31.07.2015 Views

Steven Pinker -- How the Mind Works - Hampshire High Italian ...

Steven Pinker -- How the Mind Works - Hampshire High Italian ...

Steven Pinker -- How the Mind Works - Hampshire High Italian ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

114 | HOW THE MIND WORKSity. The distributed input representation, in which a concept does not getits own unit ("parrot") but is represented by a pattern of activity overunits for its properties ("fea<strong>the</strong>red," "winged," and so on), allows for automaticgeneralization to similar concepts and thus nicely fits thie law ofassociation by resemblance. And if all parts of <strong>the</strong> mind start off as <strong>the</strong>same kind of network, we have an implementation of <strong>the</strong> blank slate.Connectionism thus offers an opportunity. In seeing what simple neuralnetworkmodels can and cannot do, we can put <strong>the</strong> centuries-old doctrineof <strong>the</strong> association of ideas to a rigorous test.Before we begin, we need to set aside some red herrings. Connectionismis not an alternative to <strong>the</strong> computational <strong>the</strong>ory of mind, but a varietyof it, which claims that <strong>the</strong> main kind of information processing doneby <strong>the</strong> mind is multivariate statistics. Connectionism is not a necessarycorrective to <strong>the</strong> <strong>the</strong>ory that <strong>the</strong> mind is like a commercial computer, witha high-speed, error-free, serial central processing unit; no one holds that<strong>the</strong>ory. And <strong>the</strong>re is no real-life Achilles who claims that every form ofthinking consists of cranking through thousands of rules from a logic textbook.Finally, connectionist networks are not particularly realistic modelsof <strong>the</strong> brain, despite <strong>the</strong> hopeful label "neural networks." For example, <strong>the</strong>"synapse" (connection weight) can switch from excitatory to inhibitory,and information can flow in both directions along an "axon" (connection),both anatomically impossible. When <strong>the</strong>re is a choice between getting ajob done and mirroring <strong>the</strong> brain, connectionists often opt for getting <strong>the</strong>job done; that shows that <strong>the</strong> networks are used as a form of artificialintelligence based loosely on <strong>the</strong> metaphor of neurons, and are not a formof neural modeling. The question is, do <strong>the</strong>y perform <strong>the</strong> right kinds ofcomputations to model <strong>the</strong> workings of human thought?Raw connectoplasm has trouble with five feats of everyday thinking.The feats appear to be subtle at first, and were not even suspected ofexisting until logicians, linguists, and computer scientists begaft to put<strong>the</strong> meanings of sentences under a microscope. But <strong>the</strong> fdats givehuman thought its distinctive precision and power and are, I think, animportant part of <strong>the</strong> answer to <strong>the</strong> question, <strong>How</strong> does <strong>the</strong> mind work?One feat is entertaining <strong>the</strong> concept of an individual. Let's go back to<strong>the</strong> first departure of neural networks from computerlike representa-

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!