Volume 8 (will be Vol 7 on site)


How the Brain Creates Predictive Models of the Environment
Kaitlyn Tuvilleja

 Abstract
Researchers, utilizing technologies like fNIRS (functional near-infrared spectroscopy) on 6-month-olds and EEG
(electroencephalography) on infants viewing image sequences, can uncover how young brains craft and apply predictive
models to react to events. Adult brain learning is also examined through two types: model-free learning (trial and error) and
model-based learning (implementing predictive models). By analyzing data from various age groups, researchers can examine
how brains craft predictive models of the environment and leverage those findings for future implications. Dissecting the way
the brain constructs predictive models of the environment at distinct ages is crucial for developing enhanced educational
practices: this paper examines the development of predictive cognitive models from infancy through adulthood using
neurological studies, highlighting their implications for enhancing educational strategies and adaptive behaviors.

Introduction
The way that we react or the choices we make in certain
situations is hypothesized to be guided by internal models.
This can trigger the body’s flight-or-fight response, curiosity,
and even hunger. For example, when going to the doctor, the
doctor will hit an individual’s knee to check their reflexes.
What may seem like a knee-jerk reaction is a realization of
the brain’s predictions about immediate danger, analogous to
how a pianist’s fingers can predict where the next note is
through repetitive practice and auditory exposure to their
piece. The brain actively constructs a "call for action" that
drives us to continuously shape the brain’s internal models
(Kayhan et al., 2019). This way, the internal map that the
brain has created can guide future actions and decisions. The
brain’s activation process for predictive models can even be
thought of as a construction zone: a place where it actively
repairs and reshapes the mental map in response to
unexpected changes. To support this theory, O’Reilly’s team
from Oxford University selected a handful of adults and
presented a target object that would change positions. The
positions included both predictable locations, learned through
repeated exposure and unexpected ones. The study revealed
that “Activation in the parietal cortex when an immediate
motor response was programmed as participants had to
update their internal models to accommodate the change of
target locations” (O'Reilly, 2013). When the change was
predictable, the parietal cortex, a region of the brain
responsible for processing sensory information and
coordinating motor responses, quickly adjusted the planned
motor commands. However, a surprise triggers the brain’s
anterior cingulate cortex–an area responsible for updating
internal models based on error detection–to activate.
Essentially, when unexpected movements occur, they trigger
specific brain regions in charge of movement planning and
adaptation, evidence to the hypothesis of the brain creating
internal models. This factors into the ability to create internal
predictive models, and how individuals learn and react to the
world through these models, starting from infancy.

Predictive Model Creation in Infants
Current research explores how infants develop the capacity
to construct mental maps. As of 2015, a functional near-
infrared spectroscopy (fNIRS) study was conducted on
infants (Gallagher, 2023). An additional study conducted by
E. Kayhan at the University of Potsdam demonstrated that at
6 months old human brains already create a predictive model
of the environment. More specifically, “after a learning period,
when images were unexpectedly omitted, infants showed
activation in the occipital cortex, as if an image was
presented, suggesting that they generated predictions about
the visual input”(Kayhan et al., 2019).

Researchers were interested in how babies process and
adapt to change when prompted by unfamiliar environments.
To explore this, they conducted a study with sixty 9-month-
old infants. Dr. Kayhan and his team had predicted that “if
participants formed predictions based on the repeated
observations of the predictable stimuli, they would show a
prediction error response when their predictions were violated
by the unexpected appearance of the cues” (Kayhan et al.,
2019). The study involved showing pictures to the babies and
would utilize a system that combined sound and brainwave
monitoring (audio-visual EEG). The pictures featured a bee,
but the sequence changed to test the babies' predictions.
First, the babies were shown repetitive images of a bee
(expected sequence). This established a control data set for
the scientists. Next, the babies were then shown a
predictable surprise which included a bee image followed by
an image that could be associated with the bee (i.e. a flower).
Finally, the babies were shown an unexpected sequence
where the bee image was followed by an image that wouldn’t
make sense (i.e. a truck). Scientists expected that babies'
brains would be more surprised if a pattern was not followed,
inducing a stronger electrical response called an Nc wave.



The first session involved participants to observe pre-
determined choices without rewards. From this, researchers
were able to measure state prediction errors (SPEs) – the
difference between predicted and actual outcomes – and
reward prediction errors (RPEs) – surprises associated with
unexpected rewards. By examining the errors, researchers
were able to gauge the participant's initial understanding of
the system. The second session gave participants the liberty
to implement their own choices with the potential of a reward.
This tested whether patients could utilize the knowledge in
the previous session. Researchers revealed that 13 out of the
18 participants were successfully able to employ model-
based learning to execute optimal choices. Suggesting a
clear preference for constructing predictive models over trial
and error approaches. Participants acquired optimal decision-
making through model-based learning which is illustrated
through their ability to adapt to the system's modifying tasks,
even when rewards were provided. This would demonstrate
the idea that “participants would acquire knowledge about the
transition probabilities during session 1, despite the absence
of any rewarding outcomes. This state knowledge can
therefore be only acquired through model-based learning,
potentially updated via an SPE” (Gläscher, 2023).

Future Implications for Learning
Like infants, these findings can be used to configure
educational plans for students. As children grow older, their
brain develops. To aid these developments, it should be
important to continue implementing lessons that can be
designed to build upon and challenge these predictive
abilities.

Brains do not inactively experience the world but actively
work to construct models to predict what will happen next. 

Before the experiment, researchers had predicted that babies
would portray a weaker Nc wave (less electrical activity) by
seeing the repeated bee images. NC waves detected
whether attention was suppressed during the trials (if infants
didn’t pay attention during trials). Interestingly enough, there
was not a significant difference in Nc wave strength between
the update and no-update trials within the enclosed section
(Fig. 1). This shows that infants might be processing given
information more deeply than expected, almost challenging
previous knowledge on how infants create predictive models
of the environment. 

These results can potentially be incorporated in lower
education fields, potentially setting a foundation for surprise-
based learning. Teachers can implement some surprise or
novelty in learning experiences that could benefit children's
learning. For example, allowing for more open-ended
questions can encourage students to explore different
approaches to problem-solving scenarios, sparking curiosity
toward unexpected results. By honing in on a child’s ability to
create predictive models, educators can create more
effective and engaging learning practices that foster critical
thinking and problem-solving skills.

Predictive Modeling in Adults 
However, before revamping academic frameworks based on
children’s predictive models of the environment,
understanding the brain’s decision-making is essential. In
part, adult brains can learn from reinforcement learning:
model-free and model-based (Otto et al., 2015). Model-free
RL directly uses past experiences to figure out what actions
are rewarding, while model-based RL builds a mental map of
how the world works and uses this map to decide what action
to take. To further explore this phenomenon, scientist
Gläscher strived to find how predictions are created through
“trial-by-trial neural signals that reflect the dynamics of this
learning” (Gläscher, 2023). 

The study consisted of 18 Caltech adult students with normal
vision) and no neurological or psychiatric conditions (20/20
and not colorblind). Within two sessions, researchers
examined how participants developed optional decisions in a
reward-based environment.

Figure 1.  Shows the brain signal from the vertex of the head for the
different trial types. The update line shows the highest amplitude
meaning the brain produced a stronger Nc wave with the update

trial than no-update and expected (Kayhan et al., 2019).

Figure 2. Illustration of the decision-making process of scientist
Gläscher and his research team at CalTech’s study on model-free

reward learning and model-based reward learning (Gläscher)

Brain Matters・Volume VII      68



This “auto-pilot” function, through its remarkable plasticity,
continues to refine its predictive models. Research on infants
illustrates the ability to effectively detect unexpected events
and update models accordingly. Meanwhile, in part, adults
utilize model-free and model-based learning for decision-
making. With knowledge of children and adult brain plasticity,  
educators can leverage these traits to create more engaging
learning environments and foster groundwork for adult
learning styles. These findings in plasticity hold the potential
for unlocking new opportunities for personalized cognitive
training programs for children. For example, educational
practices can be designed to refine and challenge a student’s
predictive models. These implications can be extended to
older individuals, as well. With this new profound knowledge
of brain plasticity, the possibilities are truly invigorating.

References 
1. Arsten, A., et al. (2012, April). Neural circuits responsible
for conscious self-control are highly vulnerable to even mild
stress. When they shut down, primal impulses go unchecked
and mental paralysis sets in. NCBI.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4774859/#fn-
group-a.f.atitle
2. Gallagher, A. (2023, March 3). Functional near-infrared
spectroscopy in pediatric clinical research: Different
pathophysiologies and promising clinical applications. NCBI.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9982436/
3. Gläscher, J. (2023, June 16). States versus rewards:
Dissociable neural prediction error signals underlying model-
based and model-free reinforcement learning. Neuron.
https://www.cell.com/fulltext/S0896-6273(10)00287-
4#secd29307264e631 Kayhan, E., et al. (2019, July 10).
Nine-month-old infants update their predictive models of a
changing environment. NCBI.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6969335/
4. Moran, R. J. (2014, January 23). The brain ages optimally
to model its environment: Evidence from sensory learning
over the adult lifespan. PLOS.
https://journals.plos.org/ploscompbiol/article?
id=10.1371/journal.pcbi.1003422
5. O'Reilly, J. X. (2013, June 14). Making predictions in a
changing world—inference, uncertainty, and learning. NCBI.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3682109/
6. Otto, A. R., et al. (2015, May 1). Cognitive control predicts
use of model-based reinforcement-learning. NCBI.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4387848/
7. Roozendaal, B., & McGaugh, J. L. (2011, December).
Memory modulation - PMC. NCBI.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3236701/
8. Šimić, G., et al. (2021, May 31). Understanding emotions:
Origins and roles of the amygdala. NCBI.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8228195/