Page 375 - Deep Learning

358 Conversion

outcomes play a role in the adjustment of the relevant confidence levels, but the
conversion process is not triggered by doubt but by the conceptual discovery
that there is an alternative way of looking at the relevant situation or domain. It
is the positive act of extending the reach of a successful conception that sets the
stage for theory change, not dismay over the accumulation of anomalies.
Indeed, the Resubsumption Theory implies that the causal sequence
between anomalies and conversion is the reverse of that envisioned by Popper,
Kuhn and psychologists in the science-inspired tradition. In the presence of a
single theory, an anomaly is merely that: yet another anomaly. Because we are
never in possession of a belief system that explains everything, we live with
anomalies and we deal with them by, briefly put, ignoring them. When a per-
son becomes aware that there is a contender theory, he becomes more moti-
vated to pay attention to anomalies. Anomalies acquire importance only if they
become reasons to choose one theory over another, which presupposes that
the contender theory has already been constructed by the time the anomalies
are seriously considered. Hence, anomalies are not triggers for theory change,
although they might guide the direction of such a change.
The main advantage of the Resubsumption Theory over alternative theories
is that it resolves the conceptual knot posed by our disturbingly powerful ability
to sweep anomalies and contradictions under the conceptual rug. By postulat-
ing that the trigger for theory change is the discovery of a previously unheeded
conflict between two already acquired belief systems, the theory of resubsump-
tion cuts the circular relation between new information and prior knowledge
and resolves the assimilation paradox: Belief revision is possible in the face of
resistance to contradictory evidence because theories initially arise in contexts
in which they are compatible with the available evidence, and because theory-
theory conflicts are not decided on the basis of logical assessment of the strength
of the relevant evidence but on the basis of intuitive estimates of relative utility.

THE PRAGMATIC IMPERATIVE

Philosophers tend to discuss belief revision and theory change in terms of logical
entities: Theories are conjunctions of propositions, predictions are logical deri-
vations and the key questions are which propositions are true, which predictions
follow from which theories and whether theory change is rational. But belief
revision happens to people, so every instance of it occurs in somebody’s brain.
Neuroscientists discuss brain functioning in terms of cause and effect, changes
in electrical charges over time and the chemical properties of synapses, and
the central questions pertain to the impact of variables like the frequencies of