Artificial Intelligence – A fool’s Dream?

...18th century Austrian Baron Munchhausen...

First it would be wise to agree on what we mean by intelligence in a given context.

The Turing test is only one measure of intelligence.

If we use our own intelligence as a yardstick, then we have to accept that artificial intelligence creations will make mistakes and errors in judgement: are we prepared to do that?

Another problem with the Turing test is that it factors in some level of "socialization". It typically takes a human the equivalent of 9 to 12 waking years (or 12-18 actual yrs) of experience to achieve a reasonable level of language skills, maturity, dexterity and socialization. We jump to false conclusions, are fooled by optical illusions, forget things, etc, in short, not what we are looking for from an "intelligent" machine.

At the present limited examples of artificial intelligence are all around us, and "Fuzzy Logic" deals well with making decisions from "vague" inputs.

As far as:

"Artificial Intelligence is when a program can overcome a precedence, which, in normal terms, is considered an illegal input. To overcome a precedence, the program would have to be able to re-write itself while running, ..."

I don't agree at all.

Regards, Greg

1. "artificial intelligence creations will make mistakes and errors in judgement" right; only for us to err usually mean some embarrassment and a new drive to "learn" or re-write our priorities or procedures, to it on the other side, an illegal input, is any situation it's not yet programmed to handle, thus, becoming idle in an infinite-loop, may be the obvious consequence, unless it can be taught, i.e, re-write it's set of handling procedures.

This is exactly the point: the ability to learn, is re-programming yourself upon stumbling a precedence, and i think we can all agree, that the ability to actively learn, is exactly what separate us from an automat, be it's complexity as it may: Deep-Blue did not prove it's more intelligent than it's human counterpart, all it proved is that chess is a calculation. immensely complex, astronomically complex, but a clear-cut, zero-grays calculation.

2. "limited examples of artificial intelligence are all around us, and "Fuzzy Logic"" Fuzzy logic algorithms are classified by their creators as "Heuristic Search-And-Decide" automatons, not much different than a double or triple bet in lottery or Toto games.

yuvalmate,

First of all, we do not deal with the unexpected (what you call illegal inputs) by any reprogramming at all. What you are incorrectly calling reprogramming is merely the incorporation of new knowledge into an existing overall "program". Your logic escapes me in no small part because you are using some of your own terms and definitions.

Programs have long been dealing with "illegal inputs" as part of their error handling routines, and any case of falling into an infinite loop is a result of a programming bug or weakness, having nothing to do with the specific input itself. Software that modifies specific portions of its operation (calling different subroutines, changing database entries and/or definitions, reassigning variables, moving pointers etc) on the fly is also nothing new. Additionally, hardware and/or software that mimic neural networks have been implemented to various degrees and are being worked on as we speak.

"Fuzzy logic algorithms are classified by their creators as "Heuristic Search-And-Decide" automatons, not much different than a double or triple bet in lottery or Toto games."

Comparing Fuzzy Logic to a double or triple bet in lottery or Toto games displays more ignorance of the subject than I care to deal with here.

I respectfully suggest you stick to standard terms and definitions in their appropriate context, study the topic further, and read a few books on Cybernetics, Fuzzy Logic, and AI (and programming too for that matter), since your concept of having to rewrite parts of a program whenever an unforeseen input occurs is deeply flawed.

On the other hand, much progress is currently being made in AI by stepping back and looking at how the "primitive" brains of lower animals and insects process information. As to our own behaviour, it is not at all unknown for humans to "become idle in an infinite loop" when faced with unexpected "inputs".

Finally, the possibility exists that you are very knowledgeable on this overall subject but are hindered by a lack of fluency in English which causes confusion in interpretation on my part. If that is the case, I apologize, but that doesn't mean I understand you.

Regards, Greg

The shape or state of my English presentation, and the extent of my knowledge, has nothing to do (believe it or not) with the principles discussed here: my poor understanding was, of presenting the idea that the feasibility of realising AI is non-obtainable, no matter how much resources are poured into the research of it, because to teach a machine learn, invent, improvise, etc, would have to be a subset of our own understanding of self, and this, dear friend, has a severe, grave, philosophical handicap: understanding your own behaviour is a context entanglement of you and your changing environment, and any interpretation resulted is only valid for the moment, not as a rule to be taught to other beings, machines included.

For me, and for my obvious ignorance, it very hard to except that a mechanism , once shaped and programmed with a given range of tasks in advance, be suddenly able to "invent" new ways to perform in a changing environment, or as I called it: (re-write it's own handling procedures while running), and in that, I tend to agree with the above mentioned teacher.

I'm aware that the labeling something as "impossible" may even considered by some as inflammatory or ignorant, but I truly believe it is so, as the barrier (writing time vs running time) mentioned on the original post, is just one principle challenge no one yet met, but not the greater one: How to teach a machine something you're struggling to understand about yourself (and with no outside aid, allow me to add)... or even harder yet: how to teach a machine, the meaning of context-based interpretation, without the need to implement our whole life's experience into it, (and which by the way will never be fully relevant to this given machine - See Godel's rule)

The ability of programs to ignore Illegal input-range or input-type, demonstrates no viable way to negotiate with a precedence, as this would necessitate their knowledge-base to re-interpret what they already know, as we have to do on daily basis.

and for that matter comparing us to a machine Presenting us (...is not at all unknown for humans to "become idle in an infinite loop" when faced with unexpected "inputs"...) as machines for that matter, proves nothing to this debate, at it is already acceptable that humans or other organisms may be considered as machines in some aspects: we're also prone to malfunction in so many other ways, not to mention degradation of wear and tear... , besides, Doesn't physics an cemistry apply? Let it go, this is not the issue.

. . .

Now, I can understand (and sympathise with) that you represent in this discussion, the hopeful majority (or minority for that matter) which hoped that, one day, given enough resources, we won't have to think act, as the machines we created will do this for us. Fine. You may be one of those who think that if no major breakthrough it the field is achieved yet, it still leaves some hope for the future, and what I'm here to say (based mainly on Godel's rule to be frank) is that it's principally, and philosophically impossible, hence all of my trivial examples and techno mumble.

. . .

As once said about aliens, that they're here, but hide well, not to freak us out, it was then replied that such tautology won't fly: if the can't show or leave a message, they're simply not here.

I maintain the just about same, about any challenge today's AI is facing: in the matter of Context Recognition and Interpretation, even in the smallest simplest demo range, for example: meaningful poetry or prose, 3D-Object Recognition, spoken sentence interpretation, etc, etc, and please, don't give me "even humans cannot write meaningful poetry of prose" or "I don't understand London Cockneys", This has nothing to do what I'm referring to.

Let's be fair in this debate, and refer to the issues, not to each other.

Hi Yuval and Greg,

I want to jump in here too, I think. While I agree that it's unlikely and potentially impossible to get an AI that can be as flexible and adaptable as a human being to dramatic and short-term changes in environment and expectations, we already do have programs that "learn" and adapt themselves in some ways to changes in their environment. I typed one into my high school's time-sharing system in BASIC in 1974, working from a listing of a game called "Zoo", which played 20 questions about animals. Self-modifying code is no longer uncommon in various controls applications. If you have a car newer than about 1995 with an automatic transmission, you've experienced a car which learns to shift "better" based on experience. Programs which ignore (or go to an error-handling routine) unexpected or "illegal" input are also quite common. (I actually wish for more error-checking rather than less in many applications. Though being possessed of a name which violates many English-language naming conventions, I would also wish that they would be more careful about defining error states...)

I am an ex-knowledge engineer and also have a long-time hobbyist interest in context-recognition and spatial interpretation for computers. I dream of having a program which can recognize and interpret the limited-and-formal visual "grammar" of mechanical design drawings. Something that could, for example, generate a parts-list or even a parts-count from a cross-section or exploded view of an assembly of say, 15 parts. Right now, it simply can't be done except by addressing one sub-domain of mechanical design at a time, and even then it's difficult, expensive and time consuming.

However, that this task can can be done for a few sub-domains implies that it ought, with one or at most two additional levels of abstraction, to be possible to do it in general. There are programs out there which can "recognize" gears in a gear train, and take actions based on that knowledge. Or "recognize" types of electronic components, consider the gap between them in a design, and warn the designer that they are too close together to provide adequate heat dissipation. And I have heard of (but not seen demos of) "spy" systems which look at satellite photos to identify certain sorts of vehicles or structures.

There are plenty of programs out there which use techniques such as rule induction, fuzzy logic, and genetic or adaptive algorithms developed by the branches of computer science usually lumped in under AI. These programs often exhibit both "intelligent behavior" in a single domain, and "learning", again within a well-defined domain. So it must be possible. At the same time, the programs now in widespread use are limited to a specific, usually fairly narrow, domain.

The Turing test as originally conceived specifies "general conversation" between a human and a program, and many existing programs can fool shallow or narrowly-educated testers. Several AIs or Artificial Personalities have been set up to interact with humans in certain ways, and the humans easily and ordinarily come to think of their on-line "health coach" or "therapist" as a real person. This might well be considered as having "passed" the Turing Test.

What do you guys think?

Ann Marie (Anna) O'Connell

Ann Marie,

"What do you guys think?"

I agree with your general definition of AI and what you say.

The biggest problem "understanding-wise" is that the OP started out with his own narrow and rigid definition of AI, without sharing that precise definition with us before he launched into all the very specific reasons he thought why it couldn't be attained.

Regards, Greg

Was delighted to read your comment, and I must admit to have loved Greg's in spite of his manner (mainly because of his enthusiasm and somewhat "burning urge") and evidently my poor English expression (I'm somewhat an "impulsive" writer, rushing the message through, many times without even checking my grammar) - but nevertheless, there is a main point I believe I have, which so far was not met - that is the difference between calculation and thought. Yes, there are Auto-Routers in PCB Design since the eighties (Calay on Digital's PDP -1986 and on); there are 2D pattern and contour recognisers; phoneme recognisers in speech and command recognition, Video (not just IR) moving body isolators; chess counterpart programs and all kinds of feedback regulators, it's all common knowledge, but these calculate but don't think. They can recognise, find and compare, decide and not choose if you may, not invent, inspire and create.

It really boils down to our definition of intelligence. We all know Expert-Systems appear to behave in an intelligent manner, and for those who think Turing's test is valid for this definition, just consider the "Chinese Box Paradox" and calm down. Admittedly, to some, the appearance is good enough. Fine.

How confident will they be if an-expert system such as an Auto-Pilot (today's can preform the whole shabaam including landing) will be applied without human supervision (aground or in the cockpit) ? How confident will they be when ES is applied nuclear launch control without human supervision?

I believe the research should first concentrate of theoretical feasibility, with any new type of gadget in the matter considered in this aspect.

I don't consider myself a Technofob, On the contrary, I love invention and novelty, but I believe the top-post presented the core of my arguments and the sum of it all lies not in a technical debate, but instead tries to describe a deep, bothering Principe, which cast a real doubt that intelligence is possible in man-fashioned mechanism. Can a language be intelligent? No. a Language can express or demonstrate it's creator's intelligence. The same about any invention which is a subset if it's creator's sum of knowledge.

Consider what's called machine's states: a toggle is a two-state machine. a chess-board is a 10 to the power of 104 (?) machine (or more, who cares) very impressive, but given and limited. Is one more intelligent than the other?, then go further and consider a door-knob or a wheel, both having unlimited number of machine-states. What of these? See my point?

...Chinese Box Paradox...

http://www.catesfamily.org.uk/letter-to-peter/chapter5.html

as a nice reference, one of many

...Chinese Box Paradox...

The rough description is: A sealed booth with a narrow slot, through which you insert your question in Chinese (Not really relevant to the point, but it is to say, it's about a given agreeable communication method). After a while, you get your answer, again in Chinese. From that, you cannot determine weather inside the box sits a wiseman, a trained ape working with tables, columns, and typing levers, or an automate, as long as the answer you got met your question adequately.

Is appearance the goal or is it substance?

...Chinese Box Paradox...

More to the (John Searle's Chinese room argument) point:

http://members.aol.com/lshauser2/chinabox.html

...There are plenty of programs out there which use techniques such as rule induction...

Just a weak challenge, as an example to illustrate the said failed promise: natural language translation. We have OCR, on-line vocabularies, and a combination of the two, in the form of this Whatchamacallit scanning-pen, which can translate printed words, one at a time.

Beautiful. Can you take a book and translate it, say, from Hebrew to English? No. If we could only say why, I would be satisfied, because the prevention would already be shattered.

I believe we can sense it all lies in this vague realm of "contextual interpretation", which in turn, is dependant on having a dynamic world-view we may call "knowledge-base" (not to be confused with "trivia-base") and then the challenge only begins. Now Go figure..

yuvalmate,

I am in no way trying to make this a personal issue.

Simply put, I disagree with what seems to be your definition of AI, and I understand it less with each post of yours.

Cybernetics is part of the key here, but the main thing is your definition of AI: I honestly don't understand it.

I was not going out of my way to compare humans to machines, you are the one who is doing that in reverse with your constant references to human learning and capabilities. (Again the issue of cybernetics). You are applying a much more rigid interpretation to individual machines than you are to individual humans.

Without prejudice as to any belief or disbelief in aliens, to address the thrust of that thought, I will only say that again I disagree: Absence of evidence can not to be interpreted as evidence of absence.

"... in the matter of Context Recognition and Interpretation, even in the smallest simplest demo range, for example: meaningful poetry or prose, 3D-Object Recognition, spoken sentence interpretation, etc, etc, and please, don't give me "even humans cannot write meaningful poetry of prose" or "I don't understand London Cockneys", This has nothing to do what I'm referring to."

Again you confuse me with what you could mean by AI: On one hand you want to seem to define it terms of the best of overall human capabilities yet object to comparisons of the range of individual human traits. It has been demonstrated time and time again that despite our pretensions, we are not consistently logical thinkers overall in our everyday lives.

I don't have any wishful thinking on the subject of AI, but our definitions seem to differ.

On one hand you are defining AI as possessed by a machine in human terms and then say that a comparison to real humans has no place here.

If you are saying that a deterministic, logical program will never be able to think, learn, and "create" the way humans do and I am inclined to agree. But, that is an overly narrow definition of AI.

Some years back, in the early days of introducing robotics into industry, many misapplications occurred from the misconception that robots could simply replace people in many tasks. They examined tasks that people did and designed robots to do them. This effort was by and large an utter failure. Later, when they designed tasks around what robots were best at, and/or limited the use of robots to things they were best at, their potential was finally realized. I mention this because your definition of AI seems to mirror the idea that such a machine could "replace" people in any situation. My definition is much broader, with the bar set much lower. That is that successful applications of AI would be limited in scope, based on the specific capabilities of that particular implementation, matched to the requirements of a given "task" or framework of related "tasks". I would no more expect an AI machine to be able to deal with all exigencies than I would a city dweller understand the nuances of farming.

I know you object to comparisons with people, yet your first paragraph (and your basic argument) is based on that very comparison.

Regards, Greg

...your definition of AI: I honestly don't understand it...

...to the effect of: 'Intelligence is self-taught information system entity, which has an operating model which serves as a "world-picture" (a symbolic representation of the rules governing it's negotiable reality) and within the world picture the self is an integral part of the symbolic system (something to the effect of a self-awareness embedded within, maybe as a self preservation, or as a reference-post, mechanism...

I hoped We wouldn't have to go into the definition of "Information-System" but it is something to the effect of: A calculating-viable entity, with a memory storage and retrieval facilities, adequate for self preservation, interaction with other such entities.

In "calculation-viable entity" I refer to the ability of sorting and isolating common attributes in a random group of objects, as well as comparing between non-common set of attributes in such a group. As counting in a given order or a given numerical system is not on this list, because it's a subject to a higher hierarchy of "math" which is one of many possible such orders, the "calculation-viable entity" is a given way to describe the ability to recognise patterns and attributes from a chaotic assortment of objects, and prioritise or define them in any "useful" order. something like that (I'm more familiar to these definitions in Hebrew). Again - Please excuse my English, I do my best now to be as concise as I can.

...Absence of evidence can not to be interpreted as evidence of absence...

I totally agree.

In the example given, I was referring to "Occam's Razor" determination principle, that is to say "If any small demonstration of a thinking machine appeared, ever since the seventies, it would suffice. The lack of it would suggest something is wrong with the basic premise of the research, and evidently of the development. Having that said, the abundance of other inventions since, wonderful as these are, contributed exactly what, for the ability of a machine to think?

...

I know you object to comparisons with people, yet your first paragraph (and your basic argument) is based on that very comparison...

Let me re-iterate the core of my argument: I suggest that the ability to think is not exclusively human, but it is void in machines created by humans, because it is a subset of their sum of knowledge, not abilities.

Comparing the abilities of humans and machines (created by humans) serves no point of relevance to me, because in that sense, a human can also be defined as a machine, prone to this mortal coil, who cares? What I'm interested in, lies in the understanding of thought, inspired thought, not just a mere calculation.

It lies in the understanding of gist, comprehension, invention. Not in the rules of symbolic communication, but in it's conception within an inquisitive mind. In the fashion learning and reasoning occur.

There is, admittedly a fine line between a tool's ability and it's user's will (that is also to include an identity commanding the body, if it makes any sense), and I have a strong feeling that (it has to do with) some form of self-awareness, that is the hardest trait to sprout in a machine, especially by it's creator/designer, because of Godel's rule.

I can accept that you can equip a machine with a self-preservation mechanism, but it is harder for me to except that you can give it a willing identity such as an ego, for instance. Yuval

Hi Yuval,

Now we have it, You are clearly defining AI to meet the highest bars: inspiration, and even more importantly, self awareness.

The only creatures that we presently know of who exhibit actual self awareness besides us, are certain of our close relatives among the great apes, dolphins, and possibly some of the whales and elephants, but the jury is still out on the whales and elephants I believe.

There is an old saying to the effect of "Be careful what you wish for, lest you get it", and creating a truly self aware machine may be one of those wishes that I for one am not about to make.

I assume though it will happen at some point, but may, as is the case with our close animal relatives be largely "written off" or not universally accepted for all its consequences because to do so, poses moral dilemmas we would prefer to avoid. It may, as goes the theme of a number of science fiction stories, take place incrementally in outward manifestations so that when it is actually achieved, its a "surprise". However, I do believe it will happen because we have barely scratched the surface now, and have only really been at it a few decades or so. It may first arise in a man-made biological entity, some type of neural network, or some massively linked parallel computer, but I don't see it happening with Von Neumann style computers and software, so at least in that we agree.

The trend seems be toward some coalescing of the inorganic and human engineered biological processes as a means to a number of applications, and AI may well be one of them.

At some point, I think the writing is on the wall that we will take charge of our evolution. How that might happen is relatively easy to imagine now, but what the implications are is not. Another related topic is the so called "cyborg" and we are well on the way there also, with our prostheses, false teeth, pacemakers, etc. and I have no idea where that will lead over the next century, except its probably safe to say it will only increase at least until we master biological engineering further.

We are heading into the great unknown, at an ever increasing speed.

Regards, Greg

...I assume though it will happen at some point...

There is, you know, this on-going supposition (I first heard of it in the late eighties, presented as a whole school of AI thought) that "given enough power" that is to say above a given threshold of memory volume and processing-speed, such self-awareness is inevitable to appear by istelf... Again, intuitively, it reminded me of the (above mentioned) Munchhausen trick

...some massively linked parallel computer...

I once read that the main hinder in parallel processing is too, contextual, and Ada Semaphores can only orchestrate the flow of entangled data, as the contextual issue is left to building Context-Daemons as an aid to relieve the programmer from the Hugh task of scanning the myriad of machine-states arising from the program. Of course, building the daemon is entirely up to the programmer's imagination, and so on, not unlike a Russian doll...

...so at least in that we agree...

We apparently agree on much more, alas, my own, personal, deep fear, is that this turning-point of "finally taking care of business" is nearer than most people care to see: We might not have the centuries it would take to witness the birth of true AI or autonomous cyborgs (as opposed to the existing bionic devices, to this or that extent).

With a global deteriorating security sate and political cooperation, we might not be able to pass through this necessary narrow-window of opportunities, or as my father said: you have to maintain a working body, in order to attain any state of happiness.

Or, as my friend Giora said once: "Maybe the bottom line is, that man is not the wisest creature in nature..."

You`re right, Greg, the first step is to find a common accepted definition for what AI is/can be/will be. I`ve been working with Fuzzy Logic for several years for which yuvalmate gave a definition. But in my opinion AI can or will (never ??) be more than a subset of programmers natural intelligence and therefor only a subset of this which can be easily proven by making a program error.

Another question is where intelligence (whatever that is ...) starts - is a 200 year old windmill equipped with a centrifugal governor "intelligent" because the inventor was so intelligent to build one ? Or does AI start in the 20th century, being constricted to software based technology applications ?

Interesting discussion - Regards Uwe

...be more than a subset of programmers natural intelligence...

This is so much to the point. It was indeed quoted off both Alan Turing and Conrad Tsoeze (The German inventor of "Plancalcul" the first all-purpose stand-alone programming language in the thirties) that: When a Mechanism is a subset of it's inventor's sum of knowledge (as it always is, by definition), it can never really contribute or advance to any significant attribute, other than a mere extension of the existing" or something like that.

In that sense alone, much of the people's phobia from the robots taking-over (as portraid in "I Robot" with will Smith) is fictional, not to mention the possible advance of intelligent machines. This is not to say that programming self-awareness of self-preservation procedures is impossible, as any positive or negative feedback automaton proves, since the early 19th century.

...Conrad Tsoeze (The German inventor of "Plancalcul" the first all-purpose stand-alone programming language in the thirties) ...

Should be: "Konard Zuse", "Plankalkul", and "forties"

See: http://www.epemag.com/zuse/part5.htm

and: http://www.rheingold.com/texts/tft/3.html

You're right, I nearly forgot: "it would be wise to agree on what we mean by intelligence in a given context"; It was once defined, maybe by Claude Shannon or Von Neumann, or any other father of what was later called "The Information Theory" (One of the followers of Godel, I can't recall now): it is something to the effect of: 'Intelligence is self-taught information system entity, which has an operating model which serves as a "world-picture" (a symbolic representation of the rules governing it's negotiable reality) and within the world picture the self is an integral part of the symbolic system (something to the effect of a self-awareness embedded within, maybe as a self preservation, or as a reference-post, mechanism.

This of course, is a schematic, formal, rational definition, it doesn't take into account phenomena such as a soul, urges, impulses, etc., but it is an attempt, done in the fifties, to define a psyche of an intelligent, aware, entity.

I correlates many philosophical and logical issues, such as the "Chinese-Box" paradox which is based on the Turing Test, but nevertheless may contribute to the matter at hand. Yuval

yuvalmate,

Thanks for your continued posts to clarify what you meant.

I was referencing YOUR definition, since that is what counts here, not the definition held or once expressed by others.

Regards, Greg

I simply adhere to the definition quoted, as it was the only one I met so far, to satisfy my confusion and doubts, in this matter.

Specifically, I have no original definition of my own, just a vague feeling of self and being, which may sometimes correlate to this or that definition. Yuval.

Hi Yuval,

"Specifically, I have no original definition of my own, just a vague feeling of self and being, which may sometimes correlate to this or that definition."

That is precisely why it is so difficult to respond directly to your original post.

For some mainstream definitions of AI, which are much broader than yours:

http://en.wikipedia.org/wiki/Artificial_intelligence

Once again I mention cybernetics, as a part of it deals with these very issues you have raised, in the comparison of machine "thinking" to human thought (I read Norbert Wiener's book about 40 years ago) yet you haven't mentioned cybernetics at all. Norbert Wiener founded the field of cybernetics and this was the underpinning for much of AI.

http://en.wikipedia.org/wiki/Norbert_Wiener

http://en.wikipedia.org/wiki/Cybernetics

BTW Yuval, please do not take my strident manner personally. I just try to get right to the point in as clear a way as I can.

Regards, Greg

Thanx, Yuval

Yuval,

You're welcome!

I am well aware I err on the blunt side, yet I don't have the patience of some here to spend a paragraph or two dancing around an issue trying not to offend at any cost, and finally finishing up with only a murky, oblique point buried in niceties.

And as you said in one of your posts (#16), I am enthusiastic (and passionate about many subjects).

For me this is a technical forum not a "feel good" group therapy session.

Engaging and drawing out details from others is one of the ways I learn, and I do love to learn.

Regards, Greg

That's perfectly fine with me. As a matter of fact, I prefer to argue passionately with someone who disagrees, than some feeble exchange of niceties with someone who agrees. My main drive in life is to learn and comprehend, not to convince.

Once we have to leave this world, we have nothing to "hold our hand" so-to-speak, but our most beloved. for some - It is the knowledge of life itself, the accumulated experience of living.

Ditto

P.S,

To the point of this discussion, I believe we both know it's a principle division ever since the end of WW2 and the birth of ENIAC, and reflected well in literature and debate. As you hinted, it is impossible to determine in advance if something is possible or not, while I maintained there is a deep principle philosophical doubt, if not a prevention to this prospects.

What would your assessment be as to the prospect of further development in the near future? Where would your bet be (in reference to R&D fields)?

Artificial intelligence is nothing more than a hobby of those having not yet fully exploited their natural intelligence - which concerns to nearly everyone of us, me included .... ;)

Indeed, one of the goals for the realisation of AI, was declared in principle as "Teaching a machine to think, in a similar way as we do" but to frankly, we don't even know how we do that, (in a formal consistent way) in order to be able program it into a machine. Besides, the formality of thinking is a philosophical matter

AI will arrive immediately after the "paper-less toilet" ....

I would design an AI to deal with any input so, there would be no such thing as an invalid input. People make logical errors all the time but, the fundamental principles of logic that we apply to everything never change. Likewise, the fundamental logic that an AI would apply to everything would never need to be changed. Only the assumptions (mathematical relations) that an AI creates as it attempts to make sense of the world would need to be correctable. These formula would constitute an extended logic system. AI would not have to behave like people in order to be intelligent either. An AI suffuciently experienced with human interaction would be able to pass a Turring test but, a Turring test would not be a reliable measure of the intelligence of an AI. Especially if the AI is smarter than the person conducting the test.

"...AI would not have to behave like people in order to be intelligent either..."

I would have to agree

"...Especially if the AI is smarter than the person conducting the test..."

Would you explain your difference between "smart" and "wise" ?

("to say Deep Blue is not thinking because it's a machine, is like saying a jet-plane is not flying because it doesn't flap it's wings")

"Would you explain your difference between "smart" and "wise" ?"

I don't really see a difference between the notions of "smart" and "wise". If someone is deemed "smart" then they may also be deemed "wise"; if someone is deemed "wise" then they can also be deemed "smart". The opposite of "smart" is "stupid" which is synonymous with the opposite of "wise", "unwise".

We may sometimes hear someone say something to the effect of "A smart gadget", or "This is a Clever machine", but hardly to the effect of "A wise machine", and I think that intuitively, people will sense some difference.

As said elsewhere on this discussion, an apparatus may demonstrate the wisdom of it's designer, but, can it be wise itself? To some, there is no difference, and this is at the core of the discussion about Artificial Intelligence: Can a machine really think, or is it just calculating.

Indeed, as was many times hinted above, we may consider calculating as thinking par-excellence ("to say Deep Blue is not thinking..."), and I would have to agree, but, is there some other quality or trait in the thinking process, which may attribute "intelligence" per-se, such as the ability to invent for instance, and therefore so defined, and as such, mentioned as a "Hard" Requirement for machine-intelligence, and opposed to a "Soft" Requirement for machine-intelligence, such as the Deep-Blue quote implies.

So, it appears to be "wise" to define in advance, what is your definition of 'intelligence', and thereby of 'machine-intelligence', as this is something I have learnt from this discussion.

Once you define that, it's going to be a much easier discussion, as we can all see in above, from the Original Post and forward.

There is an on-going, 60-years-old debate about this issue, which only the future will resolve, but maybe such a discussion, anywhere, can clarify something about the definition of thinking, intelligence, and wisdom, for either humans, dolphins, or machines.

"We may sometimes hear someone say something to the effect of "A smart gadget", or "This is a Clever machine", but hardly to the effect of "A wise machine", and I think that intuitively, people will sense some difference."

I've heard people say things like "that's a wise design feature" or "wise implementation" but only from engineering types. My sense is that such comments by anyone don't refer to the gadget or machine but to the person who designed it.

"As said elsewhere on this discussion, an apparatus may demonstrate the wisdom of it's designer, but, can it be wise itself?"

Wisdom is an emergent property of intelligence rather than a design feature. While the efficiency and performance of the design features of an intelligent apparatus is limited by the wisdom of it's designer(s), the wisdom that emerges from that intelligent apparatus is not.

"Can a machine really think, or is it just calculating."

Well one thing's for sure, a thinking machine can't really think without calculating. I remember reading an article on brain research many years ago. There are a bunch of nerve fibers directly between the left and right hemispheres of the brain. Women have more fibers then men do which may explain why women are more articulate than men. Some researchers studied patients whose left and right hemispheres were either partially or completely separated. It seems that these people have a crippled sense of awareness in general as well as inhibited self awareness suggesting that self awareness emerges as different parts of the brain, which are not self aware, 'observe' each other back and forth. This leads me to think that the characteristics of consciousness and self awareness in people aren't necessarily characteristics of intelligence but of perception and aren't required characteristics of intelligent machines.

"So, it appears to be "wise" to define in advance, what is your definition of 'intelligence', and thereby of 'machine-intelligence', as this is something I have learnt from this discussion. Once you define that, it's going to be a much easier discussion, as we can all see in above, from the Original Post and forward."

I define intelligence as the ability to identify and resolve any kind of problem.

When Ada Lovelace was asked to comment on Charles Babbage's Analytical Machine, she said: The machine cannot invent anything, it can only do as it was instructed.

I use this historical example to illustrate that this fundamental question is present and debated, ever since the earliest attempts to build computers able to do more than mere calculating.

Before Babbage's machine design (It was never built because it relied on mechanical gears which couldn't be fashioned accurately enough in his time), there were quite complicated calculators, far more developed than Pascal and Leibnitz Number-Crunchers. Babbage was attempting to design a "general-Purpose" Equation-Solver, programmable for each task apart.

This might reveal some of the problem's nature: If a Binary Machine (Just for instance, because Analogue Machines are about as prone to this) is programmed in advance, it's run will result with a given number of machine-states, be that hundreds, billions or more. Given and therefore limited.

Now, imagine it has to deal with a situation it was not yet programmed to handle, be that as trivial as it may. This Will force it to increase it's given number of machine-states. It can only increase it, if you re-program it, to include this new situation. But then, again, it is done in advance, prior to running.

Once running, it returns to act as an automat, that is, obeying the old and new rules combined, but unable to invent,(in this aspect, at least) new ones, because for this, it needs to stop running the current program, return control into a programming mode, etc, etc.

So, If you say that calculating is thinking all the same, fine.

You may even say that calculating is a form of thinking, because us humans calculate from time to time and we do so in our thought, that's fine too.

But if you can show me a design which can invent something on purpose, (Not by a random or lucky-occurrence, just as you can sometimes see faces and forms appear in the clouds), THEN I'm impressed.

Some very interesting programs were written since the eighties: poetry writing, pattern designing, music composing, but all these were mutant relatives of Eliza in the sense that they were designed for appearance, not for substance.

It's fair enough to say: who cares? I like the outcome, it's good enough as-is. Just like this art expert who saw the paintings of an elephant and only later told it made them.

But if you are interested in the essence of the subject, This dilemma is not as easy as it appears, and you might find out it's related to the chances to mankind to survive in the near or far future, no less.

"This might reveal some of the problem's nature: If a Binary Machine ... is programmed in advance, it's run will result with a given number of machine-states, be that hundreds, billions or more. Given and therefore limited.

Now, imagine it has to deal with a situation it was not yet programmed to handle, be that as trivial as it may. This Will force it to increase it's given number of machine-states. It can only increase it, if you re-program it, to include this new situation. But then, again, it is done in advance, prior to running."

I can see why you would need to reprogram a calculator but, why would you need to reprogram an AI? An AI is supposed to be able to do what the programmer does; create problems and specifiy the rules for solving them. The AI wouldn't need to be re-programmed unless there is a flaw in it's logic.

"So, if you say that calculating is thinking all the same, fine."

Of course, calculating isn't thinking but, thinking is calculating.

"But if you can show me a design which can invent something on purpose, (Not by a random or lucky-occurrence, just as you can sometimes see faces and forms appear in the clouds), THEN I'm impressed."

First of all, randomness is not a product of nature but of our own limited perception of cause and effect. Invention is a process of problem solving and problem solving is a process of trial and error- cause and effect. Thus the process of invention can appear as random as any phenomena whose process cannot be directly observed.

...AI is supposed to be able to do what the programmer does; create problems and specify the rules for solving them. The AI wouldn't need to be re-programmed unless there is a flaw in it's logic...

- I would love to hear how, - even on a very small-scale demonstration -

...The AI wouldn't need to be re-programmed unless there is a flaw in it's logic...

Indeed, this might be the holy grail for the avid seekers of AI, but, I maintain, how? You may think that functional-electronics can determine it's behaviour thus getting rid of "software" as a program, or (for that matter), any other kind of Guidelines for conduct, you may even think it can be done by mechanics,

Then, please keep in mind that "Firmware" can be considered a program par-excellence (remember Jacquard's looms?), and Brass-Minds were created ever since Pascal's time, if not earlier, and their mechanisms functionally act as programs, only not re-programmed so easily.

Because to re-direct their operation, you'd have to, (remember the computer example in post-40) stop it, return control to the technician, re-assemble it to include the new function, etc, etc...

...First of all, randomness is not a product of nature but of our own limited perception of cause and effect...

It was many times demonstrated that the determination of a pattern in a "random" arrangement, is really subject to OUR association: You may see a car in the clouds or on grains, and we know cars are products not organisms, - so in that sense I agree, but I mentioned our ability to determine a pattern, not nature's ability to create an apparent one

Ok Yuval lets get down to basics. A computer is a binary calculating machine dressed up in software so as to seem to be "Intelligent" The very idea of machine intelligence is that of an oxy-moron. Two conflicting parts. To be intelligent you have to have independence of thought. No machine can think. A computer is only as intelligent as its programming. After all look at the many times a computer has when required to give an answer failed to give the correct one. The machine can only refer to its memory in order to find its next instruction witch has to be carried out in the strict sequence listed in the programme. Garbage in Garbage out. The glossy addons only make the machine look to be intelligent by deception. After all it all comes down to binary arithmetic in the end.

Almost. Since it's not a decided issue, what I can say is that you expressed sharing my definition of intelligence. Greg's immediate, instinctive reply in #2 sums the whole discussions in a nutshell: what's you definition of intelligence?

Mine as expressed here, is considered the "hardest" definition.

The "softest" definition, Does has distinction between appearance and substance. something to the effect of "If it seem intelligent to an intelligent entity, then it is intelligence".

This is not a weak argument at all, philosophically, and has to be taken into account, once and again.

...Does has distinction between appearance and substance...

Should be: "Does not make a distinction between appearance and substance..."

CR4 just ran a four-part interview with AI researcher Selmer Bringsjord. If you're interested in reading what he had to say, here are the links.

Engineering Zombies (Part 1)

Engineering Synthetic Minds (Part 2)

Dumb Humans and Ruthless Machines (Part 3)

AI in the Garden of Good and Evil (Part 4)

Thanks Moose. AI has been an on-going interest for the last twenty years for me.

To return a favour, Here's Pamela McCorduck's site. She wrote the celebrated Non-biased account on the subject called "Machines Who Think"

You're welcome, Yuval. And thanks for the link to Pamela McCorduck's site, too.