![]()
Visiting guest. Why not sign in?
|
|
Approximating Life About Alice and its Creator | |
Approximating Life
The NY Times has an extremely interesting article about the creator of the Alice Bot; Approximating Life. Like all the other articles that go so far behind the scenes, it does reveal some surprising details -- notably Richard Wallace's habits (a euphemism you'll understand after reading the bit about his manic depression). I don't know if its a strange side effect of success, or a cause of it, but either way it's got him two Loebner prizes! The second page of the article goes into technical details, showing that human questions rarely start with anything else than one of 2,000 words. The database of 40,000 responses can then be matched as appropriate. ""Wallace's skill is not merely as a programmer but as the author of thousands of sharp one-liners for Alice."" Wallace has some quite drastic options about human ability in conversation, claiming that we over-estimate too much. I couldn't agree more, humoristic one-liners get you out of trouble in pretty much any situation. It's a shame that development on the bot has been so jagged, just like Wallace's life... |
|
just a huge programmed database?
It seems to me that these types of advances in AI consist of bigger databases and more creative programming techniques for how the program should respond to questions. There is not any real learning going on, or any actual understanding where the program can match a word to a concept, and infer other things about it based on the match. I still think the key to an AI "breakthrough" is a machine that learns the way a human learns. I think when we finally build an intelligent, conscious machine, we will have to teach it the way children are taught. Although, I hope it will learn much faster than a child. Rob |
|
re: just a huge programmed database?
> "I still think the key to an AI "breakthrough" is a machine that learns the way a human learns. I think when we finally build an intelligent, conscious machine, we will have to teach it the way children are taught. Although, I hope it will learn much faster than a child." Not necessarily it would learn faster; maybe each learning step would last for many years of trial and error. The real advantage is, once the electronic mind has learnt everything, you could easily clone it by replicating its physical support (i.e. its actual hardware, software and data), and that's something you can't do with a human brain. |
|