Page 47 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

34 CHAPTER 2 Mind, Brain, Autonomous Agents, and Mental Disorders

in real time, can generate the behavioral data as emergent properties. This conceptual
leap from data to design is the art of modeling. Once derived, despite being based
on psychological constraints, the minimal mathematical model that realizes the
behavioral design principles has always looked like part of a brain (Fig. 2.1). I first
experienced such a derivation of brain mechanisms from psychological hypotheses
when I was a freshman at Dartmouth College in 1957e58. It was a transformative
experience that shaped the rest of my life https://youtu.be/9n5AnvFur7I.
Thepast60 years of modeling have abundantly supported the hypothesis that brains
look the way that they do because they embody natural computational designs
whereby individuals autonomously adapt to changing environments in real time.
Therevolution in understanding biological intelligence is thus, more specifically, a rev-
olution in understanding autonomous adaptive intelligence. The link from behavior-to-
principle-to-model-to-brain has, in addition, often disclosed unexpected functional
roles of the derived brain mechanisms that are not clear from neural data alone.
At any stage of this modeling cycle, the goal is to first derive the minimal model
that embodies the psychological hypotheses that drive the model derivation. Such a
“minimal” model is one for which, if any model mechanism is removed, or
“lesioned,” then the remaining model can no longer explain a key set of previously
explained data. Awise theorist should, I believe, strongly resist “throwing in” known
neural mechanisms that are not yet in the minimal model if there is no functional
understanding of why they are needed. Once the link between mechanism and
function is broken in this way, the ability of the current minimal model to drive
further model refinements will be lost.
In particular, once a connection is made top-down from behavior to brain by such
a minimal model, mathematical and computational analysis discloses what data the
minimal model, and its individual and species variations, can and cannot explain.
The data that cannot be explained are as important as those that can be explained,
because they demarcate a “boundary between the known and the unknown.”
Analysis of this boundary focuses a theorist’s attention upon design principles
that the current model does not yet embody. These new design principles and their
mechanistic realizations are then consistently incorporated into the model to
generate a more realistic model, and one that has always been able to explain and
predict a lot more psychological and neurobiological data. If the model cannot be
refined, or unlumped, in this way, then that is strong evidence that the current model
contains a serious error, and must be discarded.
This theoretical cycle has been successfully repeated multiple times, and has led to
models with an increasingly broad explanatory and predictive range, including models
that can individually explain psychological, neurophysiological, neuroanatomical,
biophysical, and biochemical data. In this specific sense, the classical mind/body
problem is being solved through principled, albeit incremental, refinements and
expansions of theoretical understanding. One can think of these incremental refine-
ments as a way that a theory can try to carry out a kind of “conceptual evolution”
by analyzing how various environmental pressures may have driven the biological
evolution of our brains.