Registration � Society of Mind

Greg Detre

Thursday, February 06, 2003

gdetre@fas.harvard.edu

Special student (graduate), Harvard, Philosophy/CS

Courses taken last term:

computational semantics � computational linguistics � intelligent machines � formal systems and computation � context-aware computing

Courses taken previously:

philosophy of mind � epistemology � psychology of language and cognition � animal behaviour � neuroscience � history of philosophy

Strong C++

Academic interests:

AI, connectionism, the mind-body problem and the Turing test

I�m going to try a preliminary definition of common sense:

common sense is a measure of how well you can deal with new problems in the environment you�re used to.

High-level common sense is a compendium of methodologies, facts and representations that derive their effectiveness from their diversity. When we talk of �common sense�, we are evaluating how well these work in totality. Common sense is practical � it�s about being successful in our everyday behaviour, where we are constantly being presented with problems that are �new� at least in some respects. At the very least, new problems require us to realise how the current situation is similar to past situations, how it�s different, and consequently how to improve on last time�s behaviour. New problems may require us to form whole new representations and models, or to leverage current knowledge in new ways. For these reasons, common sense is inextricably bound up with recognising and responding appropriately to contexts. Finally, mentioning �the environment you�re used to� is crucial, since common sense is necessarily domain-specific. In the case of human-level common sense, the problems that we have to deal with are hugely complex and varied, partly because of the complexity of our environment, and especially because they involve other humans. Secondly, these problems are inherently shaped by how human bodies interact with their environment � they are problems shaped by our intellectual and sensorimotor limitations. If we want our machines to display human-like common sense, then we may well have to embody them enough for them to interact and learn from our environment in much the same way that we do.