What is Thought?
Wednesday, November 3, 2004
Building 3 Auditorium - 3:30 PM
(Refreshments at 3:00 PM)
Dr. Eric Baum, will talk about "What is Thought?". Can the strong AI/ Turing picture be extended to a plausible model of all aspects of mind, such as understanding, creativity, language, reasoning, learning, and consciousness? I propose a candidate realization as follows: Meaning is the computational exploitation of the compact underlying structure of the world, and mind is execution of an evolved program that is all about meaning. The computational learning literature has explained concept learning as stemming from a formalized Occam's razor: a compact program (Occam explanation) consistent with enough examples of a concept yields generalization to new examples. These results are extrapolated to the conjecture that meaning results from finding a compact enough program behaving effectively in the world; such a program can only be compact by virtue of code reuse, factoring into interacting modules that capture real concepts and are reused metaphorically or pleiotropically. For a variety of reasons, including arguments based on complexity theory, developmental biology, evolutionary programming, ethology, and simple inspection, this compact Occam program is most naturally seen to be in the DNA, rather than the brain. The genome is the compact source code, the brain embodies its executable. Learning and reasoning are then fast and almost automatic because they are constrained by the DNA programming to deal only with meaningful quantities. Evolution itself is argued to exploit meaning in related ways, explaining why it is so computationally efficient. Words are labels for meaningful computational modules, explaining why they are so rapidly learned. The differences between human and ape cognition are naturally viewed as largely due to nurture, stemming from massive programming that humans have accumulated over millenia on top of the DNA code-- the cumulative discovery of new meaningful modules being made possible by language.
The many aspects of consciousness are very naturally and consistently understood in this context. For example, qualia (the way things feel subjectively) have exactly the appropriate nature and meaning that evolution coded in the DNA so that the compact program behaves effectively. This theory is consistent with vast data from a variety of fields and will continue to confront empirical tests as technology progresses, for example in measuring gene expression, in brain imaging, and as new psychophysics experiments are proposed. This talk is based on the new, eponymous book (What is Thought? MIT press 2004) and requires no prior familiarity with computer science.
Eric B. Baum has held positions at the University of California at Berkeley, Caltech, MIT, Princeton, and the NEC Research Institute. He holds a BA and MA from Harvard and a PhD in physics from Princeton. He has published extensively in theoretical physics, machine learning, machine reasoning, cognitive science, and DNA computing. He is currently developing algorithms for cognitive computing related to the ideas in What is Thought?.
IS&T Colloquium Committee Host: Tony Gualtieri