On February 14-16, 2011, we watched “Watson“, an IBM system of computers the size of a two-car garage, compete against the all-time best champions on the game show Jeopardy!®. On subjects from diseases to characters in Beatles® songs, the computer was impressive, pressing the buzzer faster than its human opponents, and getting so many answers correct that its occasional gaffe seemed like a deliberate act designed to preserve the hope and dignity of our species. The technology was impressive! In just a few years, systems like Watson will fit in our mobile phones (making today’s “smart” phones seem like stone tools.) We will come to rely on their ability to parse language and retrieve factual information as a way to extend our own memories, just as we now have electronic organizers that help us to remember phone numbers and appointments.
It is clear that we will soon see uses of Watson-type systems that will greatly advance how easily we perform a number of common tasks. For instance, Watson interprets the “questions” (answers, actually) on Jeopardy! with very powerful algorithms that seem to understand human language. This part of the system will find immediate employment in technical and customer support systems, where just understanding the question and figuring out who (or what) should provide assistance is more than half the battle. Watson’s ability to quickly search a large and diverse database, and to weigh the probable correct answers will also prove valuable as a decision tool. Just having the top three probable responses to choose among in difficult situations will help us in areas as different as picking a mortgage, routing air traffic, and diagnosing diseases.
Looking beneath the “hood” however, the real-world flaws in Watson are readily apparent to those with a background in computer databases, search and natural language processing (ok, well, to me, at least, since these fields were central to my system designs for my first company, Infonautics, and its product, “Homework Helper”, way back in 1990.) First, Watson relies on a professionally-constructed database of information. Want to beat Watson at Jeopardy? Just make it search the Internet in addition to its own databanks. The mass of conflicting data, divergent views, outright lies, and incomplete reference in which we live our lives is still well-beyond the capability of even our smartest machines. Watson may be able to compute a clue, but when it comes to humor, sarcasm, inuendo, and propaganda, Watson is clueless. Next, show Watson a painting as a clue, and ask not who painted it, and when, but what it means, or what feeling it evokes. To humans, the Mona Lisa’s smile is inscrutable. To Watson, every work of art is just a catalog entry.
All of this is not a criticism of Watson or its creators. It is, however, a criticism of those who conclude that when it comes to intelligence, if Watson can beat Ken Jennings and Brad Rutter, it is “game over” for our species, and we are weeks away from SkyNet, the malevolent computer network of the Terminator movies that takes over the world. In a recent article in the National Law Journal, Robert C. Weber, IBM General Counsel wrote about using Watson’s technology to aid in the courtroom. “If a witness says something that doesn’t seem credible, you can have an associate check it for accuracy on the spot.” Lawyers already know that human memory is faulty, and that eyewitness testimony is unreliable. Juries don’t. How will juries respond when attorneys can point out every flaw, every miscue in a witness narrative? This is not a question of whether our legal system should use technology – it does, and it should, to provide the best quality of counsel to our clients. Mr. Weber continued, “Deep QA [Watson’s programming to understand questions] won’t ever replace attorneys; after all, the essence of good lawyering is mature and sound reasoning, and there’s simply no way a machine can match the knowledge and ability to reason of a smart, well-educated and deeply experienced human being. But the technology can unquestionably extend our capabilities and help us perform better.”
The question for our society is how we value those things that Watson cannot compute and factor them into our resolution of disputes – dignity, respect, integrity, altrusim. Until we can add those to Watson’s programming, the humans in the system need to value and uphold the human values that so-often get lost in the law.
— Lawrence A. Husick, Esq.