- I have been following IBM’s Watson and it’s technologies; Final Jeopardy was of course, a must read for me.
- There are only very few books that are interesting read, even after you know the ending. “Final Jeopardy” by Stephen Baker is definitely one of them. We all know the result of the Jeopardy tournament with Watson, Jennings and Rutter.
- But the journey from a “fly on the wall” perspective is a fascinating read.
- A ring side seat the book provides for a vicarious ride through this age’s most interesting AI/NLP event is priceless …
- Stephen has covered all the important aspects of Watson – from building a brand to the rationalization behind Watson’s Avatar to the algorithms and betting strategies – the breadth is amazing. There are tons of side stories and vignettes that add to the main theme. And Stephen has weaved them skillfully into the main context that while we do get the details of stories, we do not get side tracked. I was looking for architecture and technical information, of course, this is not a technical architecture book. I hope IBM writes a red book on the Watson architecture and internals.
- The book takes us through the initial ideas to build a machine that competes in Jeopardy to the final match between the machine & two humans. Stephen has done a good job in pacing through the various aspects of the machine, that we can feel the agonies and ecstasies of the team & the intricacies involved in teaching a machine how to participate in a game where the nuances are challenging.
- Note : I wrote this blog on March 1, 2011. It is Feb 2013 and I am rereading the book and updating the blog. We are getting there …
The Philosophy of an “Intelligent Machine” :
- As we read through the book, we get a good glimpse of the prevailing attitude, arguments and different camps of building intelligent machines – the so called pragmatists vs. visionaries.
- From the team’s perspective, “they didn’t need a genius, … but build the world’s most impressive dilettante” “battling the efficient human mind with spectacular flamboyant inefficiency”. This makes lots of sense, “transistors are cheap and plentiful…”
- The team also realized that this is as much a story about the people who built the machine, as the mechanics and mechanisms of building a machine. Many of the campaigns were run more by instincts and less by formal marketing analysis “the way Watson could never compute … from the gut”! Many skeptics thought “the machine was too dumb, too ignorant, too famous and too rich – who knows nothing and cannot form theories”
- Interesting view point. But then do we expect the machines to think like humans or act like humans ? I don’t think so, in fact most probably the singularity will be achieved by machines which are very different from us, and do things in a vastly different manner. So I am fine with Watson and the way it works.
- The question of forming theories or modeling is interesting. This is what I was alluding to in my earlier blog – data with interactive context!
- And that solves the theory problem, partially. And as Stephen describes eloquently, faster computers can crunch through algorithmics and search, “turning its pile of data into science” and isn’t that part of intelligence ?
- I do not see any difference between simulated intelligence and artificial intelligence, if there is one.
- Is the knowledge from belief any different than inference from facts attained via data crunching ?
- [Update 2/10/13] There is an interesting observation about language – The flexibility of language is a strength not weakness, humans need language to be inexact otherwise we would need a vocabulary of billions. But this gives lot of trouble for the machine which can handle a billion deterministic vocabulary than nebulous contextual meanings …
The Teaching Algorithms :
- Usually in machine learning one trains the machines, but in this case it is literally teaching. How deep and how broad was the challenge – too much information does create bad answers.
The Challenges :
- Lots of interesting challenges including train wrecks, filtering not-so-good words, nested decompositions, the confidence level i.e. the ability to know what it knows – You got to read the book to get a good understanding of how the team confronted & solved them, within a very short period of time.
- The betting strategies and how the machine went about selecting the clues is an informative read
- Clue : Not that different from the best Jeopardy players !
The Match :
- What I liked about the match was the way Stephen takes the reader through it – he combines the players into a single collage overlapping temporally so that we get an absorbing progressive narration with occasional flashbacks.
The Future :
- One of the questions the team had to wrestle with, a very common question in machine learning is over fitting. Could the skills Watson acquires be applicable in other areas ?
- The answer is yes, augmenting humans in areas like health care – with it’s range rather than it’s depth.
- As I said earlier, a compelling book that should be read by both the techies and non-techies.
- The book, while very informative about the computers, is not technical at all and that is good!
- An impossible feat, but Stephen Baker achieved it !
- And, he did it combining human interest with computer technology …
- In short, let me repeat, a ring side seat the book provides for a vicarious ride through this age’s one of the most interesting events in AI/NLP is priceless …
Am interested to know what you think after reading the book …