Kuro5hin.org: technology and culture, from the trenches
create account | help/FAQ | contact | links | search | IRC | site news
[ Everything | Diaries | Technology | Science | Culture | Politics | Media | News | Internet | Op-Ed | Fiction | Meta | MLP ]
We need your support: buy an ad | premium membership

[P]
New Scientist Magazine on AI Machine

By qslack in News
Thu Mar 01, 2001 at 05:35:45 PM EST
Tags: Software (all tags)
Software

New Scientist Magazine is reporting that Artificial Intelligence Enterprises, based in Tel Aviv, Israel, has created a computer program lightweight enough to run on a simple PC laptop that can respond to questions at the level of a 15-month-old baby. They expect it to advance quickly because it learns on its own; its knowledge does not need to be programmed in, unlike many other attempts at the much-coveted Turing test.


From the article:
He predicts that Hal could carry out commands issued without rigid syntax, and could also cope with confusing but similarly structured sentences such as "Time flies like an arrow" and "Fruit flies like a banana". It might even have a sense of humour.
The program, which is small enough to run on a laptop, has no sensory input - just a stream of words coming in from a keyboard. So far, it is only capable of simple sentences of a few words. For example, if asked what game it wants to play in the park it might respond: "Ball, mummy" (it speaks English, but could learn any language).

Sponsors

Voxel dot net
o Managed Hosting
o VoxCAST Content Delivery
o Raw Infrastructure

Login

Poll
I predict computers will surpass humans in natural language in...
o 2002 0%
o 2005 4%
o 2010 7%
o 2025 12%
o 2050 27%
o Never 47%

Votes: 65
Results | Other Polls

Related Links
o New Scientist Magazine
o reporting
o Turing test
o Also by qslack


Display: Sort:
New Scientist Magazine on AI Machine | 35 comments (21 topical, 14 editorial, 0 hidden)
my comments (4.40 / 5) (#12)
by Anonymous 242 on Thu Mar 01, 2001 at 11:22:28 AM EST

Blockquotes are from the New Scientist article.
They believe that if they can give it the linguistic abilities of a five-year-old, we will soon be able to converse with our computers naturally and possibly hang up our keyboards for good.
Personally I do not consider the linquistic abilities of a five-year-old to be adequate to converse with computers naturally, at least not enough to hang up my keyboard for good.
The software-based toddler was developed by Artificial Intelligence Enterprises (Ai) of Tel Aviv, Israel. It is said to have fooled independent experts into believing they were reading conversations between an adult and a real 15-month-old child.
I would personally hesitate to call a weak-AI program Arrificial Intelligence. Obviously some people (most notably Artificial Intelligence Enterprises) disagree. I do think that achievement is somewhat monumental. If the program can process language on the level of a fifteen-month-old child, it is a tremendous acheivement and notable in it's own right. But it also isn't really AI. Once a program is developed that has the full cognitive abilities of a fifteen-month-old child (including the ability to grow in intelligence), then I will say AI has been actually created. Until then all we really have is a program that can parse language at a barely adequate level. Whoop-de-doo! We already have programs that can do some tasks (such as playing chess) at a much higher level.

Growing intelligence (3.00 / 3) (#25)
by retinaburn on Thu Mar 01, 2001 at 12:18:57 PM EST

From the article I gleaned that it does 'learn'. It sounds like its told stories and asks questions based on those stories.

I think that we are a young species that often fucks with things we don't know how to unfuck. -- Tycho


[ Parent ]
so do neural nets (4.33 / 3) (#28)
by Anonymous 242 on Thu Mar 01, 2001 at 01:15:46 PM EST

IBM had a backgammon program for OS/2 that taught itself strategy once it was instilled with knowledge of the rules. There is much more to thinking than simply being able to acquire new rules.

[ Parent ]
oh no (4.75 / 8) (#13)
by Seumas on Thu Mar 01, 2001 at 11:25:33 AM EST

Great. Has anyone here actually dealt with a 15 month old baby? And we're going to be impressed by a supposed AI that replicates the questioning/answering capability of a 15 month old? Your basic baby of that age drools, screams, cries, soils itself, throws up and says "no" a lot.

With those standards, I'd say every version of Windows has accomplished this level of AI for years, showing true prior art.
--
I just read K5 for the articles.

Great! (3.00 / 4) (#15)
by maketo on Thu Mar 01, 2001 at 11:27:49 AM EST

It can say "gooo" to all questions posed!
agents, bugs, nanites....see the connection?
This is utter bullshit (4.45 / 11) (#17)
by MarkCC on Thu Mar 01, 2001 at 11:29:18 AM EST

Sorry, but this is absolutely, utter crap. It's another overblown attempt to generate PR so that they can sell utterly worthless crap to gullible customers.

My wife is a professional researcher who specializes in natural language processing. Because of her, I'm pretty familiar with the state of the art in AI when it comes to linguistics.

First of all: the "much coveted" turing test is only much coveted by people outside the field. People in computational linguistics have long since concluded that the ability to fool a human being is not at all meaningful.

Second: The current state of the art in natural language understanding is absolutely nowhere near what these guys are claiming.

Even the description here is contradictory: on the one hand, it claims to have understanding at the level of a very young baby. As a daddy of a very young baby, let me tell ya, their language abilities are very primitive at that age. Then, they go on to make claims about one of the classic difficult problems in handling language, which is semantic ambiguity (fruit flies/time flies).

If it can truly resolve the semantic ambiguity (which you should note, they carefully hedge by saying "he predicts that it can...") then its abilities are clearly far beyond that of the supposed 15month old.

So, at best, they're trying to make extravagant claims while also trying to state things in a way that give them a plausible excuse when it doesn't work.

This is just a crock. Don't fall for it.

Fruit flies like a banana. (3.00 / 4) (#22)
by i on Thu Mar 01, 2001 at 11:52:28 AM EST

I'm sorry, but I see no semantic ambiguity here. Given dictionary meaning of words, you just cannot interpret "fruit" as a noun. It must, unlike "time", come with an article ("a" or "the"). From here the reasoning is straightforward. You don't have to even know what "fruit", "flies", or "like" mean.



and we have a contradicton according to our assumptions and the factor theorem

[ Parent ]
uh.. (4.33 / 3) (#26)
by westgeof on Thu Mar 01, 2001 at 01:09:39 PM EST

Fruit is a noun, and doesn't need an article to make it so. (At least in US English) When I talk about fruit in general, I say something like "I like fruit", not "I like the fruit." And if I threw it, you would see fruit flying.
I think what you're refering to here is actually a grammar thing. That sentance would be perfectly valid either way if it read "One fruit flies like a banana", keeping the subject/object plurality equal. (I think the ambiguity comes from the fact that "fruit" is both singular and plural)
Anyway, sorry for the ramble, especially since this is just a nitpick, but I couldn't help myself.

As a child, I wanted to know everything. Now I miss my ignorance
[ Parent ]
Really? (3.33 / 3) (#27)
by ucblockhead on Thu Mar 01, 2001 at 01:10:33 PM EST

"Fruit falls like a rock."


-----------------------
This is k5. We're all tools - duxup
[ Parent ]

Am I missing that you're not missing this? (4.66 / 3) (#29)
by _cbj on Thu Mar 01, 2001 at 01:54:50 PM EST

The semantic ambiguity isn't with "Time flies like an arrow," that's just the setup to the punchline. It's between fruit flying as a banana would fly and drosophila liking fruit.

[ Parent ]
Re: Fruit flies like a banana (3.66 / 3) (#30)
by galen on Thu Mar 01, 2001 at 02:04:02 PM EST

Everyone seems to miss the third way to parse "Time flies like an arrow."
The word "Time" is a second person verb, not a noun.

Time flies like an arrow. Time arrows with a stopwatch.
[ Parent ]
I share your skepticism, but... (4.00 / 2) (#23)
by YellowBook on Thu Mar 01, 2001 at 11:59:51 AM EST

I'm quite skeptical of these claims as well -- apparent unknowns claiming to have done something known to be hard through a fairly straightforward method. It shares a lot with the alleged RSA crack we had a month or so ago.

But are you sure there's a consensus in all NLP circles that the Turing Test is meaningless? This article seems to suggest at best that it's not at all a settled question. Maybe it's only considered meaningless in some social circles? Academia works that way -- lots of people can be doing work on more or less the same thing and not talking to eachother. Maybe this is one of those cases?

(I personally don't have a strong opinion on whether the Turing Test would be definative, I am just interested in knowing if there's really that much of a consensus against it these days.



[ Parent ]
The Imitation Game (5.00 / 2) (#34)
by mdavids on Thu Mar 01, 2001 at 09:24:57 PM EST

I would expect Alan Turing to dismiss the Turing Test as meaningless. In the paper that started it all, Computing Machinery and Intelligence. he wrote in response to the question "Can machines think?":

"If the meaning of the words 'machine' and 'think' are to be found by examining how they are commonly used it is difficult to escape the conclusion that the meaning and the answer to the question, 'Can machines think?' is to be sought in a statistical survey such as a Gallup poll. But this is absurd."

So he proposed a completely different question, one that it was possible to sensibly answer: How well might a hypothetical digital computer play the imitation game? He then proceeded to outline his expectation of the potential capabilities of computers.

"I believe that in about fifty years time it will be possible to programme computers with a storage capacity of about 10<sup>9</sup> to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning. The original question, 'Can machines think?' I believe to be too meaningless to deserve discussion. Nevertheless I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted."

"The Imitation Game" was subsequently conflated into "the Turing Test" by philosophers, completely ignoring Turing's point that in the future we will almost certainly speak of "intelligent" computers "thinking", but that is because the generally accepted meaning of the words will have changed.

The initial problem; the problem in arriving at a satisfactory definition of words like "think", and "intelligence", never mind the fact that the Turing Test is also purported to be a proof of machine "consciousness", remains. We don't evaluate the airworthiness of a plane by seeing how many people can be fooled into thinking it's a bird. We know what "fly" means beyond saying it's "that thing a bird does".



[ Parent ]
Suspicious of this (3.66 / 3) (#18)
by Simon Kinahan on Thu Mar 01, 2001 at 11:30:21 AM EST

There's no URL for this company given in the article, and google cannot find their webpage, and the article is very thin indeed on detail, except in saying that they've been constantly patching the program to get it to its current level.

I am very suspicious of AI claims in general, and this one looks no better than any other. Anyone remember cyc ?

Simon

If you disagree, post, don't moderate
AIE (4.00 / 1) (#31)
by oleandrin on Thu Mar 01, 2001 at 03:34:36 PM EST

The scientist interviewed in the article, Jason Hutchens, originated the MegaHAL project (now on Sourceforge), and now works for Artificial Intelligence Enterprises.

[ Parent ]
"Intelligence" (4.00 / 1) (#33)
by mdavids on Thu Mar 01, 2001 at 08:16:24 PM EST

The tone of the article is odd, considering the pretty sensible view Hutchens apparently held at the time he wrote the megaHAL README:

"Q: Hey man, this is freaky! This program is learning! Is my computer becoming real intelligent, or is there a real person connected to this thing? It's scaring me, man! This computer is being blasphemous. It's possessed by Satan! Etcetera...

"A: You're anthropomorphising. Go get a dictionary, find out what the word means, then make yourself a nice cup of tea and have a quiet lie down."

I'm wondering if it's too early for the April 1st press releases. Or perhaps it's just the sensationalists at the New Scientist writing the kind of story they want without letting the facts get in the way.



[ Parent ]
AI Work (3.83 / 6) (#19)
by ucblockhead on Thu Mar 01, 2001 at 11:33:51 AM EST

AI is littered with neat toys that didn't scale. They expect it to advance quickly. I suspect it won't. Typically what happens with these AI projects is that everything seems amazing when they limit it to a narrow domain, but the minute they try to expand it to anything real, it fails.

I'm also highly suspicious because you can also get a gorilla or chimpanzee to talk at the level of a 15-month-old baby. Both gorilla's and chimps can learn on their own as well. Obviously, that is not enough.


-----------------------
This is k5. We're all tools - duxup

-1, and a poll complaint (3.66 / 3) (#21)
by DesiredUsername on Thu Mar 01, 2001 at 11:51:49 AM EST

I voted -1 because I've seen so darn many of these "it does X now, soon it will be standing on it's head and doing taxes" AI stories. I fully expect AI to come to fruition AND I understand the need for micro-domains. That doesn't mean I lend this story any creedence.

As for the poll: What exactly would it mean for computers to *surpass* humans in natural language? It's more natural? More language-y? They add therbligs as a part of speech? Get rid of prepositions? What?

Play 囲碁
"Make a baby AI and teach it" (3.75 / 4) (#24)
by WinPimp2K on Thu Mar 01, 2001 at 12:10:58 PM EST

(Sorry if this is a dupe)
Back in the late 60's/early 70's there was an AI program called Parrie (I think). Parrie was a program that behaved like a paranoid schizophrenic. The developers were confident that they could cure 'Parrie' and would then have a fully human level AI.
So now an outfit in Israel has a 'baby AI' named Hal? All it needs to understand is
"Open the pod bay door".

I think this story is exactly one month early. I also think K5 needs a humour section.
Incidentally, 'Parrie' had a job. 'He' was a postal worker. (I'm not making this up I remember reading about it in Analog way back when...)

Kurzweil AI (3.00 / 2) (#32)
by ucblockhead on Thu Mar 01, 2001 at 05:38:45 PM EST

If you feel like installing a lot of crap and have Windows, there's an "AI" chatterbot you can play with here.
-----------------------
This is k5. We're all tools - duxup
baby AI? Emacs can do better! (3.50 / 2) (#35)
by nickp on Fri Mar 02, 2001 at 01:43:47 AM EST

Just try "M-x doctor" in emacs. You get great AI and it's actually useful: psychotherapy! Just to convince you how good it is, here are some excerpts from my conversation with it (these are genuine lines, I start):

...

I think we are friends, but I am not sure.

Is it because you are not sure that you came to me?

Yes.

I see... Well, why do you say this is so?

Because she seems to avoid me.

Does it bother you that she seems to avoid you?

Yes.

Yes, hmmm... so, what makes you believe this is so?

Well, I am getting very anxious about it.

Are you sure that you are getting very anxious about it?

...

Yes. I'm glad you're getting back to actually trying to help me.

Maybe your plans have something to do with this.

Yes, I want to get married.

Do you really think you might want to get married?

I do.

Is it because of your life that you say you do?

Yes. I'm tired of being so lonely.

Are you lonely often?

Yes.

...

"Gravitation cannot be held responsible for people falling in love." -- Albert Einstein

New Scientist Magazine on AI Machine | 35 comments (21 topical, 14 editorial, 0 hidden)
Display: Sort:

kuro5hin.org

[XML]
All trademarks and copyrights on this page are owned by their respective companies. The Rest 2000 - Present Kuro5hin.org Inc.
See our legalese page for copyright policies. Please also read our Privacy Policy.
Kuro5hin.org is powered by Free Software, including Apache, Perl, and Linux, The Scoop Engine that runs this site is freely available, under the terms of the GPL.
Need some help? Email help@kuro5hin.org.
My heart's the long stairs.

Powered by Scoop create account | help/FAQ | mission | links | search | IRC | YOU choose the stories!