Artificial Intelligence

What threshold does a chess program need to break to say it has achieved AI? The search for AI is making big strides. Almost daily we hear of the latest advances. Wondering though... What criteria is used to proclaim: this computer is simulating human behavior?

One thing is for sure: the measure of AI won't be the computer's performance rating, unless it's a *very* big and *sudden* jump in rating, like suddenly into the high 3000s. That's where computer people have gone wrong all these decades: doing the same things in basically the same way they've always done them, based on math, algorithms, digital computers, and brute force. It stands to reason if you want to produce artificial intelligence, you first have to do something particularly intelligent that hasn't been done before. All of the academic system, financial system, and even normal science in every advanced country is against really novel ideas or really free thinking. That's what happens when you put money and other superficial, short-sighted goals ahead of long-term progress: not only do morals and social systems suffer, but also science.

What you're asking is how does one test or measure intelligence, which is not known, since "intelligence" does not even have a generally accepted definition.

----------

(p. 505)

   "What happened," he continued, "is the universities trans-

formed themselves in the 1980s. Formerly bastions of intel-

lectual freedom in a world of Babbittry, formerly the locus of

sexual freedom and experimentation, they now became the

most restrictive environments in modern society. Because

they had a new role to play. They became the creators of new

fears for the PLM. Universities today are factories of fear.

They invent all the new terrors and all the new social anxi-

eties. All the new restrictive codes. Words you can't say.

Thoughts you can't think. They produce a steady stream of

new anxieties, dangers, and social terrors to be used by politi-

cians, lawyers, and reporters. Foods that are bad for you. Be-

haviors that are unacceptable. Can't smoke, can't swear, can't

screw, can't think. These institutions have been stood on their

heads in a generation. It is really quite extraordinary.

   "The modern State of Fear could never exist without uni-

versities feeding it. There is a peculiar neo-Stalinist mode of

thought that is required to support all this, and it can thrive

only in a restrictive setting, behind closed doors, without due

process. In our society, only universities have created that--

so far. The notion that these institutions are liberal is a cruel

joke. They are fascist to the core, I'm telling you.

Crichton, Michael. 2004. State of Fear. New York, New York: Avon Books.

Exactly. That will work. Just first elucidate in detail the method of implementing "human-style Search and Evaluation." 

Just make a computer do inlogical decisions, and it will be able to outperform humans in all way and be undestructable, like terminator.

I agree academic institutions are against novel ideas and free thinking. Insightful points you make Spod. Such as the definition in itself is not generally agreed. That being said ... do not programers have indicators to proclaim .. aha AI! I understand many will dispute the claim. Somehow it seems inevitable two sides will emerge.

No, they don't, at least no formal method I've ever heard of, and I'm extremely familiar with that field. I believe the best you could do now is to give human IQ tests to machines, but that has already been done with overall poor results, and IQ tests themselves are flawed and written by more people who don't know what intelligence is.

----------

(p. 144)

   In sum, WAIS-R scores constitute a narrow slice of intellectual

functioning. The test does an outstanding job of evaluating the verbal

component of intelligence, but falls short in the holistic areas such as

spatial thinking, pattern recognition, and figure completion. To borrow

an analogy from music, the WAIS-R tests the ability to reproduce and

recognize simple melodic lines but does not ask the subject to know or

produce complex chords.

   The WAIS-R is also totally oriented toward one mode of thinking,

namely, convergent-thinking. The entire focus of each question is to

converge on a single correct answer. But it is also a component of

intelligence to start at a single point and diverge in as many directions

as possible, as when answering the question "What would be the con-

sequences if clouds had strings hanging down from them?" This form

of thinking, called divergent thinking (Guilford, 1967), requires fluent

thinking for optimal performance. It is akin to creativity and, while

correlated with convergent thinking, the conceptual differences are

obvious. Not one question on the WAIS-R even approximates measur-

ing divergent thinking, one more indication of the restricted content of

the test.

   The narrow focus of the WAIS-R can be further highlighted by

reference to Gardner's theory of multiple intelligence. Gardner (1983)

has identified seven forms of intelligence, only two of which are mea-

sured to any extent by the WAIS-R. The two that are valued most

highly in modern western culture, and that are therefore incorporated

(p. 145)

in the WAIS-R, are linguistic and logical-mathematical intelligences.

But, according to Gardner, there are five other kinds of intelligence

that are as important: spatial, musical, bodily-kinesthetic, and two

forms of personal intelligence--interpersonal (knowing how to deal

with others) and intrapersonal (knowledge of self). From this expanded

list, the WAIS-R touches briefly on spatial intelligence (Block Design)

but completely ignores the rest.

   While both logical-mathematical and linguistic intelligences are

important in the United States today and thus appropriately empha-

sized on the WAIS-R, other types of intelligence may come to

be valued some day. For example, in Japanese society, interpersonal in-

telligence--involving the ability to work well in groups and to arrive at

joint decisions--is highly valued and so may it be here one day. The

WAIS-R is a good measure of those dimensions of intelligence cur-

rently valued in the United States and many other Western industrial-

ized societies, but in its narrow emphasis on left-brain thinking, it

overlooks the many alternative modes of intelligence outlined by

Gardner and others.

   A content analysis of the WAIS-R reveals that it is almost totally

a measure of things already learned. In general, the subject is not asked

to learn anything new and apply it to ensuing problems as the test

progresses. Thus, there is little in situ or on-the-spot learning.

Gregory, Robert J. 1987. Adult Intellectual Assessment. Boston: Allyn and Bacon, Inc.

I read of one answer.. when a computer shows "an original thought" it can be said to have AI. But then it's said there are no new ideas under the sun (which I disagree with)

added: as related to chess... when a computer makes a new, or original move it can be said to achieve AI. But who or what decides "original"?

Agree Sqod.

Intellgence has som many things to it - for instance curiousity. I never met anyone I'd consider intelligent that wasn't curious, i e. taking pleasure in understanding and learning new things.

Don't think AI are curious at any level ATM lol.

Another one is your ability to adjust longheld views in the face of new evidence. I guess this would be no problem for AIs since they dont have any ego. But then on the other hand, if they changed their views every time they saw some new evidence, maye they would get sort or erratic and weird.

you can just combine random words and it would be original.

dancing green yellyfish. that's an original thought right there. not middle of the road for sure. 

There it is again: humans somehow know what is relevant to a problem, and what isn't, but we simply don't know how to program that or even model it.

----------

(p. 115)

   The second prong is by way of cognitive folding. Horgan and Tienson

claim that cognition cannot be partitioned into isolated domains. Any

part of commonsense knowledge might be called upon in dealing with

(p. 116)

any of a vast number of cognitive tasks. Thus any fully intelligent agent

must be able to see what is relevant in a given situation, and to recall the

relevant information. For example, children easily fold together restau-

rant scripts and birthday party scripts to form reasonable expectations

when they go to a birthday party at a restaurant for the first time. They

don't get it all right, but they do have a fair idea of what to expect. They

are able to fold two scripts together. Nobody has the slightest idea of

how to program a computer to know in general what's relevant to what.

Is it because no one has tried? No. Schank and his colleagues have worked

on the folding part of this problem for years (Schank and Abelson 1977;

Schank 1980). Or could it be that symbolic AI is inherently incapable

of such folding? Anyway, for now, humans do it well; computers, almost

not at all.

Franklin, Stan. 1995. Artificial Minds. Cambridge, Massachusetts: The MIT Press.

 I don't think the question is too different from asking when a child becomes an adult. The two concepts bleed into each other. Practically, you can't really give a person a gold medal for finally having become an adult in all things and in every respect. Most you can do (and actually more relevantly) is say things like "Yeah, this 19yo kid is mature enough to do a good job at my pizzeria. Let's try him out."

P.S.: I'm sure there's a pizza making robot out there. My guess would be that it's a Japanese project.

Spod... very few humans know what is relevant much less posses original thought regarding observations and measurement. I see the difficulty involved. It seems you are skeptical that AI can ever be achieved? I certainly hope that to be the case.

I believe defining intelligence and defining adulthood status are totally different problems. You're describing the ubiquitous principle that in the real world there are almost always gradual transitions between states or locations or times. (Only in the quantum world, which is a world alien to our common sense, do objects not behave that way.) The problem with AI is that it is must be based on a totally different way of interpreting incoming real-world data than we currently know how to model, so it's not a matter of degree. Faster processors, more processors, more data, and better algorithms are simply the wrong approach, I believe. That's why I say that *maybe* a sudden *jump* in Elo rating might indicate intelligence, because it would suggest that something fundamentally different is being done to process the incoming information in an intelligent manner.

What approach would you suggest Spod? One that begins with organic matter perhaps?

No, in fact I claim I already know how to model human-style thinking. Last year I submitted a $1 million+ proposal to an agency seeking breakthrough ideas in AI, and the proposal was rejected, I believe without even considering anything I wrote, I believe because my primary collaborator, a Ph.D. in the field, lacked one of the additional qualifications they thought was absolutely necessary to be able to do satisfactory research. That in turn was because the investors (read: "money rules"), who probably didn't know a darned thing about the field, put in additional qualification to make themselves feel safer about how they were spending their money. They put in that qualification one month *after* they requested proposals! That agency ended up sending an e-mail rejection to the wrong person on our team, and to the wrong address besides, and never responded when that team member requested a direct response and clarification. Go back to my first posted comment about universities above. I believe the foundations of strong AI are already here, but the existing system is incapable of bringing them to light. I'm very upset about all this, of course, but don't interpret my cynicism for the system to be cynicism about the possibility of strong AI.

Hey guys... very informative stuff. Thx. Hope the forum continues on. The sharing of information is appreciated.

So a last question today for Spod... when your program does get developed ... you must have fundemental criteria to declare success. I understand you may not wish to discuss this... for propriatory reasons. Your concept seems an original one.

Yes, yes, and yes. First, it's not a program because programs are merely algorithms + data structures, and have fundamental limitations, mostly based on doing only what they are told in advance (including being told how to learn), which means they cannot inherently break out of their preset programmed limitations on their own. Second, yes I have my own definition of intelligence that I had intended to use to measure the system's intelligence. Third, yes, it's proprietary. Fourth, yes, it's an extremely novel approach, which is one reason I'm fairly sure it was exactly what the agency really wanted.

Hey!!! 3 exclamations. I sincerely wish for your success. Please don't get discouraged... all great ideas get rejected several times over.

"Faster processors, more processors, more data,"

-- yeah ok. Probably true only after a certain level has been reached. It hasn't yet.

"and better algorithms are simply the wrong approach, I believe."

-- your belief stands in conflict with actual progress with AI.

"The problem with AI is that it is must be based on a totally different way of interpreting incoming real-world data than we currently know how to model"

-- not sure where you got that idea from. There's a difference between AI and magic.

"That's why I say that *maybe* a sudden *jump* in ELO rating might indicate intelligence"

-- That would be a curiosity but not particularly interesting or impressive for modern AI research.

"because it would suggest that something fundamentally different is being done to process the incoming information in an intelligent manner."

-- You mean like a different algorithm? Sure. But not a mystery to the programmers who developed the algorithm. Again, AI isn't magic. It's an engineering problem.

Different question which might help you think this through. Are cars 'good' cars?

i) If you think they are then when did cars become good? Can you think of a turning point for bad cars to good cars? Was it obvious before the turning point? Did people just prior to that see a previous advance as THE turning point?

ii) If you don't think cars are good then maybe you'll want to set the turning point at flying cars? If we're playing that game then I'd prefer to set the turning point at warp-drive-enabled, time travelling cars with laser cannons. (ideal for the school run). Just sayin'.