Subject: Frame Problem for Bats From: Harry Erwin <harry.erwin@xxxxxxxx> Date: Tue, 7 Apr 2009 08:12:20 +0100 List-Archive:<http://lists.mcgill.ca/scripts/wa.exe?LIST=AUDITORY>The frame problem is discussed in Shanahan, M., 1997, Solving the Frame Problem, MIT PRESS, summarised very briefly here. The AI community uses the 'frame problem' as shorthand for the difficulties that arise when the effects of actions and events are represented in formal logic. AI researchers start from two premises: "The best way to understand intelligent behaviour is as the product of correct reasoning on correct representation." and "The best way to explicate the notions of correct representation and correct reasoning is through formal logic." This leads to a situation calculus that reasons about situations, fluents (properties that can change), and actions. The naive situation calculus has an enormous number of axioms (approximately #fluents x #actions in number) that mostly state: "When you take this action, that fluent doesn't change." Much of the work on the situation calculus has revolved around ways to drop all those 'stays the same' axioms. I'm constantly exposed to this perspective in my robotics work. My background, however, is in bat behaviour. Now I suggest echolocating bats are reasonably intelligent animals that live mostly in an internal model of their external environment. (See references to the Wiederorientierung phenomenon for the argument.) A motor action by a flying bat changes almost everything about the mapping between the external environment and its internal representation. Hence, for a flying bat the situation calculus has few 'stays the same' axioms that might be dropped. Yet bats are able to make a living. It's unlikely bats represent actions and fluents as discrete 'things'. I suspect almost everything in their internal model is a distributed pattern of neural activity. Either the patterns are stable, and the transformation from what they hear to pattern updates is very complex; or the transformations are stable, and the patterns change wildly as they manoeuvre in the air. How do you reason on that? So, has anyone thought about what intelligent behaviour in animals says about the frame problem?