Path: utzoo!attcan!uunet!samsung!uakari.primate.wisc.edu!xanth!mcnc!rti!ntpdvp1!sandyz From: sandyz@ntpdvp1.UUCP (Sandy Zinn) Newsgroups: comp.ai Subject: Re: Plaiting a Plexus of Processes (Was: Re: Inferring meaning from use ...) Summary: when is an error not an error? when it's a difference that makes a difference Keywords: Pribram, Minksy, representation, bouncing Message-ID: <373@ntpdvp1.UUCP> Date: 11 Apr 90 06:59:55 GMT References: <2080@skye.ed.ac.uk> <352@ntpdvp1.UUCP> Organization: Northern Telecom DMS-10 Div., Raleigh, NC Lines: 139 > > (Sandy Zinn) writes: > >>If one assumes that infants have rudimentary patterns of squirms which > >>direct their interaction with the world, then "errors" accumulate through > >>the use of comparator functions on input squirms. > > (Ken Presting) writes: > Bateson's model is probably flexible enough to do the job, but the notion > of natural systems making "errors" is unsettling to me. We have agreed > that the knee-jerk reflex makes perfect sense in its natural setting. I > have argued that *every* natural phenomenon makes sense in its natural > setting. This would apply to biological phenomena as well as to physical. With your emphasis on *setting*, which I am set on myself, I understand your being unsettled. But fear not, my position is more subtly bizarre than this! > "Error" is one of the normative concepts, and I think AI is trying to > explain how a system which always makes perfect sense in one system of > descriptions can be said to make mistakes in another. In fact, this is a pretty good description of intelligent activity in general. > For example, a > program always does what its instructions specify - in the frame of > reference of the CPU, there are no software bugs. Even invalid > instructions are handled according to plan, with a program check or trap. > That's what's supposed to happen, and it does. But the program is also > part of a system of descriptions defined by its specifications. *That* > is the frame of reference in which the term "bug" has meaning. Ah, that is the sort of frame of reference I was assuming. Only, in this case, the program specs are in the infant, and the "bugs" are in the operation of the environment. Sometimes, of course, there *are* errors in the specs, which makes the detection of bugs problematic... > Also, I'm not sure how to make "expectations" innate. Certainly there is > no problem with innate reflexes or instinctive behavior, such as suckling, > or crying when hungry. Would you say that an error is detected when > sucking on a pacifier does not eliminate hunger pangs? Yes, I would, I do, and I did. A fine leap on your part! > I might go > along with this, actually. I would point out that the "expectation" > in this case is defined by all three plaits of the plexus, and is not > strictly epistemic. (It's implemented, I would say!) I might go along with this, potentially. (I did.) > "Differences" bother me less than "errors", because there is no normative > judgement in saying that two events are different. But before two > things can be compared, they must each be identified. Depending on how > it's used, "difference" can be intentional... > Detecting difference can depend on *reference*. I mean different = divergent from the specs. Intention unintended. > In the suckling case, we can describe the infant as a biological system > which implements all its expectations, not with a squirm for each one, > but all of them in the structure of all its squirms. Unsatisfied > expectations need not be individuated to add to the squirming. No. This is too simple. I don't think there is a nice set of individual squirms, or carrying circuits, or synaptic wave-form processors, for each items in the specs. But neither is it global wiggling: there's a can of worms here, but it's not in this neophyte mind, but rather in our neophyte relationship to our wanting to know. (She goes for a can opener.) * * * * * * * * * * * * * * * * * * * * > (Stephen Smoliar) writes: * > >Where Minsky may depart from Bateson and Pribram, however, is in his desire to > >push the processing of differences to a "meta-level:" * > > * > > The ability to consider differences between differences is * > > important because it lies at the heart of our abilities to * > > solve new problems. This is because these "second-order- * > > differences" . . . * * > Here's another apparently logical concept appearing in a model of the * > mind. * * > >I think this comes very close to Edelman's model of memory as RECATEGORIZATION > * > Recategorization sounds more like *learning* than *memory*. * * * * * * * * * * * * * * * * * * * * * Now, here, HERE, is where I think our wrestling mat lies. A difference that makes a difference (Bateson's rough & tender phrase), and Minsky's differences between differences, may be logical concepts, but have mercy! These are also ideas that unfold a fantasy. Here's the scenario. Streams of information coming in, striped, let's say, by perceptual physiology. Brain stem collation centers add a few more stripes, in a different color. The info gets split into several different channels, so that different sets of squirms get bounced off of it, overlaying the stripes with various grids. (geez this is risky!) The squirms that get bounced off are *memories of differences that make a difference* -- applying the grids isolates the difference to a striped square -- you've got analog info going digital -- and that digitization is the breaker switch for selection. The "paths" of selection are where the specs are encoded -- and some of the bounces of the selections get fed into squirm-bounce-generators, so that they become filters for the next round...plexus plaits, plied! Now, differences of differences -- ah, here's the fantasy. Suppose one day a new squirm bounce sector arose, one which had the ability to reach in on the path of nearly any little striped square, and *magically* turn it off or on, regardless of pre-existing relations in the braid. Probably it would divert half the stream in the little striped square to its own lair, leaving the rest to flow. Of course, rather than targeting one little square, it would target some pattern of squares, maybe having some generic patterns for its initial stock, to give it a better than average chance of success at diverting patterns that would -- I hesitate to say "make sense", because then of course I presuppose some context for meaning, but there's room here I think, and maybe a need, for some prior organization -- but eventually these diversions are going to refine their technique, and what we have, ladies and gentlemen, is selections of categories. All squirm-made. Now the child can juice up that renegade sector, let loose a bounce, and say of the diverted squirms, "This is me. That other stuff ain't." Pure fantasy. But the child gets on in the world, bouncing the bounces off the bounces, squealing with delight when the bounced diversions have real similar patterns. AHA! But I didn't know to go looking there except I knew I needed a difference of differences. A hierarchy. A metadifference. Sets of selections to select. A normative or two. Concepts as binaries. Binaries as concepts. All our maps select features, contain clues. So all our discussion, with all our different terms, and agendas, creates real movement. (Fantasy always has moved us.) @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ Sandra Zinn | "The squirming facts (yep these are my ideas | exceed the squamous mind" (except tonite) they only own my kybd) | -- Wallace Stevens -- me