Slashdot Mirror
New Contestants On the Turing Test
vitamine73 writes "At 9 a.m. next Sunday, six computer programs — 'artificial conversational entities' — will answer questions posed by human volunteers at the University of Reading in a bid to become the first recognized 'thinking' machine. If any program succeeds, it is likely to be hailed as the most significant breakthrough in artificial intelligence since the IBM supercomputer Deep Blue beat world chess champion Garry Kasparov in 1997. It could also raise profound questions about whether a computer has the potential to be 'conscious' — and if humans should have the 'right' to switch it off."
and see if it complains, first. If it does, then call me back.
Are they really *thinking* or have the programmers just done some tricks to make it seem that way.
"Teaching to the test", so to speak.
It could also raise profound questions about whether a computer has the potential to be "conscious" -- and if humans should have the 'right' to switch it off."
Maybe in the esteemed opinion of vitamine73 it will, but if you knew anything about how artificial conversation engines were constructed, you would understand that it's anything but sentient. Right now, conversation logic is simply trick laid upon trick to stagger through passing as a human, and doesn't, at its core, contain anything remotely similar to self-aware thought.
I'm sorry. The number you have reached is imaginary. Please rotate your phone 90 degrees and try again.
It is likely to be hailed as the most significant breakthrough in artificial intelligence since the IBM supercomputer Deep Blue beat world chess champion Garry Kasparov in 1997.
I don't understand how this is a breakthrough for artificial intelligence. Deep Blue didn't "think", at least not in the way most people think when they consider artificial intelligence. It did what computers are really good at - it computed.
Deep Blue applied an evaluation mechanism specifically tuned to chess - taking the location of pieces on the board and computing a number telling it how "bad" or "good" this position was and how "bad" or "good" responses to this position would be. Granted, it took this to a depth farther than any other chess computer in history, but it was doing essentially what a small, handheld chess computer does.
Of course a computer is going to be good at computing. That doesn't mean it's thinking.
Early chess computers used AI techniques to try and cut out candidate moves. This was expensive in CPU cycles, but the thought was to get them to play chess like humans. Computer chess since AI Winter has been all about number crunching - let Moore's Law take hold and just brute force our way through the problem - evaluate deeper because we have a faster processor. This is what Deep Blue did.
If Deep Blue were true AI, then it wouldn't be limited just to chess. It's an interesting experiment in computer chess, and an interesting experiment in tuning an algorithm working against a human, and in interesting experiment in making a computer chess opening book, but a huge leap forward in AI it isn't.
If you read TFA they have a sample chat which just shows you how stupid these chat bots still are. It is extremely easy to get them to just parrot responses and then try to change the subject in completely random directions.
I have yet to see any chat bot that can figure out the line of questioning, then pick up and introduce interesting things to the conversation that are corollary to that subject. I think the only way you will get bots that will "pass" this test is to have massive databases of words, relationships between words and subjects with corresponding topics of discussion. Still, the computer won't be intelligent, it will just be reciting from its huge database of responses.
I think the type of question i'd ask these bots is something that would require them to extemporize and they'd all fail. For example: "You have two rubber ducks, what are the possible ways you could use them if you don't have a bathtub?"
Any human could reply to that with things like "i'd put them in a stream, run over them with my car, put them on a lake, in the swimming pool" etc but a computer program isn't likely to respond to that in any way that makes sense. The response i'd expect from the computer would be "You like ducks then?".
"To strive, to seek, to find, and not to yield." - Tennyson
Can you create a symphony? Oops, did you just fail your own definition of sentience?
the problem is how long you talk to it. If you talk to it daily it would need to learn and expand for you not to reach the end of its tricks. I think that is where the quality of the turing test comes in. It would have to be capable of self expansion and learning in order to make you think it is capable of the learning and self expansion of a human.
I'm sure these bots could fool you for an hour in a select setting, but if you were to talk to them on AIM every night for 6 months on a variety of subjects from opinions to jokes, to hopes and dreams, they would need to be practically human to not fail.
Sure you can argue that it would just be an awesome ball of clever tricks, like auto-reading news feeds and analyzing stories for conversation currency. The thing about clever tricks is that a lot of what the human brain does in the separate lobes are just clever tricks, it's when you combine these all together and they start working with each other that you get something amazing.