenges, others have a more “socio-political” character based on different views, 
perceptions, and ideologies of different AI researchers. The following points seem obvious once 
they are as explicitly formulated as it is done in this article. However, in the past, the issues AI 
researchers chose to investigate led to considerable confusion and misunderstandings, particularly 
amongst younger researchers having entered the field of AI only once it was already as diverse and 
heterogeneous in its objectives, methods, and ideologies as it is today. I myself at the beginning of 
my scientific career can be included into this group. Thus, outlining the challenges to face and the 
stumbling blocks to overcome can be very valuable for the sake of better clarity and to avoid a 
repetition of errors that could have been avoided by learning from the past. 
 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   41 

7.1. Decipher the complexity of the brain 
The most important reason for the quite limited “brain-inspiredness” of existing AI 

approaches lies most certainly in the complexity of the brain and the difficulty to understand it. 
Existing models in brain sciences are often still only abstract and descriptive instead of a functional 
character and contain blank spots and inconsistencies. T. Deutsch [34] describes the problem as 
follows: “Various sciences have gathered a vast amount of information on the human mind. An 
unmanageable flood of different theories, ideas, and approaches derived from psychology, 

neurology, and philosophy could be used as starting points. The problem is: They are piecemeal – 

no holistic model is provided.” 
What is particularly missing is an understanding of the intermediate levels of cognitions, i.e., 

how neural correlates can actually result in cognitive functions. B. Goertzel [45] formulates this 
problem as follows: “On the one hand, the cognitive sciences provide very clear advice regarding 
what the overall “conceptual architecture” of an Artificial General Intelligence (AGI) system 

should be like ... We know what the major regions of the brain do, and we also have a decent 

working decomposition of human cognition into a list of interacting yet significantly distinct 

faculties. This high-level architecture can be emulated in AGI systems. On the other hand, the 

cognitive sciences provide a variety of suggestions regarding specific low-level mechanisms for 

carrying out intelligent processing, such as perception, learning, and memory. However, the low-

level messages from the cognitive sciences are more controversial than the high-level ones for two 

reasons. First, there is less agreement on them among contemporary experts. And second, it's not 

always clear that emulating human psychological or neural behavior is a practical approach to 

implementing intelligence on radically un-brain-like hardware. Cognitive theorists recognize that 

there is more to the human mind/brain than its high-level architecture and low-level mechanisms. 

However, the cognitive sciences to date have had relatively little to say about the crucial 

“intermediate level” of intelligence. This is the main reason that the cognitive sciences don't yet 

provide really powerful prescriptive guidance to AGI designers. The cognitive sciences tell us what 

major parts a mind/brain should have, and they describe some low-level mechanisms that can help 

these parts to carry out their functions, but they say precious little about how the different parts all 

work together, and how the low-level mechanisms coordinate to give rise to higher-level dynamics.” 
M. Looks and B. Goertzel [73] further indicate: “The best approach to Strong AI at present, 

we suggest, is to learn what we can from the brain about what sort of high level architecture and 

general dynamics and representations are useful for achieving general intelligence under conditions 

of limited computational resources – and then fill in the algorithmic level with representations and 

algorithms that make sense in terms of the mathematics, computer science, and computer hardware 

and software that we know. This attitude leads into an integrative approach to AI, in which one 

takes a general architecture loosely inspired by human cognition, and then uses it to bind together 

components drawn from various areas of mathematics and computer science.” The suggestion to 
fill blank spots with algorithmic solutions is without doubt reasonable and valid to achieve 
functioning systems. However, this approach has to be followed with caution as it has in practice led 
to further problems (see Section 7.5 for further details). In summary, researchers that aim at 
emulating brain functions for technical purposes need, apart from sophisticated skills in engineering, 
a relatively profound understanding of the state of the art knowledge in brain sciences. Apart from 
this, they need the analytic capacity to detect inconsistencies and blank spots in brain theories and 
the ingenuity and creativity to propose solutions how to resolve them. 
 

7.2. Find the right level of abstraction and detail 
A second point crucial to successful brain modeling is the specification of the right level of 

abstraction and detail from which to start. This decision certainly depends on the specific 
requirements of the system in mind (e.g., what processes need to be explained by the developed 
model). Starting with a neuroscientifically inspired bottom-up design (as in large-scale brain 
simulations) on a level with too much detail – e.g., with the modeling of molecular processes in 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 42 

single neurons when trying to understand the general information flow in larger cortex structures – 
bears the danger of becoming lost in unnecessary details not directly relevant for the given problem. 
Similarly, starting with a top-down design (as generally in cognitive architectures) and remaining 
with brain analogies on a too high level of abstraction can lead to models that are too loosely 
patterned in regards to the organization and principles of the brain. Such designs tend to contain 
rather simplistic metaphors instead of sound models of the brain/mind. In such approaches, 
researchers are in danger of falling back to classical AI methods and developing solutions of mere 
mathematical and algorithmic character. In this context, O. Sporn [118] comments that “it will be 
futile and scientifically meaningless to attempt to build intelligent systems by slavishly imitating or 

replicating the real brain. On the other hand, the extreme computational functionalism of the 

classical AI movement has done little to advance flexibility and robustness in intelligent systems. 

What is needed is a new synthesis of brain, cognitive and engineering sciences to harness the 

complexity of biological systems for the design of a new generation of more capable brain-like 

intelligence.” Similarly, in his article about Neuroscience and AGI, R. Koene [61] points out that “a 
recurring argument against borrowing from neuroscience in the development of AGI has been to 

note that the low-level design of the brain is very complex, possibly needlessly complex for general 

intelligence. But other difficulties arise when one instead chooses the most obvious alternative 

approach: Observing high-level processes and implement those”. He raises his concerns about the 
strong reliance on vastly simplified models of cognition in AGI: “The aspects of cognition that are 
well-explained by the popular cognitive architectures cited in AGI research are based, in part, on 

cherry-picked experiments and corresponding data about human cognitive processes that are the 

easiest ones to characterize.” 
 
7.3. Overcome the lack of interdisciplinarity and collaboration 
A further possible reason for the limited progress in Brain-Like AI so far can be found in the 

fact that many AI researchers – who are still mainly computer scientists and engineers – naturally 
lack deeper understanding of established brain theories. Obtaining this knowledge, e.g., by regular 
and tight cooperation with brain researchers, could in the future make a huge difference. However, 
setting up fruitful interdisciplinary collaborations which would benefit from synergies still requires 
considerable effort. In his article What Can AI Get from Neuroscience, S. Potter [99] states: “Yet, 
little attention in the AI field has been directed toward actual brains. Although many of the brain's 

operating principles are still mysterious, thousands of neuroscientists are working hard to figure 

them out. Unfortunately, the way neuroscientists conduct their research is often very reductionistic, 

building understanding from the bottom up by small increments. A consequence of this fact is that 

trying to learn, or even keep up with, neuroscience is like trying to drink from a fire hose. General 

principles that could be applied to AI are hard to find within the overwhelming neuroscience 

literature. AI researchers, young and old, might do well to become at least somewhat bilingual.” He 
further suggests: “It's time for AI to move in the brainwards direction. This could involve PhD 
programs that merge AI and neuroscience, journals that seek to unite the two fields, and more 

conferences at which AI researchers and neuroscientists engage in productive dialogs. 

Neuroscientists have not exactly embraced AI either. Both sides need to venture across the divide 

and learn what the other has to offer.” Similarly, B. Sendhoff et al. [111] point out: “New 
principles are usually found at the interfaces between existing disciplines, and traditional 

boundaries between disciplines have to be broken down to see how complex systems become simple 

and how the puzzle can be assembled. Each of us is rooted in a certain community which we have to 

serve with the results of our research. Looking beyond our fields and working at the interfaces 

between established areas of research requires effort and an active process. It is our belief that we 

have to more intensively pursue research approaches that aim at a holistic and embedded view of 

intelligence from many different disciplines and viewpoints.” 
A further fact that contributes to the problem is that researchers who are actually active in 

the field of Brain-Like Artificial Intelligence are currently still located on the borders of other 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   43 

disciplines. So far, they have spent only moderate effort on collaborating amongst each other. To 
catalyze the emergence of the field of Brain-Like AI, it will be necessary that they start to join 
forces. 

 
7.4. Avoid a misapplication of terms 
The domain of Brain-Like Artificial Intelligence is still relatively new and dynamic and only 

beginning to establish. It is up to now no homogenous research field with well-defined borders and 
it must struggle with many difficulties and barriers in addition to the ambitious challenge of 
understanding and emulating the brain based upon incomplete knowledge from neuroscience, 
cognitive sciences, and other disciplines. 

One problem with this field that has led to considerable confusion not only amongst the 
general public but also amongst AI researchers themselves is that there do not exist guidelines or 
rules for when a model may be called “brain-like” or “brain-inspired”. There are no broadly 
accepted definitions that define the attribute “brain-like” and how the attribute can be implemented 
and tested [53]. B. Sendhoff et al. [111] state: “What is Brain-Like Intelligence? Although it seems 
necessary to have a good understanding of what one wants to create before one starts, there is no 

crisp and clear definition.” Unfortunately, the term “brain-like” is therefore sometimes misapplied 
or misused either knowingly or unknowingly. Knowingly, this happens for marketing purposes – as 
the label brain-like (or similar) is often (partly unconsciously) perceived as a synonym for highly 
sophisticated and advanced hardware and software technology amongst the general public. On the 
other hand, in the past, this has also occurred partly unknowingly due to engineers' and computer 
scientists' limited understanding of the difference between the structure and information processing 
principles of the brain and commonly used computer architectures and programs. 

In both cases, this has sometimes led to a mere transfer of terms from brain sciences to AI 
rather than a transfer of sophisticated concepts and principles. T. Deutsch [34] advises caution in 
this context: “One has to be careful when using terms from human sciences. They tend to end as a 
label for something, which has nothing to do with the original concept they described“. Several 
textbooks, articles, and PhD theses written in the past have shown that a conclusive and tight 
merging of biological and technical concepts is not that straightforward. There exists literature that 
starts with a description of biological, neural, and cognitive principles in the first chapters and 
finishes with technical solutions in the last ones without a direct, notable correlation and connection 
between the former and the latter [5]. In other approaches that claim to be brain-inspired, some 
correlations between certain mechanisms in the brain and the developed models can be deduced. 
However, what is still often missing is a clear statement about which parts of the models have taken 
inspiration from the brain and which elements were guided by mathematic-algorithmic methods and 
engineering considerations only [100, 107, 121]. 

 
7.5. Prevent a drive against the traffic 
While Brain-Like AI generally aims at implementing concepts from the brain for the design 

of technical systems, there exists on the other hand a research niche where engineers aim at 
understanding the brain by studying the structure and behavior of common (not brain-inspired) 
computational systems and programs [8]. Although such approaches can in certain cases bring some 
new interesting insights [111], in practice this has on occasion been counterproductive in the past as 
it has misguided researchers who did not have a clear picture of the differences between today's 
computational machines and the brain [137]. It should be noted that – if these computational 
systems and programs were not designed by taking inspiration from nature and the brain/mind – it is 
scientifically “dubious” to draw conclusions about the functioning of the brain from these designs, 
at least, without further consultation with brain scientists and their theories about brain function. L. 
Miller states that AI researchers have to be criticized for this in particular as this has not produced 
theories whose adequacy can be tested by empirical research [82]. M. Looks and B. Goertzel [73] 
formulate this criticism in a more moderate way: “We are neutral as to how directly the brain's 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 44 

neuronal network structure and dynamics relate to its cognitive representations and algorithms, and 

also as to how closely the brain's knowledge representation resembles formal logic and how closely 

its dynamics resemble logical inference. These are very important questions, but neuroscience has 

not yet progressed far enough to give us the answers. No one knows how an abstract proposition 

like “Every boy has a dog that every girl likes to call by a special name” is represented or 

manipulated or learned through experience in the human brain, and until we know this, we won't 

know the extent to which the conceptual premises of the most popular neural net or logic based 

approaches to AI are correct.”  
To be fair, it has to be mentioned that there has started to evolve a research niche called AI-

Inspired Biology [24] where cognitive scientists and biologists have joined forces with AI 
researchers with the aim to illustrate ways in which AI can influence research on natural cognition 
and formulate new research questions in biology and brain sciences. Treating such topics in an 
interdisciplinary team instead of just amongst engineers certainly has higher probabilities of success. 
 

7.6. Tolerate and respect differences in scientific ideology 
Artificial Intelligence is not a homogeneous research domain but actually deeply divided into 

sub-fields with different ideologies, methodologies, and targets. It requires a profound 
understanding and knowledge about the history and developments in AI to gain an awareness (and 
maybe also tolerance) of the differences in approaches and aims of the distinct sub-disciplines. This 
fact is particularly problematic in the field of Brain-Inspired AI. While other AI domains are 
focused on approaches based on mathematic, algorithmic, and computation theories, Brain-Like AI 
aims (or at least should aim) at using the (human) brain as archetype. It therefore has different 
objectives and methodologies and targets an absolutely different level of complexity than the 
mathematic-algorithmic approaches of the more “classical” AI. The problematic of the differences 
in objectives, ideology, and methodology of different AI domains comes into play when a 
researcher of one sub-field evaluates the work of a researcher of the other sub-domain. In practice, it 
seems to happen rather frequently that editors and funding agencies assign research articles or 
scientific research proposals to reviewers with an “opposing ideology” about how research in AI 
should be done. In certain cases, this can lead to a misinterpretation and at times an unjustified 
rejection of scientific work of related but opposing disciplines. This phenomenon is not particularly 
new in AI (and is probably also common in other fields of science). Furthermore, like today, already 
in 1960, the competition for research funding in AI was intense. As pointed out in [70], there have 
thus always been personal animosities between AI researchers of different groups. In 1969, for 
instance, M. Minsky and S. Papert published their famous work on Perceptrons [84] criticizing 
computational models of the nervous system, which showed “with unanswerable mathematical 
arguments that such models were incapable of doing certain important computations” [70]. This 
killed most research in neural computing for the following 15 years. Much later, S. Paper admitted 
in an interview  [70]: “Yes, there was some hostility behind the research reported in Perceptrons ... 
part of the drive came from the fact that funding and research energy was dissipated ... money was 

at stake.” In recent times, these conditions have turned out to be particularly hard for scientists 
working in Brain-Inspired AI. As they are still a minority compared to other AI research 
communities, the statistical probability that their work is judged (and sometimes misjudged) by 
experts of opposing fields is relatively high. Of course, misjudgment of scientific quality of work 
does not only occur in a one way direction from classical AI to Brain-Like AI but also the other way 
around. A. Sloman [111] points out that “lessons learned from and within classical Artificial 
Intelligence remain relevant to the research program of creating brain-like intelligence and the 

reaction against everything related to classical AI may even have held up progress.” However, as 
the field of Brain-like Artificial Intelligent is still very novel and dynamic and can so far only count 
on a small community of researchers compared to traditional AI, it is in this case still a quite 
unequal fight, which can drastically slow down necessary progress.  
 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   45 

7.7. Establish suitable measures for validating progress and success 
As in any other field of engineering and science also in Brain-Like Artificial Intelligence, the 

validation of progress and success is a crucial element [112]. In Brain-Like AI, the aims, 
methodologies, and the level of complexity of problems are different from other fields of AI. 
Accordingly, different evaluation mechanisms for judging the value and success of developed 
approaches have to be found.  

Applied AI tries to achieve some kind of computational behavior that can solve mainly 
circumscribed and limited problems that would need cognitive capabilities when done by humans. 
Brain-Like AI aims at using similar organizational structures and information processing principles 
like the brain to achieve such tasks. For example, it is impossible to isolate the “person-recognition-
brain-circuits” from the “fruit-recognition-circuits” or the circuits for planning a travel route from 
the ones taking a decision about what to eat for breakfast. Thus, Brain-Inspired AI cannot focus on 
isolated problems. It has to aim at solutions targeting human cognitive capabilities at a more general 
and global level. It has to target topics such as human perception, situation assessment, or human 
decision-making in a more holistic way and cannot just focus on singular topics such as “face 
recognition” or “path planning” in a specific context.  

In Applied AI as it focuses on very specific and circumscribed problems, an evaluation and 
comparison of results amongst each other is relatively straightforward. For example, for a particular 
classification problem, the number of true and false positives can be determined and compared in 
different approaches. A qualitative comparison of machine performance with human performance is 
generally out of the scope of the evaluation. 

In contrast, in research fields like Brain-Like AI and also Artificial General Intelligence 
(AGI), things are not that straightforward. In AGI (before Strong AI), which has been around since 
the beginning of AI, the challenge of effectively evaluating performance has been troublesome for 
scientists for decades. Strong AI has the aim of achieving a general level of intelligence instead of 
just measuring the performance of an approach in very specific sub-tasks. A. Adam [1] comments 
that “if the aim of AI is primarily to create an artificial mind, then the success or failure of the 
whole AI project should be judged against this one goal”. S. Legg [72], who provides an extensive 
overview and discussion of suggested methods for measuring biological and artificial intelligence, 
points out that the first fundamental problem for judging artificial intelligence is that “nobody really 
knows what intelligence is”. R. Davis [30] claims that the basic question is not just “What is 
intelligence?” but equally important also “Why is intelligence?” R. Sternberg et al. [119] point out 
that “viewed narrowly, there seem to be almost as many definitions of intelligence as there were 
experts asked to define it”. This problem is certainly “especially acute when we need to consider 
artificial systems” – e.g., the computer – “which are significantly different to humans” [72]. In the 
past, the most prominent approach for evaluating „intelligence“ in the field of AGI (before Strong 
AI) was the Turing test, originally suggested by A. Turing in 1950 [123]. According to the Turing 
test, a machine is “intelligent” if a human judge cannot distinguish its answers from the answers of a 
person in a communication without direct physical contact. However, in later years, the Turing test 
has been heavily criticized [147] and has been subject to further changes and extensions, e.g., the 
total Turing test [106] which also incorporates video signals and the possibility to pass physical 
objects “through the hatch” to allow the interrogator to evaluate the subject's perceptual abilities and 
its “manipulative” abilities. This has led to an extension of earlier efforts using computer vision and 
robotic object manipulation. However, passing such extended Turing tests can just be – if at all – a 
necessary but not a sufficient condition for intelligence [85]. The Turing test is based upon the 
(doubtful) assumption that “once having a sufficiently precise theory of the mind, it becomes 
possible to express this theory in a computer program. If the program's input-output behavior 

matches corresponding human behavior, this is evidence that some of the program's mechanisms 

could also be operating in humans” [106]. One major problem with the whole Turing test is that it 
can just check for an indistinguishability of human behavior and the „behavior“ a program emits in 
a particular situation. However, it is important to note that just because two things are 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 46 

indistinguishable from a certain perspective, they do not at all have to be the same. Judging progress 
and success in AGI thus remains a hotly debated issue. 

Similar to AGI, finding adequate evaluation mechanisms in Brain-Like AI is a puzzling task. 
Like in AGI, the objectives in Brain-Like AI are much more complex than in Applied AI. Therefore, 
it is much more difficult to achieve quantitatively comparable results in the short term. After all, in 
Brain-Like AI – if done properly and without too many “workarounds” – the overall functionality of 
a model only starts to emerge when the whole system starts to sufficiently resemble the 
organizational structures and information processing principles of the brain. Furthermore, a detailed 
comparison of the performance of developed brain-like models with methods used in Applied AI or 
other fields of engineering – as frequently requested from reviewers – is often simply not feasible 
due to the fact that the problems addressed with brain-like models often simply do not match in 
terms of complexity with problems addressed in Applied AI and related domains for the following 
reasons:  

 
• Unequal Performance Comparisons: The aim of Brain-Like AI is usually to provide more 

general, “global” solutions and not only task-specific solutions as normally addressed in 
Applied AI. A brain-like model targets for instance the problem of (machine) perception in 
general [138] and not just the distinction of ten specific faces in a number of video frames or 
single images. Thus, research efforts spent until reaching a more global solution – which also 
needs cooperation with brain-sciences on unresolved questions of brain functioning – are 
certainly much higher. Quantitatively comparing this “general purpose” solution to a “single 
purpose” approach that is highly specialized and optimized to one single task is certainly not a 
fair competition if the evaluation concerns only the task for which the “special purpose 
system” has been optimized. 

• Lack of Systems to which to Compare: Brain-Like Artificial Intelligence targets particularly 
challenges and applications for which today there are still no satisfactory solutions provided 
by current technical approaches. In order to compare a particular brain-inspired approach with 
other purely mathematically/algorithmically-based technical approaches for such a “complex” 
application, it would therefore be necessary to do both:  
1. Develop the brain-inspired approach; 
2. Develop, adapt, and customize purely technical approaches for the given non-trivial 

problem – developments that can, if done properly, themselves easily take several years 
or even decades for each chosen approach; 

This would however need an enormous effort that could only be achieved by combined major 
resource and time investments from various research groups with different backgrounds 
having a common interest in amicable competition. Applying for funding for such an 
undertaking seems to be however outside the current funding frameworks. 

• Different Requirements for Comparison from Different Disciplines: Brain-Like AI faces 
an additional difficulty in comparison to other sub-disciplines of AI and engineering as it is 
very interdisciplinary. B Sendhoff et al. [112] point out: “We are not clear yet, whether we 
shall position ourselves more within the scope of brain science or technology. In the first case, 
success has to be judged by neurophysiological or psychological experimental data as is the 

case in computational neuroscience. In the second case, the target is to provide evidence that 

the realized system accomplishes its intended requirements. Of course, in this case we have 

the initial burden to clearly define what the intended requirements are against which we want 

to judge our progress. The fact that there is a continuous transition between both extreme 

stand-points makes the measurement process even more ambiguous.” 

 
Seeing the difficulties of quantitative comparisons of Brain-Like AI approaches to other 

conventional existing technical approaches and considering the fact that we are at the very 
beginning of the development of the field of Brain-Like AI make it necessary to reach for validation 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   47 

methods that are different from the ones common in more classical fields of engineering and 
Applied AI. As outlined above for the field of AGI which faces partially similar challenges, the 
discussion of how to do this is still ongoing. Nevertheless, the fact that in Brain-Like AI we can 
judge both “brain similarity” and “system accomplishment” – as indicated in point three of the 
itemization above – can actually be considered as an advantage concerning validation criteria in 
comparison to AGI. While AGI usually still has to rely on external observations of emitted 
„intelligent behavior“ only (e.g., the Turing test), which has proven to be problematic, Brain-Like 
AI can additionally count on an analysis of internal similarities of developed models with the brain 
on a neural/cognitive level. Together with the emitted function of the implemented technical system, 
the analogies in organizational structures and information processing principles of the underlying 
model can be compared to the ones of the brain to obtain a judgment of its quality. It is important to 
note that an ideal brain-like model should show both similarities in externally emitted function and 
in internal organization and processing! Such an evaluation becomes particularly interesting if 
learning mechanisms are integrated into a model. If these learning mechanisms were actually 
sufficiently „brain-like“, similar structuring and function principles should evolve after this learning 
as they can be observed in the developed (human) brain. In the long run, such “self-developed” 
similarities could be an ultimate proof for adequate brain-like design. 

 
8. Recommendations for a way to go in future 
As outlined in the last section, the field of Brain-Inspired Artificial Intelligence is not 

completely non-problematic for different reasons. While part of the difficulties are issues of 
“science-policy” and could therefore be addressed by broadening researchers' still sometimes a bit 
too one-sided view on the topic via a promotion of interaction between disciplines, the most severe 
drawback is still the fact that the brain is presently not sufficiently understood on all levels to 
technically emulate its complete function based on existing neuro-cognitive models alone. In the 
future, symbiotic developments of AI researchers together with brain scientists might be able to 
change this fact. However, this process will require time. B. Goertzel [45] describes this problem 
and suggests a possible way out of this dilemma as follows: “Given the current state of the cognitive 
sciences, the present-day Artificial General Intelligence (AGI) designer has several recourses: 

 

• Firstly, he can simply wait until the cognitive sciences advance further, and give more 
thorough prescriptions for AGI design.  

• Secondly, he can ignore the cognitive sciences and attempt to design an AGI on other 
grounds – e.g. based on the mathematics of reasoning, or based on general considerations 

regarding the dynamics of complex self-organizing systems. Of course it's worth reflecting 

that many of these „other grounds“ – such as mathematical logic – were originally 

conceived as qualitative models of human thought. But still, in spite of this historical fact 

and the strong intuitive feelings associated with it, the empirical cognitive sciences have not 

yet substantiated any deep connections between mathematical logic and human cognition.   

• Or, thirdly, he can seek to create an AGI design that is consistent with the information 
provided by the cognitive sciences, but also introduces additional ideas filling in the gaps 

they leave.”  

 
According to the understanding of Brain-Like Artificial Intelligence as followed here, the 

third suggested approach seems to be the most sensible one for the moment and is therefore the one 
advocated in this article. However, to avoid confusion, it should always be transparent which parts 
of the model are actually brain-inspired and which are those that are based on such „technical 
workarounds“. Accordingly, my recommendation for a procedure to follow in the future for the 
development of Brain-Like AI systems is the following: 

 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 48 

1. Use knowledge about the organizational structures and information processing principles of 
the brain as basis for model and system development as far as available.  

2. Based on this, advance the knowledge in brain sciences by reporting detected inconsistencies 
and blank spots in existing brain theories. 

3. In case the emulation of the brain is not possible at a certain point due to blank spots, 
supplementary logical engineering considerations are allowed to create a functioning 
technical system. However, it must always be pointed out clearly what “workarounds” have 
been used and at which place. In fact, such workarounds can in certain cases even enhance 
the knowledge about the brain by bringing up new hypotheses which have been missing in 
the big picture about the functioning of the brain. Nevertheless, one always has to be clearly 
aware of the balancing act of drawing conclusions from such technical solutions back to the 
brain. In the past, they have far too often led to inappropriate claims about the structure and 
functions of the brain. 

4. Validate the model in terms of both achieved functionalities and similarities to internal 
structural organization and processing principles of the brain. 

5. When having proposed a new potentially relevant hypothesis concerning particular brain 
functions, aim at bringing this hypothesis to the attention of the respective discipline of brain 
science (e.g., via peer-reviewed journal publications) with the encouragement to 
experimentally verify it with empirical brain science data. 
 
During the process of increasing the synergies between Artificial Intelligence and brain 

sciences in the way proposed above, a bootstrapping process is likely to occur. Brain research will 
provide better models that can serve as basis for more advanced AI systems. More sophisticated AI 
systems will give brain scientists the tools to make further discoveries and interpret them [99]. 
 

References  
[1] A. Adam. Artificial Knowing: Gender and the Thinking Machine. Psychology Pres, 1998. 
[2] J. Albus, H.M. Huang, E. Messina, K. Murphy, M. Juberts, A. Lacaze, S. Balakirsky, M.O. 
Shneier, T. Hong, H. Scott, et al. 4D/RCS Version 2.0: A reference model architecture for unmanned 
vehicle systems. National Institute of Standards and Technology, Gaithersburg, MD, NISTIR, 6912, 
2002. 
[3] I. Aleksander. Best brain boosts artificial life. BBC News, 1999. 
[4] The ALPAC report. Technical report, Automatic Language Processing Advisory Committee, 
1966. 
[5] S.I. Amari, N.K. Kasabov, and A.P. Assembly. Brain-like Computing and Intelligent 
Information Systems. Springer, 1998. 
[6] S. Amari. Brain-Like Computing and Intelligent Information Systems. Springer, 2002. 
[7] R. Ananthanarayanan and D.S. Modha. Anatomy of a cortical simulator. In ACM/IEEE 
Cconference on Supercomputing, 2007. 
[8] M.A. Arbib. Towards an automata theory of brains. Communications of the ACM, 15(11), 
1972. 
[9] I. Arel, D. Rose, and T. Karnowski. A deep learning architecture comprising homogeneous 
cortical circuits for scalable spatiotemporal pattern inference. In NIPS 2009 Workshop on Deep 
Learning for Speech Recognition and Related Applications, 2009. 
[10] J. Bach. Principles of Synthetic Intelligence: PSI: An Architecture of Motivated Cognition. 
Oxford University Press, USA, 2009. 
[11] A. Baranes and P.Y. Oudeyer. R-IAC: Robust intrinsically motivated active learning. In 
International Conference on Development and Learning, 2009. 
[12] E. Baum, M. Hutter, and E. Kitzelmann, editors. Artificial General Intelligence. Proceedings 
of the Third Conference on Artificial General Intelligence, 2010. 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   49 

[13] J. Beal, P. Bello, N. Cassimatis, M. Coen, and P. Winston. Naturally inspired Artificial 
Intelligence. Technical report, AAAI Fall Symposium, 2006. 
[14] M.A. Bedau, J.S. McCaskill, N.H. Packard, S. Rasmussen, C. Adami, D.G. Green, T. 
Ikegami, K. Kaneko, and T.S. Ray. Open problems in artificial life. Artificial Life, 6:363-376, 2000. 
[15] J.L. Bermudez. Thinking Without Words. Oxford University Press, USA, 2003. 
[16] K. Boahen. Mimic the nervous system with neuromorphic chips. Scientific America, pages 
55-63, 2005. 
[17] N. Boltuc and P. Boltuc. Replication of the hard problem of consciousness in AI and Bio-AI: 
An early conceptual framework. In AI and Consciousness: Theoretical Foundations and Current 
Approaches, 2007. 
[18] R. Brooks. Intelligence without reason. In 12th International Joint Conference on Artificial 
Intelligence, pages 569-595, 1991. 
[19] D. Bruckner, J. Mitterbauer, and R. Velik. Perception and modeling of scenarios in ambient 
automation networks with Hidden Markov models. In 6th International Conference on Ubiquitous 
Robots and Ambient Intelligence, 2009. 
[20] D. Bruckner, R. Velik, and Y. Penya. Machine perception in automation: A call to arms. 
EURASIP Journal on Embedded Systems, 2011. 
[21] W. Burgstaller, R. Lang, P. Poerscht, and R. Velik. Technical model for basic and complex 
emotions. In 5th IEEE International Conference on Industrial Informatics, volume 2, pages 1007-
1012, 2007. 
[22] W. Burgstaller, R. Lang, P. Poerscht, and R. Velik. Technical model for basic and complex 
emotions. In 5th IEEE International Conference on Industrial Informatics, volume 2, pages 1007-
1012, 2007. 
[23] N.L. Cassimatis. Adaptive algorithmic hybrids for human-level Artificial Intelligence. 
Frontiers in Artificial Intelligence and Applications, 157:94, 2007. 
[24] J. Chappell, S. Thorpe, N. Hawes, and A. Sloman, editors. AI Inspired Biology. SSAISB: 
The Society for the Study of Artificial Intelligence and the Simulation of Behaviour, 2010. 
[25] D. Charles, C. Fyfe, D. Livingstone, and S. Mcglinchey. Biologically Inspired Artificial 
Intelligence for Computer Games. IGI Publishing, 2007. 
[26] A. Clark and R. Grush. Towards a cognitive robotics. Adaptive Behavior, 7(1), 1999. 
[27] J. Copeland. Artificial Intelligence: A Philosophical Introduction. Wiley-Blackwell, 1993. 
[27bis] J. Copeland. What is Artificial Intelligence. AlanTuring.net, 2000. 
[28] K.J.W. Craik. The nature of explanation. Cambridge Univ Pr, 1967. 
[29] D. Crevier. AI: The Tumultuous History of the Search for Artificial Intelligence. Basic 
Books, 1993. 
[30] J. Davis. Biological sensor fusion inspires novel system design. In Joint Service Combat 
Identification Systems Conference, 1997. 
[31] H. De Garis. CAM-BRAIN - The evolutionary engineering of a billion neuron artificial brain 
by 2001 which grows/evolves at electronic speeds inside a cellular automata machine (CAM). 
Towards Evolvable Hardware, pages 76-98, 1996. 
[32] T. Deutsch, R. Lang, G. Pratl, E. Brainin, and S. Teicher. Applying psychoanalytic and 
neuro-scientific models to automation. In 2nd IET International Conference on Intelligent 
Environments, volume 1, pages 111-118, 2006. 
[33] T. Deutsch, A. Gruber, R. Lang, and R. Velik. Episodic memory for autonomous agents, 
IEEE Conference on Human System Interactions, pages 621-626 2008. 
[34] T. Deutsch. Human Bionically Inspired Autonomous Agents. PhD thesis, Vienna University 
of Technology, 2011. 
[35] D. Dietrich, G. Fodor, G. Zucker, and D. Bruckner, editors. Simulating the Mind - A 
Technical Neuropsychoanalytical Approach. Springer Wien New York, 2008. 
[36] W. Duch, R.J. Oentaryo, and M. Pasquier. Cognitive architectures: Where do we go from 
here? In Second Conference on AGI, 2008. 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 50 

[37] G.M. Edelman. Neural Darwinism: Selection and Reentrant Signaling in Higher Brain 
Function, volume 10. Neuron, 1993. 
[38] N. Forbes. Imitation of Life: How Biology is Inspiring Computing. Cambridge MA, MIT 
Press, 2004. 
[39] D. Floreano and C. Mattiussi. Bio-Inspired Artificial Intelligence: Theories, Methods, and 
Technologies. The MIT Press, 2008. 
[40] S. Franklin and F.G. Patterson Jr. The LIDA architecture: Adding new modes of learning to 
an intelligent, autonomous, software agent. Integrated Design and Process Technology, 703:762-
1004, 2006. 
[41] H. De Garis, C. Shuo, B. Goertzel, and L. Ruiting. A world survey of artificial brain projects, 
Part I: Large-scale brain simulations. Neurocomputing, 3(29):3-29, 2010. 
[42] C. Goerick. Towards cognitive robotics. Creating Brain-Like Intelligence, pages 192-214, 
2009. 
[43] B. Goertzel. From complexity to creativity. Exploration in Evolutionary, Autopoietic, and 
Cognitive Dynamics, 1997. 
[44] B. Goertzel, M. Looks, and C. Pennachin. Novamente: An integrative architecture for 
Artificial General Intelligence. In Proceedings of AAAI Symposium on Achieving Human-Level 
Intelligence through Integrated Systems and Research, Washington DC, 2004. 
[45] B. Goertzel. A comparison of the Novamente AI design with the human mind-brain. 
Novamente LLC, February 2005. 
[46] B. Goertzel. Patterns, hypergraphs and Embodied General Intelligence. In International Joint 
Conference on Neural Networks, 2006. 
[47] B. Goertzel. Opencog Prime: A cognitive synergy based architecture for embodied Artificial 
General Intelligence. In ICCI-09, 2009. 
[48] B. Goertzel, R. Lian, I. Arel, H. De Garis, and S. Chen. World survey of artificial brain 
projects, Part II: Biologically inspired cognitive architectures. Neurocomputing, 74:30-49, 
December 2010.  
[49] J.D. Han, S.Q. Zeng, K.Y. Tham, M. Badgero, and J.Y. Weng. Dav: A humanoid robot 
platform for autonomous mental development. In 2nd International IEEE Conference on 
Development and Learning, pages 73-81, 2002. 
[50] A. Hanazawa. Brain-Inspired Information Technology. Springer, 2010. 
[51] J. Hawkins and D. George. Hierarchical temporal memory: Concepts, theory and 
terminology. Whitepaper, Numenta Inc, 2006. 
[52] D.O. Hebb. The Organisation of Behaviour. New York: Wiley, 1949. 
[53] C. Herrmann and F. Ohl. Cognitive adequacy in brain-like intelligence. Creating Brain-Like 
Intelligence, pages 314-327, 2009. 
[54] E. Horvitz and M. Barry. Display of information for time-critical decision making. In 11th 
Conference on Uncertainty in Artificial Intelligence, pages 296-305, 1995. 
[55] A. Hussain. Cognitive computation - An introduction. Cognitive Computation, 1(1):1-3, 
2009. 
[56] E.M. Izhikevich and G.M. Edelman. Large-scale model of mammalian thalamocortical 
systems. PNAS, 105:3593-3598, 2008. 
[57] J. Johnston. The Allure of Machinic Life: Cybernetics, Artificial Life, and the New AI. MIT 
Press, 2008. 
[58] M.A. Just and S. Varma. The organization of thinking: What functional brain imaging 
reveals about the neuroarchitecture of complex cognition. Cognitive, Affective, & Behavioral 
Neuroscience, 7(3):153-191, 2007. 
[59] S.C. Kak. Can we define levels of Artificial Intelligence? Journal of Intelligent Systems, 
6:133-144, 1996. 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   51 

[60] M.M. Khan, D.R. Lester, L.A. Plana, A. Rast, X. Jin, E. Painkras, and S.B. Furber. 
SpiNNaker: Mapping neural networks onto a massively-parallel chip multiprocessor. In IEEE World 
Congress on Computational Intelligence, pages 2849-2856, 2008. 
[61] R. Koene. Agi and neuroscience: open sourcing the brain. Artificial General Intelligence, 
pages 401-406, 2011. 
[62] J.L. Krichmar, A.K. Seth, D.A. Nitz, J.G. Fleischer, and G.M. Edelman. Spatial navigation 
and causal analysis in a brain-based device modeling cortical-hippocampal interactions. 
Neuroinformatics, 05:197-222, 2005. 
[63] J.L. Krichmar and G.M. Edelman. Brain-based devices for the study of nervous systems and 
the development of intelligent machines. Artificial Life, 11(1-2):63-77, 2005. 
[64] R. Kurzweil. The Age of Spiritual Machines. Penguin, 2000. 
[65] R. Kurzweil. The Singularity Is Near: When Humans Transcend Biology. Viking Adult, 
2005. 
[66] J.E. Laird, A. Newell, and P.S.S. Rosenbloom. Architecture for general intelligence. 
Artificial Intelligence, 33:1-62, 1987. 
[67] R. Lang, D. Bruckner, R. Velik, and T. Deutsch. Scenario recognition in modern building 
automation. International Journal on Intelligent Systems and Technologies, 4(1):36-44, 2009. 
[68] R. Lang. A Decision Unit for Autonomous Agents Based on the Theory of Psychoanalysis. 
PhD thesis, Vienna University of Technology, 2010. 
[69] P. Langley. An adaptive architecture for physical agents. In IEEE/WIC/ACM Int. Conf. on 
Intelligent Agent Technology, pages 18-25, 2005. 
[70] LearningSpace. Machines, minds and computers. The Open University, 2011. 
http://openlearn.open.ac.uk/mod/oucontent/view.php?id=401236. 
[71] C.Lebiere J.R. Anderson. The Newell test for a theory of cognition. Behavioral & Brain 
Science, 26:587-637, 2003. 
[72] S. Legg and M. Hutter. Universal intelligence: A definition of machine intelligence. Minds 
and Machines, 17(4):391-444, 2007. 
[73] M. Looks and B. Goertzel. Mixing cognitive science concepts with computer science 
algorithms and data structures: An integrative approach to Strong AI. In AAAI Spring Symposium, 
Cognitive Science Principles Meet AI-Hard Problems, 2006. 
[74] M. Lungarella, G. Metta, R. Pfeifer, and G. Sandini. Developmental robotics: A survey. 
Connection Science, 15(4):151-190, 2003. 
[75] W. Maass. Imitation of life: How biology is inspiring computing by Nancy Forbes. Pattern 
Analysis & Applications, 8(4):390-391, 2006. 
[76] P. Maes. From Animals to Animats 2, chapter Behavior-based Artificial Intelligence. The 
MIT Press, 1993. 
[77] H. Markram. Perspectives. Nature Reviews Neuroscience, 7:153-160, 2006. 
[78] J. McCarthy. What is Artificial Intelligence? Stanford University, Computer Science 
Department, 2007. http://www-formal.stanford.edu/jmc/whatisai/whatisai.html. 
[79] P. McCorduck. Machines Who Think. Natick, M.A. and Peters, A.K., 2004. 
[80] W.S. McCulloch and W. Pitts. A logical calculus of the ideas immanent in nervous activity. 
Bulletin of Mathematical Biology, 5(4):115-133, 1943. 
[81] G. Metta, G. Sandini, D. Vernon, L. Natale, and F. Nori. The iCub humanoid robot: An open 
platform for research in embodied cognition. In 8th Workshop on Performance Metrics for 
Intelligent Systems, pages 50-56, 2008. 
[82] L. Miller. Has Artificial Intelligence contributed to an understanding of the human mind? A 
critique of arguments for and against. Cognitive Science, 2:101-127, 1978. 
[83] M. Minsky. Semantic Information Processing. Cambridge, MA: MIT Press, 1968. 
[84] M. Minsky and P. Seymour. Perceptrons: An Introduction to Computational Geometry. The 
MIT Press, Cambridge MA, 1969. 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 52 

[85] J. Moor. The Turing Test: The Elusive Standard of Artificial Intelligence. Springer 
Netherlands, 2003. 
[86] H. Moravec. Mind Children. Harvard University Press, 1988. 
[87] T. Munakata, editor. Fundamentals of the New Artificial Intelligence. Springer, 2008. 
[88] A. Nestor and B. Kokinov. Towards active vision in the dual cognitive architecture. 
Informationtheories, page 9, 2004. 
[89] N. Nilsson. The physical symbol system hypothesis: Status and prospects. 50 Years of 
Artificial Intelligence, pages 9-17, 2007. 
[90] N.J. Nilsson. The Quest for Artificial Intelligence and History of Ideas and Achievements. 
Stanford University, 2010. 
[91] B. Palensky. Introducing Neuro-Psychoanalysis towards the Design of Cognitive and 
Affective Automation Systems. PhD thesis, Vienna University of Technology, 2008. 
[92] P. Pangaro and H. von Foerster. Cybernetics - A definition. Encyclopedia of Computers, 
1990. http://www.pangaro.com/published/cyber-macmillan.html. 
[93] M. Pelikan. Hierarchical Bayesian Optimization Algorithm: Toward a New Generation of 
Evolutionary Algorithms. Springer, 2005. 
[94] C. Pennachin and B. Goertzel. Artificial General Intelligence, chapter Contemporary 
approaches to Artificial General Intelligence, pages 1-28. Springer, 2007. 
[95] L.I. Perlovsky, B. Weijers, and C.W. Mutz. Cognitive foundations for model-based sensor 
fusion. Proceedings of SPIE, 5096:494-501, 2003. 
[96] R. Pfeifer and C. Scheier. Understanding Intelligence. MIT Press, 2001. 
[97] R. Pfeifer and F. Iida. Embodied Articial Intelligence: Trends and challenges. In Lecture 
Notes in Computer Science. SpringerLink, 2004. 
[98] D. Poole, A. Mackworth, and R. Goebel. Computational Intelligence: A Logical Approach. 
New York: Oxford University Press, 1998. 
[99] S.M. Potter. 50 Years of Artificial Intelligence, chapter What Can AI Get from 
Neuroscience? Springer-Verlag Berlin, Heidelberg, 2007. 
[100] G. Pratl. Processing and Symbolization of Ambient Sensor Data. PhD thesis, Vienna 
University of Technology, 2006. 
[101] M. Raibert, K. Blankespoor, G. Nelson, R. Playter, et al. Bigdog, the rough-terrain 
quadruped robot. Proceedings of the 17th International Federation of Automation Control, pages 
10822-10825, 2008. 
[102] R.C. O'Reilly and Y. Munakata. Computational Explorations in Cognitive Neuroscience: 
Understanding of the Mind by Simulating. Cambridge, MA: MIT Press, 2000. 
[103] E. Reingold and J. Nightingale. Artificial Intelligence Tutorial Review. PSY371, 1999. 
[104] C. Rösener. Adaptive Behavior Arbitration for Mobile Service Robots in Building 
Automation. PhD thesis, Vienna University of Technology, 2007. 
[105] J. Rosbe, R.S. Chong, and D.E. Kieras. Modeling with perceptual and memory constraints: 
An EPIC-Soar model of a simplified enroute air traffic control task. SOAR Technology Inc. Report, 
2001. 
[106] S.J. Russell, P. Norvig, J.F. Canny, J.M. Malik, and D.D. Edwards. Artificial Intelligence: A 
Modern Approach, volume 74. Prentice Hall Englewood Cliffs, NJ, 1995. 
[107] G. Russ, D. Dietrich, and C. Tamarit. Situation dependent behavior in building automation. 
In Workshop of EurAsia-ICT Advances in Information and Communication Technology, pages 319-
323, 2002. 
[108] S. Russell and P. Norvig. Artificial Intelligence: A Modern Approach. Upper Saddle River, 
New Jersey: Prentice Hall, 2003. 
[109] A. Schierwagen. The way we get bio-inspired: A critical analysis. In From Brains to 
Systems, 2010. 
[110] J.R. Searle. Minds, brains, and programs. Behavioral and Brain Sciences, 3(03):417-424, 
1980. 



R. Velik - AI Reloaded: Objectives, Potentials, and Challenges of the Novel Field of Brain-Like Artificial Intelligence 
 

 

   53 

[111] Sendhoff, B.; Körner, E.; Sporns, O.; Ritter, H. & Doya, K. Creating Brain-Like Intelligence: 
From Basic Principles to Complex Intelligent Systems, SpringerLink, 2009. 
[112] B. Sendhoff, E. Körner, and O. Sporns. Creating Brain-Like Intelligence: From Basic 
Principles to Complex Intelligent Systems, chapter Creating Brain-Like Intelligence, pages 1-14. 
Springer, 2009. 
[113] S.C. Shapiro, W.J. Rapaport, M. Kandefer, F.L. Johnson, and A. Goldfain. Metacognition in 
SNePS. AI Magazine, 28(1):17, 2007. 
[114] L. Shastri and V. Ajjanagadde. From simple associations to systematic reasoning: A 
connectionist representation of rules, variables and dynamic bindings using temporal synchrony. 
Behavioral and Brain Sciences, 16:417-417, 1993. 
[115] H. Simon. The Shape of Automation for Men and Management. New York: Harper & Row, 
1965. 
[116] M. Sipper. Machine Nature: The Coming Age of Bio-Inspired Computing. McGraw-Hill, 
2002. 
[117] J.F. Sowa. Conceptual Structures: Information Processing in Mind and Machine. Addison-
Wesley, 1984. 
[118] O. Sporns. Creating Brain-Like Intelligence: From Basic Principles to Complex Intelligent 
Systems, chapter From complex networks to intelligent systems, pages 15-30. Springer, 2009. 
[119] R.J. Sternberg, B.E. Conway, J.L. Ketron, and M. Bernstein. People's conceptions of 
intelligence. Journal of Personality and Social Psychology, 41(1):37, 1981. 
[120] R. Sun and F. Alexandre. Connectionist-Symbolic Integration: From Unified to Hybrid 
Approaches. Lawrence Erlbaum, 1997. 
[121] C. Tamarit-Fuertes. Automation System Perception - First Step towards Perceptive 
Awareness. PhD thesis, Vienna University of Technology, 2003. 
[122] R. Thibadeau, M.A. Just, and P.A. Carpenter. A model of the time course and content of 
reading. Cognitive Science, 6:157-203, 1982. 
[123] A.M. Turing. Computing machinery and intelligence. Mind, 59(236):433-460, 1950. 
[124] A. Sloman. Some requirements for human-like robots: Why the recent over-emphasis on 
embodiment has held up progress. Creating Brain-like Intelligence, pages 248-277, 2009. 
[125] A. Sloman. If learning maths requires a teacher, where did the first teachers come from. In 
Proc. Int. Symp. on Mathematical Practice and Cognition, AISB, pages 30-39, 2010. 
[126] A. Sloman. Comments on Annette Karmiloff-Smith's (1992) book: Beyond Modularity: A 
Developmental Perspective on Cognitive Science. 
http://www.cs.bham.ac.uk/research/projects/cogaff/misc/beyond-modularity.html, 2011. 
[127] L. Smith. Cognition as a dynamic system: Principles from embodiment. Developmental 
Review, 25(3-4), 2005. 
[128] J.A. Starzyk. Frontiers in Robotics, Automation and Control, chapter Motivation in 
Embodied Intelligence. InTech, 2008. 
[129] R.R. Vane and D. Griffith. Cognitive automation solves many AI-hard problems. In AAAI 
Spring Symposium, 2006. 
[130] R. Velik. A model for multimodal humanlike perception based on modular hierarchical 
symbolic information processing, knowledge integration, and learning. In 2nd IEEE International 
Conference on Bio-Inspired Models of Network, Information and Computing Systems, pages 168-
175, 2007. 
[131] R. Velik and D. Bruckner. Neuro-symbolic networks: Introduction to a new information 
processing principle. In 6th IEEE International Conference on Industrial Informatics, pages 1042-
1047, 2008. 
[132] R. Velik. Neuronal Pattern- and Object Recognition: Methods of Analysis in Time and 
Frequency Domain. VDM Verlag, 2009. 
[133] R. Velik. A Bionic Model for Human-like Machine Perception. Suedwestdeutscher Verlag 
fuer Hochschulschriften, 2008. 



BRAIN. Broad Research in Artificial Intelligence and Neuroscience 

Volume 3, Issue 3, October 2012, ISSN 2067-3957 (online), ISSN 2068 - 0473 (print) 

 

 54 

[134] R. Velik. A Bionic Model for Human-like Machine Perception. PhD thesis, Vienna 
University of Technology, 2008. 
[135] R. Velik, D. Bruckner, R. Lang, and T. Deutsch. Human-Computer System Interaction: 
Backgrounds and Applications, chapter Emulating the Perceptual System of the Brain for the 
Purpose of Sensor Fusion, pages 17-27. Springer Berlin/Heidelberg, 2009. 
[136] R. Velik and D. Bruckner. A bionic approach to dynamic, multimodal scene perception and 
interpretation in buildings. International Journal of Intelligent Systems and Technologies, 4(1):1-9, 
2009. 
[137] Velik, R. Quo vadis, Intelligent Machine? BRAIN. Broad Research in Artificial Intelligence 
and Neuroscience, 2010, 1, 1-9. 
[138] R. Velik. Towards human-like machine perception 2.0. International Review on Computers 
and Software, 5(4):476-488, 2010. 
[139] R. Velik and G. Zucker. Autonomous perception and decision-making in building 
automation. IEEE Transactions on Industrial Electronics, 57(11):3625-3652, 2010. 
[140] R. Velik and H. Boley. Neurosymbolic alerting rules. IEEE Transactions on Industrial 
Electronics, 57(11):3661-3668, 2010. 
[141] R. Velik. The neuro-symbolic code of perception. Journal of Cognitive Science, 11(2):161-
180, 2010. 
[142] R. Velik. Why machines cannot feel. Minds and Machines, 20(1):1-18, 2010. 
[143] R. Velik, G. Zucker, and D. Dietrich. Towards automation 2.0: A neuro-cognitive model for 
environment recognition, decision-making, and action execution. EURASIP Journal on Embedded 
Systems, 2011(11), 2011. 
[144] R. Velik. From simple receptors to complex multimodal percepts: A first global picture on 
the mechanisms involved in perceptual binding. Frontiers in Cognitive Science, 3:1-13, 2012. 
[145] P. Wang. Rigid Flexibility: The Logic of Intelligence. Springer Verlag, 2006. 
[146] P. Wang and B. Goertzel. Introduction: Aspects of Artificial General Intelligence. In Advance 
of Artificial General Intelligence, pages 1-16. IOS Press, Amsterdam, 2007. 
[147] S. Watt. Naive psychology and the inverted Turing test. Psycoloquy, 7:14, 1996. 
[148] J.Weng, W.S. Hwang, Y. Zhang, C. Yang, and R. Smith. Developmental humanoids: 
Humanoids that develop skills automatically. In 1st IEEE-RAS International Conference on 
Humanoid Robots, 2000. 
[149] N. Wiener. Cybernetics. J. Wiley, 1948. 
[150] H. Zeilinger. Bionically Inspired Information Representation for Embodied Software Agents. 
PhD thesis, Vienna University of Technology, 2010. 
[151] N. Zhong. Intelligent Information Processing IV: IFIP Advances in Information and 
Communication Technology, volume 288, chapter Towards Brain-inspired Web Intelligence. 
SpringerLink, 2008.