Watson Wrap-up: Why Humankind's Inferiority Complex is Unfounded

Ben Templeton wraps up the Robot Jeopardy Fiasco of 2011 with a refreshingly sober dose of anti-hysteria.

- 44 Maagnum

Well, days 2 and 3 were significantly more sensational that day 1, and produced much more of a reaction than the first day. I scanned some of the comments from the public on online articles from some major news outlets. Widely regarded as the lowest-quality thinking that can be found anywhere in the world, these comment threads provided me with a few points that seemed worth discussing.

Meatbag concedes to Watson

First, the huge margin of victory for Watson: A common reaction (online and live) was ‘of course Watson won, he gets the questions in text, and he can press the button so much faster!’

Both of these things are true, but also not the real point. The reality is that actually appearing on Jeopardy is basically a publicity stunt for IBM, to capture the imagination of a society that is increasingly focused on Farmville, and decreasingly on using technology to resolve problems and conflicts. But while IBM definitely cares about engaging the public (otherwise they wouldn’t have bothered with the whole thing at all), it’s also concerned with demonstrating a huge scientific breakthrough to the technological community. To that end, the fact that Watson can answer questions posed in Natural Language is evident from the fact that it produces the correct response in 3 seconds, not by the $70,000 that it accumulates (which is, incidentally, the price of 2 of the 2,880 Power7 servers that make up the supercomputer, available for purchase on IBM’s website). And while the public is largely concerned with the perception of the supremacy of the computer, the truth is that Jeopardy is, in and of itself, trivia.

Worthwhile Computing Challenge

But while people seemed quite passionate about Watson’s victory over Mr. Jennings and Mr. Rutter, they seemed thoroughly unimpressed with the technological achievement. So I’m going to talk, very briefly, about the challenge that Watson faces, and the significance of its level of success.

Computers only speak languages which are unambiguous. In other words, when a computer programmer writes a program, the computer can only understand it if there is exactly one possible interpretation. An example of such a sentence in English might be: “If x is equal to one, then set y equal to two.” This sentence is very straightforward, and essentially only has one interpretation, under the definitions of all the involved words.

The English language (and every human language, although to interestingly varying degrees) is not a completely unambiguous language. Humans generally (although not always) can pick the correct meaning of a sentence through experience with the language, accumulated over several years. Watson lacks this experience.

Consider the following sentence: “I’m out of the building.” This could take several different meanings, all of which are logical to varying degrees. The most obvious one is that the speaker is not located in the building. Another interpretation that would make sense but for the definition of “building” is that the speaker is using “out” as if he or she were a waiter telling a customer that “we’re out of the salmon tonight.” There are plenty more, as each word has lots of definitions, and many of them work from a strictly grammatical sense.

The fact that people don’t keep a stock of buildings is what provides the computational difficulty of Natural Language Processing (or NLP, not to be confused with Neuro-Linguistic Programming, a controversial psycho-therapeutic technique that shares the same acronym). Although everybody knows that one can’t be “out of the building” in that sense, there is absolutely no indication of that in the definition of any of the words. What IBM accomplished over 4 years in Watson was getting a computer to figure out which interpretation to use with remarkable accuracy. It does this by searching through massive amounts of data and performing statistical analyses, connecting words to their contexts, seeing which words go with which other words. In this way, Watson is able to figure out which interpretations to use and what the proper context is, without the human experience of learning language.

Given that bit of insight into how Watson functions, let’s take a quick look into the significance of IBM’s achievement. There is a lot of talk of Skynet, and the robot revolution, etc. An important thing to recognize is that Watson is a breakthrough in Computational Linguistics, not really Artificial Intelligence in general.

A Few Steps Short of Skynet

At the same time, the process of connecting words into their contexts and figuring out what a sentence truly means is an incredibly important tool. The most commonly discussed application is diagnostic medicine. In this situation, symptoms are analogous to the ambiguous words of a sentence (in that symptoms, like words, can have a variety of different meanings depending on the context), and Watson could synthesize a diagnosis from the combination of a set of symptoms in their context.

Medicine is one of many cases where the ability to synthesize natural language data into a context could be game-changing. But ultimately, while extraordinary, the technological breakthrough of IBM is not something that is yet able to replace human intelligence. CNN sums it up well by positing that “Watson’s eventual commercial incarnation will be a tool, not a human replacement.”

There are lots of other concerns and issues, but the three above seemed the most important. While yes, the publicity stunt aspect of Watson was exactly that, the technology is undeniably impressive. While the physical hardware is fairly cool, Watson falls short of the most powerful computer in the world. The breakthrough of Watson is the algorithmic development in statistical analysis which allows Watson to be incredibly effective in one of the most challenging fields of computer science, Natural Language Processing.