Re: Sentience

From: Brian Mastenbrook (NOSPAMbmastenbNOSPAM_at_cs.indiana.edu)
Date: 04/11/04 

Date: Sat, 10 Apr 2004 23:18:41 -0500

In article <kN6dnUHs98lsAeXdRVn-vg@golden.net>, Cameron MacKinnon
<cmackin+nn@clearspot.net> wrote:

> Brian Mastenbrook wrote:
> > I would extend the cargo cult phenomenon beyond "just science". I've
> > seen entirely too much cargo cult programming - putting together parts
> > of programs without any real understanding of what's going on. When
> > you're helping a student who has close parens and open parens littered
> > randomly throughout their Scheme source, it's pretty obvious that the
> > student is not reasoning about how Scheme works, but putting together a
> > program that has an appearance such that it /might/ work.
>
> Infants don't learn languages by reasoning out that every sentence needs
> a (possibly implied) noun phrase and a verb phrase. They perform a lot
> of experiments with no scientific method. Mimicry and feedback seem to
> be enough.

Enough... with hardware that supports it. Our brains are simply wired
up to rapidly obtain the necessary information for language learning.
But there is a difference between doing that and understanding the
wiring itself. Our genetics do really provide a bootstrapping procedure
that we've got to understand now.

> Computer folks know that, because computer languages contain almost no
> redundancy (as used to resolve ambiguities), aren't DWIM and have
> pedantic compilers with truly awful diagnostics, learning one's first
> through trial and error is painful, and so syntax is often taught
> rigorously. But that's all hindsight and deep insight. Your students'
> strategy seems quite reasonable if their only experience is in
> biological languages with good feedback from their "conversation partner."

It's reasonable up until the 52nd time I explain what s-expressions are
and point them at a tutorial explaining the rules. At that point they
really are just coasting instead of thinking.

> What, exactly, is the cargo cult's mistake? Seabees came and built
> runways and control towers, men put funny headphones on, and the planes
> came. Mimicking this looks ridiculous TO US, because we have superior
> domain knowledge of aeronautics. Absent that knowledge, we could say
> that they err in persisting for five years when no planes have come, but
> that suggests that the cultists are to somehow know how long to wait for
> results. Given that the original was a one-time event for a culture
> which (I'm guessing) had an oral history stretching back many
> generations, even a several-generations-long experiment might not be
> unreasonable.

I don't think Feynman's point was to attack the cargo cult at all. He
was drawing an analogy between why it didn't work and the same type of
pseudoscience that doesn't work in our culture. In our case, we have
many methods which have been developed to enable us to come to an
understanding of something which appears magical. Since we don't yet
(and concievably won't for many many years due to the seeming exclusion
of accurate and large-scale measurement) understand how the embodied
mind actually processes information, that means we need to apply these
processes to the understanding of intelligence itself.

Without following scientific or systematic philosophical inquiry into
the nature of the general concept of intelligence and its
representation as part of the human mind, we cannot convincingly assert
that our experiments will bring in the cargo planes of intelligence.

> [If you'd have asked me yesterday whether I'd defend cargo cultists, I'd
> have been insulted at the implication. I never really thought about it
> before, just thought what Feynman wanted me to think.]

I don't think Feynman wanted you to think anything about them, but
instead about the advocates of pseudoscience in a scientific society.

> I meet people all the time who seem to have only the most tenuous grasp
> of logic. In fact, I'd hazard that, for the majority of any given
> society, Aristotle and Descartes might as well never have existed. It's
> not that they are completely illogical, but they certainly don't operate
> through rigourous reasoning, and often can't see the flaws in the
> systems they believe.
>
> So the interesting question is, when seeking artificially intelligent
> behaviour, are we looking for computers to act like the best of us (in
> which case, perhaps "Artificial Brilliance"?) or the rest of us?

The fact that rigorous logic is learned is actually a valid point in
the study of intelligence. It suggests that it might not be the right
level to describe what's going on. I personally focus much more on the
symbolic aspects of intelligence, with a slightly peculiar connotation
of symbol: something which has identity. I try to stake out a middle
ground between the subsymbolicists and the logicians.

> Explaining human behaviour, in isolation, in terms of neurons seems very
> difficult. But if we start at the lowly cockroach, we see something that
> wouldn't seem to take too many neurons to emulate.

But is it useful? My contention is that the human brain is not really a
well-designed device and is more of a kludge of cognitive ability on
top of a neural net than a reflection of some innate cognitive ability
in neurons.

> My comments on game AI weren't meant to imply that it represents the
> future, merely to say that some areas of AI have been advancing, because
> there's both quasi-objective metrics and motivations for the people in
> that field. AI people seek to get the right fitness criteria for the
> training or population culling within their experiments, but I think
> that it's equally important to have appropriate fitness criteria for the
> researchers themselves, otherwise they end up playing in their own theses.

Unfortunately I'm not convinced that those metrics are actually
measuring intelligence. There is a lot of fascinating stuff going on
there, but I really would hesistate to put the label of AI on, say,
genetic programming. It's not a slight to GP, but just a difference in
classification.

> Military (and eventually civil) logistics is another similar area.
> Results were wanted, money was spent, results were gotten. Since funding
> domain specific intelligence (or solutions) research has a payoff,
> whereas general intelligence is a money pit, domain specific is where
> the progress is being made.

I think it's shortsighted investment to only fund practical and not
basic research. For instance, basic physics research gave us the MRI,
and soon sustainable energy-producing fusion.

> Anyway, we may rapidly be approaching (or worse) the limits of my
> insights in this area.

Not at all; these are good thoughts.

> From your comments, I'd say you follow the progress, er, machinations,
> of the AI community. Do you have any pointers for the uninitiated to
> collections of stuff that's more toward the seabees' end of the runway
> than the natives'?

Actually, it's worse than that. I publish in AI. My homepage has links
to my papers, but they aren't necessarily terribly well written,
particularly because of the need to slip past reviewers and masquerade
as useful research.

While I don't claim to be a classicist in any strict regard,
http://plato.stanford.edu/entries/language-thought/ gives a good
history on some of the previous work in the field and the feud with the
connectionists. 

-- 
Brian Mastenbrook
http://www.cs.indiana.edu/~bmastenb/

Relevant Pages

  • Re: The Reasonable Minority
    ... physics, chemistry, biology, and medicine. ... essential to understanding evolution and evolutionary processes. ... I have some scholastic studies and work history in math, ... I prefer reading, considering, evaluation by reasoning, and openness ...
    (talk.origins)
  • Re: Evolution means people werent planned.
    ... turned you into a parsimonious Scot! ... I'm a Yorkshireman now resident in Scotland, and with Scots ancestors so we share some characteristics, I think! ... But, as I say elsewhere, different disciplines have differnt rules and not all are amenable to the same kind or level of reasoning. ... merely saying a claim is not in accord with scientific understanding means very ...
    (uk.religion.christian)
  • Re: What it takes to be a revolutionary thinker
    ... Jim Greenfield wrote: ... If you mean a professional degree in physics, ... hard questions whether the student has learned it sufficiently to ... "Secret signs and dodgy math" are often required for UNDERSTANDING. ...
    (sci.physics.particle)
  • Re: What it takes to be a revolutionary thinker
    ... Jim Greenfield wrote: ... If you mean a professional degree in physics, ... hard questions whether the student has learned it sufficiently to ... "Secret signs and dodgy math" are often required for UNDERSTANDING. ...
    (sci.physics)
  • Re: What it takes to be a revolutionary thinker
    ... Jim Greenfield wrote: ... If you mean a professional degree in physics, ... hard questions whether the student has learned it sufficiently to ... "Secret signs and dodgy math" are often required for UNDERSTANDING. ...
    (sci.physics.relativity)