Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!isi.edu!smoliar From: smoliar@isi.edu (Stephen Smoliar) Newsgroups: comp.ai.philosophy Subject: Re: What does intentionality have that AI doesn't..... Message-ID: <17153@venera.isi.edu> Date: 14 Mar 91 19:11:37 GMT References: <13503@ccncsu.ColoState.EDU> Reply-To: smoliar@venera.isi.edu (Stephen Smoliar) Organization: Information Sciences Institute, Univ. of So. California Lines: 96 In article <13503@ccncsu.ColoState.EDU> petersja@debussy.cs.colostate.edu (james peterson) writes: > > >What it is about "intentionality" the lack of which would impede the >implementation of intelligent behavior artificially is related to the >problem of "relevance." How is it that intelligent creatures are capable >of selecting from their manifold inputs that portion which will be considered >as important, and that which is to be ignored? How is it, moreover, that >intelligent creatures are able to assign relative values to parts of >the environment related to importance, and readjust these relative values >as they procede? > >Frames and scripts, it seems to me, gloss over this difficulty by assigning >relevance in advance. The hard problem is to account for how relevance >comes about in the first place, and how it develops... > >What makes assignments of relevance possible on an ongoing basis is >*motivation* --- things, parts of the environment, are relevant, important, or >interesting precisely in the context of some *purpose* (if my purposes change, >so does what is relevant); relevance is thus a function of our >reasons (or motives) for acting... Humans act for reasons, but for >reasons which do not compel or necessitate (reasons are not causes); being >free to act according to one's own plans, plans of one's own authorship, >and to change those plans on an on-going and flexible manner is what >I believe intentionality has that is needed to implement intelligence. Searle >says that intentionality and intelligence are tied to "causal powers" -- >and this is what I take him to mean -- the ability to cause actions for >reasons independent of nature's causal nexus, in a word, motivation. > Maybe we had better go back to how Searle claims to define intentionality. Unfortunately, his definition is relegated to a footnote in "Minds, Brains, and Programs:" Intentionality is by definition that feature of certain mental states by which they are directed at or about objects and states of affairs in the world. Thus, beliefs, desires, and intentions are intentional states; undirected forms of anxiety and depression are not. Matters such as relevance and causality are secondary. The primary issue is the assumption that there are these mental states and then there are world states. We can talk about things being IN world states at a casual intuitive level, so we assume we can do the same about mental states. Intentionality accounts for the ability to translate between what is IN a mental state and what it IN a world state. Looked at in this light, we can then ask if we can have machine states which admit of a similar translation. Such machine states could be said to have intentionality, which, supposedly, means, that their machines would be capable of the sort of "understanding" which Searle argues is lacking in symbol manipulating systems. On the other hand are we in a position to argue that we can build symbol manipulating machine which LACK such powers of translation? As I said in my last article, I do not think the real issue is one of what machine states have or lack. Rather, the question has to do with the relationship between machine state and machine behavior and with the question of whether or not it makes sense to talk about disembodied mental states. In other words an agent can only HAVE mental states in the first place by virtue of certain properties of its BODY. If you try to abstract away the body (as Turing assumed in his initial paper on artificial intelligence), you lose the mental states, too. Does this make sense? Actually, a more appropriate question would be: CAN this make sense? I think it can. However, the issue is not so much whether we are talking about symbol manipulation as to how we choose to look at machine states. If we view a machine state as a configuration of bits in memory--something we can freeze in time or even copy from one physical machine to another--then we may run into trouble. Such machine states are essentially divorced from the machine itself . . . particularly any peripheral hardware concerned with sensors and effectors. If, on the other hand, we imagine a more dynamic machine in which there is constant interaction between those sensors and effectors and the state of the machine, then it may no longer make sense to try to identify that machine state with something like a pattern of bits in memory. The bits are always changing, and taking away the dynamics of the situation would amount to inducing brain-death. Now this may be naive, but at an intuitive level there seems to be no reason why such a machine cannot be built from components which are basically symbol manipulating systems. Thus, symbol manipulation does not appear to be the critical issue. The critical issue involves the nature of the processing concerned with the management of sensors and effectors and the assumption that such processing is highly dynamic. If we allow such dynamic qualities into the system, we should be able to play the game Minsky has laid out in THE SOCIETY OF MIND, building a system which manages those sensors and effectors up from relatively low-level processing components, very much in the spirit with which Brooks builds his robots. The question is not whether or not our machines have intentionality but whether or not they have bodies through which they interact with what Searle calls "states of affairs in the world." -- USPS: Stephen Smoliar 5000 Centinela Avenue #129 Los Angeles, California 90066 Internet: smoliar@venera.isi.edu