Linking the Bayesian brain, perceptual
uncertainly, and cognitive prediction at the
wheel
The nervous system’s ability to dynamically coordinate cognitive
functions in a world of sensory uncertainty is one of the
greatest evolutionarily-given tools \cite{Jensen_2012}. Although from
experience, our perceptual world looks stable and determined, many
factors indicate that coordination with incoming sensory information is
less reliable than it appears \cite{Geisler_2002}. For example, the variability in
the density of the receptors of the retina -as well as the structural
limitations in neural organization and the variety and variability of
information introduced in the early stages of sensory coding- mean that
the nervous system must dynamically and effectively deal with the
resulting uncertainty. Appropriate uncertainty management generates
accurate perceptual representations of the world. Thus guiding
decisions and actions in a more adaptive manner. Therefore, perception
is a process of active unconscious and probabilistic inference \cite{Friston_2012,Moutoussis_2014,Moutoussis_2014a} (Barlow 1990 --> no lo encuentro).
On the one hand, Bayesian methods have proven successful in describing
computational theoretical models for perception and sensory-motor
control \cite{Buckley_2017}. On the other hand,
psychophysics has provided a growing body of evidence on how perceptual
calculations in human beings follow an optimal Bayesian pattern
\cite{Buckley_2017}. This means that Bayesian coding approach
reverse the notions classically established of perception as a largely
bottom-up process of information accumulation or stimulus detection
driven by the impact of sensory signals. Suggesting instead that both perceptual content and behavioral response is determined by
top-down predictive signals. These signals would arise from multi-level generative internal
models of environmental sources, which are
continually modified by bottom-up prediction error signals. The latter ones, communicating
reciprocal exchanges of information between predicted and actual signals
across the different sensory and perceptive hierarchical levels,
ultimately impacting on behavioral execution. Thus, evidence indicates
the processes by which the human brain represents incoming sensory
information, could be understood in a probabilistic way [CITE EVIDENCE]. Thus, the Bayesian coding hypothesis, suggests the human brain could be a natural probability distribution
generator (Friston., 2009, 2010). This theoretical and
experimental approach has fundamental implications for studying the brain in everyday environments, particularly in the way we understand and describe the
neural calculations and natural dynamics of neural representations. However, direct
applications in daily life on this hypothesis are almost non-existent. Determining how and which are the requirements of the driver’s
brain to coordinate, the dynamics and natural neuronal
interactions that are useful for the effectiveness of information coding
in environments of high sensory uncertainty, is an important objective
not only for the development of neuroergonomics as a field, but also a major
advance in the field of applications in complex social contexts.
In consequence, considering the Bayesian brain framework (Friston, 2009, 2010 & 2011), if we provide
relevant information to optimize a probability calculation (e.g. the prediction of motor behavior on a highway) our brain would use
this information and modify itself from its interaction with incoming
stimuli. Thus drawing a probabilistic prediction, with higher success rate of
what could happen within the given context. In real
high-uncertainty road driving conditions, the drivers’ brain would
consider the inputs from the environment and compare them with their
previous experiences, in order to regulate and modify their behavior
according to perceived risk \cite{wilde1998risk}. In turn, optimization of
brain functioning -by integrating the most relevant variables and trying
to zero out any errors in its predictive model- can occur in a
moment-by-moment basis. Furthermore, optimization involves updating the
current model to reevaluate sensory inputs. Selective sampling of new
incoming sensory inputs will thus generate adaptive responses reducing
uncertainty and facilitating behavioral performance \cite{Friston_2015}. Hence the nervous system is prepared to make active and probabilistic
inferences of what might happen in our nearby sensory world.
Furthermore, it can update such models with new predictive information,
increasing the degree of accuracy. Environmental Psychology can
therefore use this framework in order to develop systems that can
facilitate such an optimization process. For example, using information
derived from state-of-the-art predictive model of accidents in order to provide online non intrusive data about driving
conditions \cite{basso2018real}. Given the fact that in real driving scenarios signals and
stimuli are in constant competition, transmitting information to drivers
is not an easily achievable goal.