Page 374 - Deep Learning

Belief Revision: The Resubsumption Theory 357

if a person’s belief system Th(B) does a poor job of accounting for what hap-
pens in domain B, there is no guarantee that he will construct some alternative
theory Th(A+B) that does a better job. That is, he may or may not have some-
thing to convert to. Even if a potential alternative is present in his belief base,
the possibility of applying Th(A+B) to B might go unnoticed for a lifetime.
Alternatively, the person might lack an overarching theory that says that these
two theories are mutually incompatible. He might cheerfully accept that he has
two ways of looking at domain B but not feel any need to choose between them.
Also, his environment might not provide a sufficient number of opportunities
to apply his contending theory for him to realize that the resident theory has
less utility than the contender. Finally, the resident theory might be situated in
such a way within his belief base that propagation of a switch in truth value
leaves all core beliefs unchanged. All of these conditions affect whether the
conversion of a core belief will occur.
In contrast, the ontological shift theory, in its current stage of develop-
ment, does not come with an explicit formulation of the exact mental trig-
31
gering conditions for ontological shifts. There is no obvious explanation in
the ontological shift theory for why ontological shifts are difficult or have a
low probability of occurring. To detach a link between some phenomenon X
and a category C and to create a new link between X and some other category
1
C are simple processes that do not seem unduly demanding. Why do we not
2
shift ontological categories at the drop of a hat? The difficulty of conceptual
change is attributed by Chi and co-workers to a lack of awareness of the need
for such a shift, a lack of the correct target category and “the cognitive demand
of re-inheriting all the attributes of a concept based on its new categorical
membership.” Only the third of these three – the cognitive load associated
with conceptual change – applies in the central cases in which a person is
aware of a cognitive conflict and is in possession of the correct target category
but nevertheless does not shift. In those cases, the cognitive load hypothesis,
if taken to be the whole story, predicts that conceptual change will be blocked,
and so cannot explain why it happens when it does happen.
A radical difference between resubsumption and other theories of belief
revision, conceptual change or theory change is that resubsumption implies that
dissatisfaction with a resident theory is neither a necessary nor a sufficient trig-
gering condition for change. The discovery that some other theory provides a
different way of looking at something can happen even if one is unaware of any
difficulties with one’s current beliefs. Spontaneous or opportunistic retrieval in
combination with structural alignment do not require doubts about the resi-
dent theory. During competitive evaluation, satisfaction and dissatisfaction with