Wednesday, February 16, 2011

Final Jeopardy: Our computer overlords are here

This is the my final of three posts on the historic match of computer vs. human in the game of Jeopardy. The first two posts are here and here. I will also speculate below on when and how computing power of Watson's caliber, and beyond, will find its way into our daily lives.

So  Watson won the IBM Challenge. A lot of us expected it, especially after Watson started dominating in the previous round. It was great to see Ken Jennings surge midway through the final round; I am a little disappointed, however, that he didn't make a larger wager on Final Jeopardy. He wouldn't have won, but it would have been closer.

Watson had quite a few wrong responses today, most particularly in a category in which the answers were all supposed to be the names of keys on a computer keyboard. There are two things interesting about this; the first is that Watson didn't understand the category to start with (and knew it). The second was that Watson didn't learn after noticing the correct 'questions' to the first couple of 'answers'. Machine learning is actually quite an advanced aspect of AI, and I am sure that a little bit more learning mixed in to Watson's many algorithms might have made a difference.  As with the interesting types of mistakes made in the previous days, I am sure the engineers would try to fix this, if there was to be another series in a few months time.

I wonder if there will ever be any more Jeopardy games pitting humans against powerful computers? Now a computer has won, it won't be so much of a sensation. However, it might be nice to have computer vs. computer matches every now and then. The computer's goofs and quirks (such as betting odd amounts) are still pretty interesting to watch, so I am sure there would be an audience for this.

There should be a competition to come up with ever more tricky games that humans can figure out but computers still have trouble with.

Within a couple of decades, I expect we will see computers of Watson's capabilities on everybody's desktop, and a few years beyond that, in everybody's phone. At that point in time, you will be able to buy Jeopardy software to play against, and you will be able to handicap it to match your skill level. Incidentally, I don't think this will all come about simply by computers getting faster and faster. There will be considerable speed increase, but most of the gains will come from a combination of increases in the amount of memory available, massive parallelism, architectural innovation and algorithm innovation. Perhaps even quantum computing may be part of it, and certainly spintronics optical connects and other yet-to-be-invented concepts will likely be part of the mix. I think hardware elements that emulate neurons might well be involved, and this will open up a whole new area for software. Certainly we will move from electronics on 2-D chips to fully 3-D structures; a lot of work will have to go into using energy super-efficiently so these do not overheat.

With Watson on the desktop, the supercomputers of the coming age will likely be mind-blowing in their capabilities. Hopefully they will routinely be put to uses that will have major benefits for society: Quickly inventing and testing new drugs in a fully simulated environment; ever more accurate forecasting of weather and climate change; optimizing traffic flows through cities, and maybe even helping humans negotiate lasting solutions to social problems and conflict situations.

For all of this to come about, we need a continual supply of bright students entering the fields of computer science, software engineering, computer engineering and electrical engineering. There is a shortage of such people now in the Western World, even though the opportunities in the years to come are immense.

Incidentally, in case you are wondering, the subtitle of this post is a  quote from contestant Ken Jennings that he wrote as a humorous addendum to his Final Jeopardy answer. I actually think that there is a 70-80% chance that we won't ever have actual computer overlords. I am optimistic that we will be able to evolve technology in such a manner that we remain the masters of the computers, even though the latter may approach sentience. That said, we must heed the warnings in many science fiction stories, and realize that powerful computers pose a huge risk, just like many other technologies humans have developed.

No comments:

Post a Comment