Friday, January 14, 2011

IBM super computer Watson beats Jeopardy players

This is quite a big step. Watson understand natural language questions and uses a range of algorithms to give sensible answers. Not to long from now Watson will be the world champion. More @ Engadget

6 comments:

annom said...

This is a big step. Been waiting for this to finally run.

Gives us a look into the future.

I wonder how long this (these) algorithm(s) take to run on a normal computer.

cybrbeast said...

Quite some time. It currently uses a high end super computer to give split second answers. So we need a few more Moore Law cycles before it fits into a standard computer.

Now it also uses a really brute force method, I think the algorithms could be improved a lot. Deep Blue, the chess computer that beat Kasparov used brute force. A few years later Deep Fritz managed to perform almost as well by using clever software, while running on a single desktop.

annom said...

It's not really fair to call the algorithms of Deep Blue and Watson brute force. They use a lot of computing power, but you can't brute force a game of chess, even with modern supercomputers. Let alone play Jeopardy, which is not a game of trail and error.

But I see what you mean. They can be improved. Deep Fritz was about 20x more efficient than Deep Blue.

cybrbeast said...

I didn't mean brute force in the strict way of trying every possible path, more of just throwing heaps of computing power into very unoptimized task. T

hat's of course how all these developments start. Once they know what works well and how it does that, how it could do it better, then they will be able to make it much more efficient I'd guess.

cybrbeast said...

Wiki on Deep Blue:

The system derived its playing strength mainly out of brute force computing power.

annom said...

Well yes, although others will tell you no, not really. The definition of brute force is a big vague. Some call our brain a brute force machine that is "just" wired and tweaked by a genetic algorithm for pattern recognition, mainly based on a huge database (memory).