Jump to content

  • Log In with Google      Sign In   
  • Create Account

What is really AI?


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
81 replies to this topic

#41 vallentin   Members   -  Reputation: 181

Like
0Likes
Like

Posted 03 April 2008 - 03:42 AM

Quote:
Original post by Timkin
Quote:
Original post by Rixter
Quote:
Original post by Timkin
Quote:
Original post by Rixter
AI is search.


Ignorance is bliss


I figure while we're assigning arbitrary definitions to an apparently ill defined concept, why not take the simplest?


Except that saying that "AI is search" is like saying that a house is a hammer. Search is a tool that can be used to create the end result (along with other tools, skill and creativity), but that doesn't magically transform it into the final product.

...and having said that we must accept that a 'so-called AI' that uses only search to solve a problem is not 'AI', but rather just an intelligently designed implementation of a solution to a computational problem.

This is the most common objection raised about AI: that it's just intelligent design, rather than an embodiment of intelligence... but then, are we any more (and here I state that I believe in 'design by evolution' rather than 'design by God', just to make my position unequivocally clear). So where do we draw the line?


Can't they both be true?How come we exist?Due to what?Something or someone is accountable for that Let's call Him God.


Sponsor:

#42 AngleWyrm   Members   -  Reputation: 554

Like
0Likes
Like

Posted 03 April 2008 - 05:55 AM

There's also a couple women competing for that trophy: Mother Nature and Lady Luck.

#43 Emergent   Members   -  Reputation: 971

Like
0Likes
Like

Posted 03 April 2008 - 06:51 AM

I submit that "Artificial Intelligence" is just the name for an academic field. It is a social label first and foremost, used by researchers to classify themselves and others.

Like those from all academic fields, its practitioners are dependent on grant money to survive, and judged by the number of papers they write that are accepted for publication by major journals. Those are the inputs and outputs of the academic game.

Success in the game comes partly from having good ideas, and it comes partly from having the technical expertise to execute them well. But it also depends strongly on marketing, self-promotion; you have to convince others -- article reviewers at journals, funding agencies, and sometimes even the general public -- that your ideas are better, cooler, more interesting.

And what better way to do that then by calling them "Artificial Intelligence?"

An example: What really is machine learning? There is some system which, given inputs, produces outputs; we want to select inputs which minimize some cost function defined over the outputs. That is all. Another group of academics does pretty much the same thing (using more abstract language), and they call what they do "Optimal Control."

Question: Who is more likely to get the newspaper article written about them with the sexy headline, "Smart Computers 'Learn' like Babies:"
1 - the researcher who calls what they do "Machine Learning" and talks about the computer being "intelligent,"
OR
2 - the researcher who calls what they do "Optimal Control" and talks about minimizing objective functions?

Answer: #1.

This is just one example of a general trend. You succeed in academia by making what you do sound sexy -- and calling what you do "Artificial Intelligence" achieves that.

If it sounds like I'm bashing Artificial Intelligence researchers, I'm not: It's the game that all academics play. And if it sounds like I'm more generally bashing academics, again, I'm not: Much of industry, politics, and essentially every other human endeavor is built on a strong foundation of bullshit.

I'm just pointing it out in this case, because I'd hate for everybody to waste their time philosophizing about what is essentially a brand name.

#44 Rixter   Members   -  Reputation: 785

Like
0Likes
Like

Posted 03 April 2008 - 07:49 AM

Quote:
Original post by Timkin
Quote:
Original post by Rixter
Quote:
Original post by Timkin
Quote:
Original post by Rixter
AI is search.


Ignorance is bliss


I figure while we're assigning arbitrary definitions to an apparently ill defined concept, why not take the simplest?


Except that saying that "AI is search" is like saying that a house is a hammer. Search is a tool that can be used to create the end result (along with other tools, skill and creativity), but that doesn't magically transform it into the final product.


Except isn't that like saying an encyclopedia is intelligence? (This is like a game of telephone, isn't it? :) ). I would be more inclined to say it's the producing of the solution that is intelligence, not the solution itself. I'm not exactly going to defend this "AI is search" idea to the death (it's not even originally my idea), so maybe there are other tools that can't be classified as search, but I would consider skill and creativity to be more attributes of intelligence.


Quote:


...and having said that we must accept that a 'so-called AI' that uses only search to solve a problem is not 'AI', but rather just an intelligently designed implementation of a solution to a computational problem.


Perhaps our definition of search is different as well. As you said, one does not need to know how to do something for AI techniques to work, and sometimes not even the end goal. Wouldn't that require some form of search? Wouldn't even the wiring and re-wiring of our brain be a form of search? I don't know.


Also, I don't necessarily believe that AI is a misnomer, rather I think people tend to have these Johnny 5 expectations of what it is or should be.



#45 owl   Banned   -  Reputation: 364

Like
0Likes
Like

Posted 03 April 2008 - 07:58 AM

What if I say that intelligence is the "ability of organizing information" ?

#46 Sneftel   Senior Moderators   -  Reputation: 1781

Like
0Likes
Like

Posted 03 April 2008 - 08:13 AM

Bubble sort organizes information.

#47 owl   Banned   -  Reputation: 364

Like
0Likes
Like

Posted 03 April 2008 - 08:28 AM

bubble sort isn't an autonomous creature...

#48 Sneftel   Senior Moderators   -  Reputation: 1781

Like
0Likes
Like

Posted 03 April 2008 - 08:32 AM

Ahh, the definition groweth! Currently "Is an autonomous creature AND organizes information".

Consider a program which, given a poem as input, determines how interesting it is, and then composes a witty retort to the poem using the same metric and rhyming structure. Since it isn't an autonomous creature, is it not AI?

#49 owl   Banned   -  Reputation: 364

Like
0Likes
Like

Posted 03 April 2008 - 09:01 AM

I said intelligence, not AI.

For AI we dont need to imagine much, there are plenty of game AIs that behave pretty intelligently. And we could say that they *are* intelligent to some degree.

But if there was a scale to measure inteligence on how qualitatively and cuantitatively a being (artificial or not) organizes data, then bubble sort, your program and all the games AI's would be near ZERO and humans would be at the top.

#50 Sneftel   Senior Moderators   -  Reputation: 1781

Like
0Likes
Like

Posted 03 April 2008 - 09:04 AM

Alright. Is it not intelligent?

#51 owl   Banned   -  Reputation: 364

Like
0Likes
Like

Posted 03 April 2008 - 09:10 AM

are you trolling?

#52 Sneftel   Senior Moderators   -  Reputation: 1781

Like
0Likes
Like

Posted 03 April 2008 - 09:23 AM

Not in the slightest.

#53 owl   Banned   -  Reputation: 364

Like
0Likes
Like

Posted 03 April 2008 - 09:36 AM

Quote:
Original post by Sneftel
Alright. Is it not intelligent?


Quote:
Original post by owl
I said intelligence, not AI.

For AI we dont need to imagine much, there are plenty of game AIs that behave pretty intelligently. And we could say that they *are* intelligent to some degree.

But if there was a scale to measure inteligence on how qualitatively and cuantitatively a being (artificial or not) organizes data, then bubble sort, your program and all the games AI's would be near ZERO and humans would be at the top.


Yes, based on the definition of intelligence I proposed earlier, we could say they are intelligent in a very low scale and their domain is quite restricted compared against human beings.

#54 Sneftel   Senior Moderators   -  Reputation: 1781

Like
0Likes
Like

Posted 03 April 2008 - 09:39 AM

But now your definition gets hazy. You propose ranking intelligence based on "how quantitatively and qualitatively a being organizes data", without elaboration. Clearly the numbers in a sorted list are "more organized" than the words in a poem. Doesn't this make bubble sort more intelligent by your definition? If not, how does one score, on your one-dimensional scale, how a being organizes data?

#55 owl   Banned   -  Reputation: 364

Like
0Likes
Like

Posted 03 April 2008 - 10:10 AM

Yeah well, I think I never said what kind of "quality" to measure.

I meant quality in a human sense. A poem expresing a human idea (with methaphors, paraboles, reminicenses), would be of higher quality than a bunch of sorted numbers.

We could even say that a short poem contains more information than a bast collection of sorted numbers, since a poem carries a meaning (that triggers experiences, images, sensations) and the mechanisms needed to extract that meaning is orders of magnitud more complex than the one needed to appreciate the "sortedness" of the numbers.

#56 Sneftel   Senior Moderators   -  Reputation: 1781

Like
0Likes
Like

Posted 03 April 2008 - 12:54 PM

Quote:
Original post by owl
I meant quality in a human sense. A poem expresing a human idea (with methaphors, paraboles, reminicenses), would be of higher quality than a bunch of sorted numbers.
Then this just comes down to the bog-standard definition, "thinks like us".



#57 owl   Banned   -  Reputation: 364

Like
0Likes
Like

Posted 03 April 2008 - 01:04 PM

Well yes. Do you know any other remarcable intelligence to compare with? I mean, dolphins may be intelligent but their kind of intelligence doesn't really seem to be very useful for us right now.

#58 Sneftel   Senior Moderators   -  Reputation: 1781

Like
0Likes
Like

Posted 03 April 2008 - 01:54 PM

Not offhand. But you should be aware that that's the real definition you're using, none of this "organizes information" silliness. Justice Potter Stewart (in)famously defined pornography as "I know it when I see it", and perhaps he was onto something there. What you're looking for is a definition of intelligence that agrees with all the gut feelings you've already decided on regarding the level of intelligence of various entities. Since the underlying decisions here are, in general, so capricious and impulsive--yet now so strongly held--there's no reasonable definition one can come up with to excuse them all equally. This is why coming up with a definition of "intelligence" is useless unless you (a) have a need to objectively define a metric of intelligence, (b) have an objective test to determine whether a given metric is accurate, and © are willing to have that metric disagree with you and tell you that you're wrong about something being intelligent or not intelligent. Under any other set of circumstances, it's all just semantic flailing.

#59 Timkin   Members   -  Reputation: 864

Like
0Likes
Like

Posted 03 April 2008 - 02:06 PM

Quote:
Original post by Rixter
Except isn't that like saying an encyclopedia is intelligence?

No, an encyclopedia is a collection of information (pages in a book, or in the digital age usually an electronic database + interface). My comment about 'search being a tool' was intended to convey that while search may appear as a fundamental component of many instantiations of systems labelled 'artificial intelligence' it is, in and of itself, not a demonstration of intelligence. Just as the hammer, nail and 2x4 are components that can be combined to produce a finished house.

Quote:
Original post by Rixter
I would be more inclined to say it's the producing of the solution that is intelligence, not the solution itself.


Yes, you may well need intelligence to create the design and implementation of the end product given only the possible components... but again, that isn't strictly necessary. I could create a machine that builds houses for me. Or I could create a machine that builds machines that builds houses. At some point in that chain there is likely to be some application of intelligence... but that doesn't mean that I couldn't build a system that could interact with the world, learn and adapt itself so that it becomes a house. Now where is the 'intelligence' specific to building houses?

Quote:
I would consider skill and creativity to be more attributes of intelligence


I'd consider skill and creativity to be expressions of intelligence. Look back to my original definition for why I take this stance.


Quote:
Perhaps our definition of search is different as well.


To formalise my definition, search is a procedure for determining the elements of a set that satisfy an objective... so some examples of search:
  • finding the members of a set that map to an extremum of an objective function
  • finding a partition of a set such that at least one disjoint subset returns a disjoint value of a constraint function

    A search algorithm (IMO) is then a specific procedure for solving the search problem.

    How does your definition differ?

    Quote:
    As you said, one does not need to know how to do something for AI techniques to work, and sometimes not even the end goal. Wouldn't that require some form of search? Wouldn't even the wiring and re-wiring of our brain be a form of search? I don't know.

    It may not require an explicit instantiation of a search algorithm, but all learning and adaptation that improves the performance of a system is implicitly a search through the space of possible systems. So yes, learning in the mammalian brain can be considered a search through the set of possible mammalian brains (constrained by the initial state of the current brain and the reachable set of possible brains) for one that improves performance.


    As for Emergent's comments...

    Yes, in part I agree that 'Artificial Intelligence' is a brand label, just as 'Cola' is a label for a type of carbonated drink. There are many different drinks that might be labelled 'Cola' that don't necessarily fit with the general expectation of what 'Cola' is... and there may or may not be an objective form of 'Cola', nor of 'Artificial Intelligence'. But that doesn't mean we cannot use the label to describe specific instatiations of the product.

    The problem with AI is that too many people erroneously apply this label to what they're building so it has lost some of its value (mostly this is a problem seen in business where people are trying to sell products ... look at the hype over artificial neural network based products during the early 80s). Some of those instantiations may indeed by good examples of the product... others not so good... but some of us do believe that it's worth trying to work out if in fact there *is* an objective form of AI... and rather than just philosophise about it, we go out and build systems and evaluate them and ask the question... "is it intelligent"?... and then we get into arguments over our definition of intelligence... and most of us just go back to our labs and ignore it... but sometimes we feel compelled to speak! ;)


    As a quick aside... machine learning and Optimal Control should not be considered side by side and I don't believe anyone working in OC would ever claim what they were doing was AI. If though you meant by OC merely the problem of determining an optimal control function/regulator... then ML is just a tool for doing that... as are the formal methods of OC (what we usually call Control Theory).

  • #60 owl   Banned   -  Reputation: 364

    Like
    0Likes
    Like

    Posted 03 April 2008 - 02:39 PM

    I can imagine a sort of test to measure the quantity and the (human like) quality of data being "organized" by different creatures of nature. Say for example:

    * Is social?
    * Has language?
    * Uses tools?
    * Builds tools?
    * Is adaptable?
    * etc.

    And some sort of sub-set of detailed questions regarding the kind of society they form, how many different sounds (words) their language has, what kind of tools, for which purpose, etc. Each one with certain score, thus giving you a measure of how similar to human intelligence a given intelligence is.




    Old topic!
    Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



    PARTNERS