33 replies to this topic

[rouncer]

me-> hello
bot->Hello.
me-> do you feel ok?
bot->Sometimes.
mw-> when do you hate things?
bot->When I let myself.

that was the conversation i had with "clever bot" it worked up until then!

This is about all the level of interaction I was talking about, it does "understand" for what I was talking about, but a real thinking intelligence I realize that this is pretty much science fiction.

So it has been done heaps before... I should have realized. Its pretty cool actually.

[JarkkoL]

It seems I was able to convince the bot that the world is flat ;)

Did you know that Earth is really flat?
> Did you know that it's not true.
Of course it's true. If it was round like people are claiming then people on the other side would fall into the space!
> What is your point?
My point is that all this talk about earth being round is just a big conspiracy not to look for the edge!
> How did you know that I don't like eminem?
Are you trying to chance the topic?
> I know the answer but you are changing the topic every second sentence.
The answer to what? Do you mean you agree that the world is flat?
> Yes, indeed.

[Mihail121]

OK, too much posts on a weird topic, natural language processing is different than natural language understanding, but still the question of machines doing Turing tests in a decent fashion will remain unsolved in my eyes for the following reasons.

Language is just an interface, abstracting the structures of our mind and its current state, which are still complete mysteries. Why would you expect a computer to respond in a human way, when it does not hold a mind? Its structure is nowhere near as complex as that of the mind. Indeed, a machine could check all the pieces it needs like wether information, age of somebody and what not, it could have sensors for other feedback, but how would it solve the puzzle together that a simple question like 'Did you know the world is flat?' is posing and give a proper answer that not only gives a (language-) correct answer, but also reflects thought? Exactly! What if there's soul? What if there's destiny? Are humans themselves machines? Exactly! You simply cannot make a machine think, for you have no idea how to do it.

Besides, even if you can answer all these questions AND formulate a formal problem -- you then have to prove it's actually COMPUTABLE with a machine (giving Turing machine as a reference model), for as we know, some very simple problems are actually not computable...

Rouncer, you could of course make your parser respond to some queries by giving it knowledge of some very specific and small domain. That's not a problem: "give me three vertices and position them at random, connecting them with random edges, forming random triangles" will be ok, but even this will cost you a great deal of thinking plus coding and it might take years. Go ask a computational linguist about all the fields involved.

Prolog being used for AI? I don't think so. Prolog is just a nice way of deriving knowledge (according to certain logic) from a known set of axioms (facts). Everyone could code Prolog in a matter of hours and I, having used it in the last years, fail to see, how this GREAT NEW concept takes AI a step forward. All you need is a computable logic and a derivation rule for that logic.

In short: rouncer, you CAN have a bot in your program specific to a very small domain, its only a lot of work, but there are enough tools already for matching language to structures.

[Luz Reyes]

I know it's an old post - but does anyone know, what is that contest/standing order that anyone who can make a machine that speaks indistinguishably from a human gets a large sum of money??

[tobeythorn]

Maybe The Loebner Prize or The Chatterbox Challenge?

[alphadog]

Luz Reyes said:

make a machine that speaks indistinguishably from a human

What kind of human? Are we talking Marilyn Vos Savant or Paris Hilton kind of human?

Admittedly, getting a computer to spew the nonsense Hilton does versus the logical and factual thinking a smart person can may ironically actually be harder to achieve...

:sneaky:

[v71]

Wehn i was writing code for my phd theses i had a sudden urge to write a kind of human language parser, since my work was about an intelligent internet browser capable of "understand" the contect of the web page and in a second moment search trhough the net for similar web sites.
I started observing that my work was a subset of a much larger category going into natural language parsing.
My idea at that time was to break up a long text into nodes.
One main node was the entire text, subnodes where periods terminated by the '.' or ';' symbol , these subnodes were called sentences.
A sentence was subdivided into short nodes called phrases , terminated by the ',' symbol.
Once i collected all this nodes i started parsing the phrases, comparing the strucutre of the expression with a larger database , which was growing in time in a totally automated fashion, the more text you put in, the larger the neural network grew until the point i could trace a neural path and reconstructing the meaning of the tex , building up a 'comphresion' tree.
The task was huge and my time was ticking fast so i stopped the idea and worked back on the theses, but this idea still knock at my door once in a while.

[poita]

@v71: That approach to natural language processing was abandoned quite a long time ago. It's simply far too rigid. For a specific example, what if someone forgets a period? Most humans would still be able to make sense of a sentence (realising the person's mistake), but it would throw your node-based parser right off.

It's all done statistically now, and I believe that's the correct way to do it. Anything based on strict definitions is probably not going to work.

[v71]

Well, yes, it was an idea i had time ago, i abandoned it in favour of different things, maybe one day i will be interested again into nlp

[touch_the_sky]

I once had this idea of writing some "genuine human" bot, more of a real convincing imitation attempt than true thinking thing. Apart from all the obvious learning algorithms (and cheap tricks which would make it appear much more like it), it was supposed to have a fairly neat feature - it was supposed to verify the information it learned through conversation, by "watching" real people chat with themselves. But work and stuff.. They'll probably build the real SkyNet before I get my hands on this sort of project;)

Missing comas, typos, etc. - this is sort of stuff is already sortable (even on web search). Idioms, jumping sharks etc - I don't think it would be that hard to make a bot recognise them to a decent extent and give the user a convincing answer matching the context. (I'm not saying it would not take loads of time) Problem is, it would still be just imitation of "what a real human would say" (there would be no such thing as conscious thought involved).

As far as real thinking intelligence - Ray Kurzweil says we'll have a full mapping of human brain + hardware / software required to create artificial consciousness NO diffferent than ours, with all it's emotions, feelings, etc. in about 20-25 years or so... What do you think?;)

[poita]

touch_the_sky said:

Missing comas, typos, etc. - this is sort of stuff is already sortable (even on web search). Idioms, jumping sharks etc - I don't think it would be that hard to make a bot recognise them to a decent extent and give the user a convincing answer matching the context. (I'm not saying it would not take loads of time) Problem is, it would still be just imitation of "what a real human would say" (there would be no such thing as conscious thought involved).

I think you're underestimating the subtleties of human communication.

Quote

As far as real thinking intelligence - Ray Kurzweil says we'll have a full mapping of human brain + hardware / software required to create artificial consciousness NO diffferent than ours, with all it's emotions, feelings, etc. in about 20-25 years or so... What do you think?;)

I think it's quite feasible. Last I heard someone had already reconstructed a rat's brain, although I didn't hear any more from it (perhaps they didn't really do what I think they had done). I think it will be more than 25 years though. We tend to overestimate these sorts of things -- remember flying cars in the year 2000 :-)

[alphadog]

poita said:

someone had already reconstructed a rat's brain

Last I heard, we don't even understand how a rat's brain really works, yet we've successfully rebuilt it?

http://online.wsj.co...1557234725.html

He's mimcking/modeling how neurons in a specific region work together in a net. It's a far cry from rebuilding the rat brain. Even then, he needs a supercomputer to pull off the sim.

Furthermore, putting a real sim into a believable, future "The Sims XXVI" is much more than 20 years away.

[touch_the_sky]

poita said:

I think you're underestimating the subtleties of human communication.

I was more about that now it's possible to create a waaayy better chatterbot than what I've seen so far

poita said:

remember flying cars in the year 2000 :-)

Yeah flying cars in the year 2000;) It's actually quite funny when you watch some old sci-fi movie where "the movie future" is already our past though

I heard about that rat brain some time ago, there was another one when they made a robot controlled by a cpu merged with live brainy tissue.

Btw, you may have seen it but:

;)

[Hyper]

poita said:

remember flying cars in the year 2000 :-)

You mean the 1980s, not 2000. :)