Re: simple, plain-English introduction?

B1ackwater wrote:
> On Mon, 19 Dec 2005 10:02:58 -0600, Randy <joe@xxxxxxxxxxxxxxx> wrote:
>
>
>>B1ackwater wrote:
....
> Frankly, I'm not interested in splitting hairs over 'equality'.
> Even 'identical' human twins aren't ACTUALLY equal. I'm much
> more interested in FUNCTIONAL equality. If an electronic
> intelligence performs just as well on a wide range of tasks
> as a wetware intelligence, I'd say they're functionally
> equivalent. To rephrase Turning ... if you can't tell the
> difference then there IS no difference for all practical
> purposes. Hey, no two human brains are 'wired' just the same
> either ... yet they still yeild that 'human' quality. If
> that 'wiring' happens to be in silicon and nanotubles rather
> than greasy membranes that leak chemicals ... so what ?

I used to feel the same way about symbolic AI. But after watching the
field run into one immovable object after another, I'm convinced that
building a simulation of a dog is going to be a lot harder than we
thought. Perhaps it's time to start with the dog and move toward
symbols than start with symbols and move toward the dog.

>
>>Can an "alife" system ever be capable of strong AI? Perhaps. Why not?
>
>
> All e-life NEED NOT have strong EI. I can write a pretty decent
> behavioral simulation of an ameoba without much of a sweat and
> it IS e-life - but it isn't any 'smarter' than a natural ameoba.

I double dare you. If you don't first simplify the amoeba into its most
obvious external behaviors, you'll have a herculean task. Remember, in
addition to its motility, irritability and fission, it'll need to ingest
nutrients and excrete heat and waste. Together, these'll be no mean feat.

>
>>But I'm claiming that alife has failed to prove itself to be more
>>likely to achieve strong AI than any/all other strong AI approaches.
>
>
> You're right, there's no 'proof' per-se. However we've seen the
> failure of some previous top-down approaches to EI - and that
> makes us suspicious. The more 'bottom-up' work we've subsequently
> seen generates much more 'life-like' behaviors with considerably
> less effort. As to whether we can achieve EI via "electronic
> stupidity" remains to be seen. It COULD be as much a failure as
> Minskis work in the 50s and 60s.
>
> Now clearly 'nature' managed it all using a 'bottom-up' approach.
> Of course it (a) took about a billion years and (b) was probably
> more a matter of luck than a necessary outcome of the evolutionary
> process. Add-up all the individual organisms on earth and, for
> all intents, there IS NO 'intelligent' life here ... the
> dead-stupid bacteria and plants so vastly outnumber people, chimps
> and dolphins that we may as well not exist. 'Bottom-up' is proven
> to be a good way to produce lots of stupid organisms, but not
> intelligent ones.

"Would be a good way to produce LOTS of intelligent ones" is a phase
that I would agree with. But "...ANY intelligent ones", obviously I
would not. We puddin-heads are the existence proof.

>
> Of course we can shave off a billion years and a lot of the random
> chance because we're "intelligent designers". I suspect that when
> we do produce something worthy of the title 'electronic
> intelligence' it will have been achieved through a sort of
> "middle-outwards" approach. Take the 'bottom-up' engine, but
> EDUCATE it with as many tricks as we know so it doesn't spend
> a billion years making mistakes.

Sure. But is alife the right basis for that enterprise? Or is alife
just a label that we'll slap on post-hoc since our monster has a human
face? I don't see the principles of alife as forming that critical path
as much as just taking credit for the top-down handiwork of
intelligent-designers-R-us.

>
>>Is
>>alife then *less* likely? IMHO yes, since any successful approach to
>>achieving strong AI must possess many attributes, among them a knowledge
>>representation system that is well grounded, systematic, and extensible,
>>not to mention adaptable and self-repairable. Personally, I have seen
>>no alife systems designed around these inalienable skills.
>
>
> They're supposed to DEVELOP such skills just as our species
> did. The skills are (supposedly) an emergent property of a
> suitably complex and well-structured system. Human infants
> don't have a well-grounded knowledge-representation system,
> but they've got the grey goop and wiring to DEVELOP one.

Yes, that's what's supposed to happen. But do you see systems like
Avida developing some real smarts? AFAIK, I don't. It's equally as
likely that a bunch of Apache servers running PHP will suddenly burst
into flame as Kurt Godel. Not damned likely.

>
>>Historically, it seems to be that alife has been a descriptive term, not
>>prescriptive. It characterizes systems whose behaviors were intended in
>>some way to resemble those of living systems; it doesn't under gird
>>designs whose precepts were intended to gestate into intelligence,
>>eventually manifesting as "a life".
>
>
> I agree with that. "A-Life" was closely associated with cybernetics
> way back when - the imitation of certain gross characteristics of
> living things. The goal wasn't "intelligence" at all, at least not
> any real intelligence. Much of the research went into replicating
> motor/reflexive 'intelligence' because nature seemed to do certain
> complicated things SO easily. This got incorporated into aircraft
> control systems, walking machines even missile-guidance systems.
> I suppose the 'Segway' self-balancing scooter is a direct ancestor
> of that line of reseach.

....
> The 'physical' aspect doesn't worry me - many means can produce
> the same end result. Real-life niceties like self-propagation
> can wait. (indeed, we may NOT want to include that capability).

The physical aspect worries *me*. If no foreseeable mechanism is
imaginable, then we're back to inventing another Ether. I need to see
stepwise progress toward a goal to believe that we're on the right
track. I don't see that with symbolic GOFAI, and while bottom-up
approaches look promising, and even assuming that we're on the right
track, we still have 99% of 4 billion years of evolution yet to go.

>
>>We can debate ad nauseum the criteria for "What is life?".
>
>
> Because there's no real answer ...

No meaningful constructive answer... I agree. But as with any search
algorithm, if you can't devise a precise reward metric, your brilliantly
well implemented A* algorithm is still gonna suck.

>
>>But alive or
>>not, no alife system I've ever heard of has exceeded the performance of
>>a dumbass expert system at delivering human levels of competence in even
>>a simple task (other than implementing a specific algorithm like TSP or
>>optimizing a function). That's a far cry from forming the basis for the
>>potential emergence of human level intelligence from an alife system's
>>superior flock-like behavior or its development of an intraspecies
>>phenotypic trait.
>
>
> Well GEE guy ... we only started this stuff 50 years ago ...
> 'nature' had a BILLION years or more. Show some patience.

Yeah, but I'm not going to live that long. I'd like to think we'll at
least find the a plausible game plan before I have to retire from the field.

>
> There are gobs of money to be made in strong, or even weak,
> EI. This will MOTIVATE people. Hell, how much money did that
> first generation of automatic floor-sweeping robots make ?
> Hundreds of millions I'll bet - and the things were purely
> reflexive rather than having any smarts.

Yes, but the money is available only for undertakings that can deliver
success in the short term. If the problem is too hard or the technique
fails to reward the investors quickly enough, no matter how great its
potential, the game will be called on account of rain before our slugger
gets a chance to bat.

>
>>If an agent appears intelligent then it must have intelligence ?
>
>
> If you give it a slate of everyday problems and interactions
> and can't tell the dif between it and the guy next door then
> yes, it's "intelligent" AND "human-like".

Fool me once... and you're smart enough to fool me every time? I don't
think so. Before an immature human is deemed "useful" by our
establishment (that is: worthy of employment), that person has survived
many trials: some 22 years of academic study and nearly innumerable
tests therein. Only then is an employer ready to presume their utility
in serving a practical/commercial enterprise. So it must be with
intelligent systems. One simple test (like Turing's) tells us little or
nothing of a system's intellectual capacity. And when that test
measures not performance, but the audience's gullibility, it's time to
put away childish things and devise some grown up criteria.

When a system can pass *both* a MMPI and a 3rd grade human academic
competency exam, only then will I imagine that Pinocchio will one day
grow up to be a man.

>
>>No.
>>That makes no more sense than saying, "If a thing is lively, then it
>>must be alive." I think the latter describes alife much better than any
>>alife systems have recapitulated even the most basic forms of ontogeny
>>or phylogeny.
>
>
> We've got to be careful with how we use language here. Most
> of the terms we've been using are antiques left over from an
> age of ignorance and common-sense certainty. "Life", "lively",
> "intelligence" ... they're not exact enough to use when
> speaking about modern technologies.

Agreed. That's my point in denigrating "life" or alife as having
meaning other than as a phenomenological label. Just because there's
smoke doesn't mean there's fire. Perhaps smoke is just an ambitious way
to describe the fog we're in. That's all I'm saying about alife.
Whether there can be fire, or whether we're on the right track to
creating combustion is another matter. In the end, we shouldn't
conflate pretty smoke as being evidence for even a flicker of intelligence.

....
>>
>>Is a thermometer conscious? By your definition, if it can take its own
>>temperature, it is.
>
>
> The thermometer is not conscious because it can't AFFECT its
> own temperature. A CPU that's wired-to a thermometer may be
> able to register the slightest blip on the 'consciousness
> meter' because it can be programmed to RESPOND to the temperature
> data in various ways. Life isn't just input, it's input->process->
> output.

So a thermostat that's attached to a furnace *is* conscious? After all,
it can affect its own temperature... :-)

>
>>> 'Consciousness' doesn't seem to have a 'threshold' - it's
>>> something that just creeps into the picture a little bit
>>> at a time as brain complexity increases. Dogs are LESS
>>> 'conscious' than people, but not UN-conscious.

You've admitted that self awareness alone is insufficient to confer
consciousness. You've said that a system must also be able to alter its
consciousness (i.e. perceptions). I think we could continue in this
vein for quite some time, defining and refining the nature, necessity,
and sufficiency of the nature of consciousness, especially in its role
as a prerequisite for intelligence. But I'm not sure it would be time
well spent. How self-aware must something be in order for it to be 1)
alive or 2) intelligent? And how would we know whether our conclusions
were right?

I think a great many folks would be thrilled to build a system with the
"intelligence" of many creatures that appear not to be conscious. Or
should we conveniently presume a continuum of ever rising amounts of
consciousness from the lowly paramecium up to the lofty human, not
because we observe it to be so, but because it's so damned hard to
characterize consciousness and perhaps harder still to quantify it? If
we do, I can't see what value an unsubstantiated assertion can have.
Not only will we be unable to judge whether our synthetic system
approaches a natural system's level of consciousness (since we can't
measure it), but we can't even decide how or whether consciousness
contributes to intelligence in the first place.

I just don't see the value in inventing yet another phlogiston. IMHO,
it's high time to get down and dirty (grounded and methodical). Let's
help intelligent systems to develop perceptual models of the world, and
then use IQ tests to measure the development of situated intelligence.
Have it run mazes, solve logic problems, recognize patterns, and
manipulate physical puzzles. That's how we measure the IQ of rats.
What's good enough for rodentia is good enough for cybernetia.

>>
>>Unless you can ground a concept fundamentally before extending it
>>systematically, you're destined to go round and round toying with it
>>forever -- like your metaphorical dog, forever taking measure of its own
>>tail.
>
>
> That's one of the problems with this subject - language often isn't
> up to the task. 100 years ago, 'conscious' was 'conscious' - and
> we all "just knew" what the term meant. Now 'consciousness' gets
> subdivided and subdivided again - it's become a TECHNICAL sort of
> term and an inadequate one at that. The more deeply we stare into
> our own image in the mirror, the more difficult it will become to
> describe what we see there.

I think that's the curse of reductionism. You can't know what you're
talking about before you talk about it. But ya gotta start somewhere.

Randy
.