Bryn Mawr College Center for Science In Society

To facilitate the broad conversations, involving both scientists and non-scientists, which are essential to continuing explorations of the natural world and humanity's place in it, the nature of education, the generation, synthesis, and evaluation of information, technology and its potentials, the relationships between forms of understanding.

Participants:
Doug Blank (Computer Science) Anne Dalke (English), Ted Fernald (Linguistics-Swat), Panama Geer (Math, Computer Science) , Paul Grobstein (Biology), Deepak Kumar (Computer Science), Eric Raimy (Linguistics), Kathryn Rowe (English), George Weaver (Philosophy), and Rob Wozniak (Psychology)

One Summary View (prepared by Eric Raimy; views by other participants encouraged and can be sent either by email or posted using our working group forum area):

The meeting began with a question from Kathryn for the 'scientists' in the group. The question was whether the readings from Beckett could be used as evidence of some sort and if so what kind of evidence? Paul gave the main response in that he thought Beckett is attempting to use metonomy to identify metaphorical aspects of the mind. Additionally, Beckett attempts to do this in an 'acultural' way or in a way that escapes cultural biases. Consequently, Paul thinks that this sample of writing can be used as evidence to highlight the general structures of thought and the mind.

A question was raised at this point about Beckett's use of metonomy. From Paul's point of view the metonymic structure of the narrative (jumping from one idea to another in a sequential manner instead of using metaphors to relate the ideas) provides an insight into how thought operates or is organized. The question was raised though of how to distinguish between a metonymic narrative like this from a narrative that is created by simply randomly shuffling different ideas. The group concluded that there did not appear to be any easy way of determining the difference between these two views of narratives like Beckett.

As part of Paul's discussion of Beckett, he made a distinction between conscious and unconscious mental processes. Rob pointed out that 'unconscious' can be a dangerous term because there is more than a single interpretation of 'unconscious' for mental processes. One interpretation is 'cognitive unconsciousness' which is in principle unconscious in nature. Rob gave the example of working memory as this type of mental process. Working memory is completely impenetrable to introspection. No one can describe or tell what is exactly going on as a mental process in working memory. This type of 'unconscious' is different from 'Freudian unconsciousness' which can be penetrated by guided introspection. Everyone agreed that the distinction between the two types of unconsciousness is important from a terminological point of view. There were still two points of disagreement on this topic though. The first was that 'Freudian unconsciousness' is not really acceptable to current views of psychology. Paul expressed this view even though it is not his own and Rob disagreed with it. Finally, Paul and Rob disagreed with whether these two types of unconsciousness derive from one or two distinct neuro-biological sources. Paul believes that both types of unconsciousness are derived from a single neuro-biological source and Rob believes that there are two distinct neuro-biological sources, one for each type of unconsciousness.

The discussion at this point turned to the new article "High-level perception, representation, and analogy: A critique of artificial intelligence methodology" by Chalmers, French and Hofstadter. The first point of discussion was whether the article was 'dated' or not. The group differed in their answer to this question. Doug felt that the article was not dated at all because its main points of criticism have not be addressed in AI research. Deepak on the other hand felt that current AI research includes perceptual aspects and thus lessens the relevance of the article. Although Deepak admitted that the perceptual aspects of AI research which are present today are not the same as the ones pointed out by the article, he did still hold that the article is a little bit dated.

The remainder of the discussion about this article was about the nature of representations. One of the main points in the article was that there was too much stipulation of representations in work on AI. Instead the article proposed that AI research should work on models that build the representations that are computed over instead of just stipulating the representations. An expansion of this theme is to question whether representations are necessary at all. This position is being investigated by researchers working on connectionist models. Many in the group were uneasy about lessening the role of representations though. Rob pointed out the fact that many approaches in AI have 'data acquisition' interleaved with task processing and this approach skips the question about how the representations are learned. Rob further elaborated this point by discussing the multiple nature of representations. According to Rob there are three distinct meanings/types of representations. The first can be considered to be the 'post conscious experience' that we recognize as cognitiion. The way we experience the world is a representation created by our mind. With this position there are interesting questions about how 'real' this representation is but this point more likely sharpens this definition of representation as opposed to weakening it. The second type of representation is the internal storage of information in our unconscious. There was little discussion of this type of representation. The third type of representation was symbols as representations with the example of language in particular sentences. Again there was little discussion on this definition.

The end of the meeting consisted of a discussion about the relationship between representations and computation. Although the discussion throughout this part of the meeting appeared to be contentious everyone in the group recognized the importance of both representations and computations. The different positions expressed in the meeting were just coming from different ends of the spectrum on this issue. There is much to be learned about representations by focusing on computational issues which is the main point of the article for today. There is also much to be learned about computation by focusing on representational issues. Eric summarized his point with a quote from John McCarthy, "If the representations are right, the rules will follow".