Surely everyone has heard that IBM’s Watson virtually crushed its meatbag counterparts in the epic trivia game Jeopardy. What? You haven’t? I suppose you’re not nerdy enough. Lucky for you…I am!!
But first a bit of background. Alan Turing is often considered the father of Artificial Intelligence (AI). Besides being absolutely brilliant he was one of the first to really recognize the power that computing could have on our world. Many of his predictions in AI have been very accurate. In 1950, Turing introduced a concept that has become known as the “Turing Test.” The Turing Test is considered to the be the grand challenge of AI.
Here’s the scenario. There are three players, A, B, and C. Player A is a computer, and players B and C are humans. Players A and B are not visible or otherwise physically distinguishable to player C. Player C (the interrogator) gets to ask any question to players A and B. If, via these questions, player C cannot deduce which player is the computer and which is the human, then the computer is said to have passed the Turing Test.
It is fairly clear that this grand challenge is still many years away. Most computers, even those having extremely sophisticated AI algorithms, can typically only carry on approximately a five minute conversation with a human before the human gets bored and gives up (Emacs psychiatrist anyone?). But, in a particular narrowly defined scope, we have many machines that have passed the Turing Test. For example, in 1996 Gary Kasparov (a world chess champion) lost to IBM’s Deep Blue, a supercomputer. It is easily demonstrated that, at least in the narrowly defined world of chess playing, Deep Blue passed the Turing Test. That is, a human observer, based on the chess playing alone, would not be able to distinguish between the world champion chess player, and Deep Blue.
This leads us to IBM’s Watson, and the recent Jeopardy episodes featuring him. It’s fairly obvious that in the narrow world of Jeopardy, Watson passed the Turing Test (beating the two most successful champions). In fact, not only did he beat them, he demolished them, ending with a total of $77,147 for the two game total, compared with Ken Jennings’ $24,000, and Brad Rutter’s $21,600.
It is by means an overstatement to say that Watson is an amazing supercomputer and that his AI reaches into new territory never before seen in the computing world. The natural language processing, capture of statistically significant information, in conjunction with (most of the time) correct inference is absolutely incredible!! When one considers what it takes to use 1s and 0s to represent, process, and store the kind and amount of information we’re talking about, and how much ambiguity there is in natural language, it is remarkable how far we have come.
From a theoretical AI standpoint, it really is a new era of computing in which computers cannot only provide us with loads of information (a la Google), but it can also give us the right information. The ability to interpret questions, even ones riddled with metaphor, idioms, and other tom-foolery opens doors of analysis with speed previously unknown. Further, the computer can reach into its vast database of knowledge (I call them Google Dumps), and somehow retrieve the right information (think about how long it sometimes takes you to find the information you want on Google). Being able to command that amount of information at the speed Watson is capable of has limitless possibilities.
The Not so Remarkable
Now for a bit more skeptical view. Let’s face, Jeopardy is impressive because of it’s breadth. It has virtually no depth. The questions are all simple enough to be answered by a High School Sophomore who has access to Wikipedia. Sure, when you watch Jeopardy you probably don’t know most of the answers. But if a topic comes up in which you’re an expert, it’s obvious the questions are terribly trivial. What makes Ken and Brad so amazing is their ability to remember and quickly recall trivia that spans the spectrum of topics. Nevertheless, with Wikipedia, I contend virtually anyone could correctly answer almost all Jeopardy questions on a single show in a relatively short period of time.
Furthermore, from watching the show it is clear that all three players knew the answers to almost all the questions. So why did Watson win? It was not because he knew the answers to questions that his decimal-crunching competitors did not (that would be impressive and would really inaugurate the takeover of our robotic overlords)! No, he won because he could ring his buzzer faster. The fact that a computer can respond quicker than a human is not at all a remarkable feat. That has been happening since the dawn of the computer age.
All in all, I was thrilled to watch the match. I was delighted to see Watson completely botch some super simple questions (like Final Jeopardy in game 1 in which Watson thought that perhaps Toronto was a U.S. city). Is this a major leap for AI and computing in general? I think so, even while acknowledging that in many ways it is not particularly impressive and certainly Watson is still light years behind having the kind of intelligence, reasoning, and consciousness that even the most uninspiring bipeds have.
So what say ye? Are you concerned about the influx of our increasingly intelligent binary slurping creations? Is Watson remarkable to you? Are we reaching the limits of what computers can accomplish, or only beginning? Perhaps the most relevant question of all, is this a step toward robotic consciousness, or is such a thing unachievable?