Overview of interactions

Q1. Several participants:

Your formula (3) must be completely wrong. It says that if $\alpha$ is not perceived, and if it is neither believed nor disbelieved, then it is both believed and disbelieved.

A1. John Bell (13.1):

In this case we have (writing  a  for $\alpha$):

	P(a)(t) ^ ¬ B(¬ a)(t) ·-> P(a)(t+1)
	¬ P(a)(t) ^ ¬ B(a)(t) ·-> B(¬ a)(t+1)

Suppose that, at  t , I neither believe  a  nor  ¬ a , and neither perceive  a  nor  ¬ a . Then the second axiom gives both

B(¬ a)(t+1)

and

B(¬ ¬ a)(t+1)

which is not an inconsistency in the logic but describes an inconsistent state of belief.

A fundamental part of our theory is that perceptibles are atomic, so $\alpha$ is used to range over atomic sentences only. This is clear in the paper (e.g. the semantics of the perception operator), but clearly wasn't in my (necessarily) hurried presentation.

This axiom might be considered unintuitive for the same reason as the negative introspection for belief is considered unintuitive. I don't perceive a rhinocerous in the room, so by default I believe that there is not a rhinocerous in the room. This seems an odd conclusion, as I was not aware/thinking of rhinoceri, this concept was not "cognitively active". In the case of belief, Halpern and (I believe Vardi) introduce an awareness operator and use this to distinguish between explicit belief and implicit belief. We can do a similar thing here to make the axiom more plausible. In our case, awareness would, as suggested, be something like "cognitively active". Thus I may not explicilty believe that there is a rhinocerous (Baby Mercedes, flying saucer, ...) in the room, because I am not currently using these concepts. However, I have the appropriate implicit beliefs, e.g. if asked whether there was a rhinocerous in the room, I would say that there wasn't one (and think that the question was a bit eccentric :-))

Any references on what, for want of a more precise term, I have called "cognitively active", would be appreciated.

Q2. Pat Hayes:

You are proposing a formal common sense theory of perception and belief. However, in your paper there is very little about perception, and what you say is plain wrong. At the same time, there is an enormous wealth of knowledge about perception in the psychological literature, which you do not make any use of. Why should we pay any attention to this paper at all? (Additional debate).

A2. John Bell:

(Full answer not recorded; the answer used a distinction between what the robot sees and what it says it sees).

C2-1. Murray Shanahan:

I want to defend Bell's approach: for the purpose of building robot, I'm not interested in the psychological aspects of perception or in how it's done in people. It is important, however, to understand the relationship between an agent's sensory data interpretation and its beliefs, and that seems to be what the paper is about.

Q3. Erik Sandewall:

You claim to present "a formal common sense theory of perception and belief" (first sentence of the abstract), but in the paper you suggest "that we perceive perceptual attributes". Do you also claim that all major aspects of perception can be characterized in terms of perceptual attributes? Otherwise, can you say something about what aspects of perception are covered and what are not?

(Answer not recorded).

This on-line debate page is part of a discussion at recent workshop; similar pages are set up for each of the workshop articles. The discussion is organized by the area Reasoning about Actions and Change within the Electronic Transactions on Artificial Intelligence (ETAI).

To contribute, please click [mail to moderator] above and send your question or comment as an E-mail message.