I wasn't planning on following the poll up with anything, but after finding this I feel I must. Actually I think I might do a part 3 at some point now too.
Artificial intelligence is a fascinating and at thrilling field. So far, some of the most notable achievements are an unbeatable checkers player, and a chess world champion (though technically not unbeatable). But most of the achievements still don't really feel like intelligence, at least in the form that humans have. I'm not saying that they aren't impressive, but they lack the kind of nuance that the human brain is so remarkably good at.
Solving checkers, after all, only requires knowledge of all of its 500 quintillion (5x1020) possible configurations, and choosing a path between the current one and a winning one. This is an impossible task for a human brain, but rather simple in concept and well within the limits of memory and computing power of around 100 desktop computers running for 18 years. Winning at chess is a bit more complicated (and a sufficiently large problem that there is currently no prospect of actually solving it), and requires cleverer algorithms. But the principle is still fairly simple: evaluate a set of possible moves several steps into the future and choose the next one based on which one is most likely to produce the strongest final positions. So good algorithms now need to include some forms of pattern recognition and machine learning, in addition to relying on being able to perform over 10 billion operations per second. However, successfully flirting and getting the phone number of a girl requires a lot more "intelligence" versus computing power (actually that sounds like an interesting AI project). But some of the latest AI projects are starting to achieve things that actually seem human.
Enter Watson. Yes, the next match between human and artificial intelligence will take place under the auspices of everyone's favorite Canadian, Alex Trebek. Read more about it here and here, or google it yourself.
I don't know about you, but I find this incredibly cool. It's not surprising that a computer would be good at searching a database of knowledge larger than any human could possibly dream of possessing. But to actually be able to interpret the meaning of a question and then search a database of knowledge is perhaps a quintessential example of the nuance of human intelligence. After all, it takes us quite a few years to master the concept of language, and our brains have been shaped by millions of years of evolution in order to be good at doing this task. Watson certainly isn't as good at it as we are yet, but to think that a computer can even do it makes you think that the science fiction dream of machines that behave just like humans may not be that far off after all.
The first of three matches airs tonight, so check your local listings. Here's a neat little preview:
If you're looking for more fodder for your fear of the robot apocalypse, consider the fact that the Watson project at IBM is about 7 years old. Ken Jennings and Brad Rutter, Watson's Jeopardy! foes, are 36 and 33 years old respectively. If Watson wins, be afraid, be very afraid.
This is cool. But it's still far from the human mind. I am just guessing, but I assume Watson thinks similarly to how a google search thinks. If you type most not-so-strange questions into google, you can find someone else asking a similar question that has already been answered.
ReplyDeleteI don't intend to come off like I am not impressed. I am. Especially how Watson's creators have effectively programmed into Watson lots of the intricacies of the human language. Personally, I don't think I could ever feel the same way about a machine as I do a human. There is an unpredictable element of our species that you cannot capture in artificial intelligence. I think it's amusing that we know so little about the human mind yet are trying to mimic it.
Seriously though, I am impressed.
I don't know exactly how it works either. But I know it has many algorithms for trying to determine what the answer is, and many of them are not very similar to a google search (it uses a database of books and such, not the internet, for starters). So it has to use machine-learning to figure out which ones to rely on in which situations.
ReplyDeleteI agree that Watson is still a bit gimmicky, and it certainly doesn't understand language like we do. And while we don't know much about the way that our own minds work, I don't think this means that we ought to dismiss the possibility that we might be able to create something very similar. There is a lot of cool work going on with algorithms that learn. The human mind may always be a black box to us, but if we are able to create something that can learn to adopt new behaviors just as we do, does it matter if maybe it does it in a completely different way? Do we even have to know exactly how it does it? (That last question may not be as stupid as it sounds.)
I didn't mean to make it sound like it uses the internet. The way Google searches is through some crazy linear algebra that is useful for searching through databases.
ReplyDeleteI'm not saying artificial intelligence is a bad thing either. I think it will be an important part of humanity's future. Ted makes a good point: teaching a machine to think based on a completely different way from how humans think is a very impressive task. That type of research is getting a lot of attention because organisms that have evolved over millions of years have very efficient systems (photosynthesis, our immune system). If we could make solar cells that have the efficiencies close to those of plants, our reliance on fossil fuels would plummet.
The applications are actually way bigger than I thought. For example:
ReplyDeletehttp://www.youtube.com/watch?v=uWHG7DMLurE