People remember Alan Turing for many different reasons. He was a British mathematician who worked as a codebreaker at Bletchley Park during World War 2. He also went on to become one of the pioneers of computing along with Max Newman. In 1952, Alan Turing was convicted of homosexuality. He accepted treatment with female hormones (or chemical castration) rather than go to prison and 2 years later committed suicide. In short, he was a genius who became a victim of his time. Were he born today, no-one would bat an eyelid at his homosexuality.
One of the his legacies is the Turing test. A machine could be said to be intelligent if it was indistinguishable from a human in conversation. He suggested that it would be better to come up with a learning machine (like a child’s mind) that could be taught and not something that simulates an adult mind. There is some debate as to whether a machine has ever really passed this test. I can think of a few humans who would struggle too.
Whenever I’ve seen any proffered example of artificial intelligence, I could not help but be disappointed. Today, however, I attended a very interesting session on IBM’s Watson semantic supercomputer. As supercomputers go, its $3m cost and 4TB of storage are pretty modest. Built on standard hardware and software, it resembles Turing’s child-like mind that can be taught. Indeed, before it can answer any sensible questions in a particular domain, it needs to be fed with information. Lots of it. Even that’s not enough, Watson needs to do further research on everything it reads.
Once loaded up with information, Watson is ready for questions. The first step in being able to answer any question is to parse it into the component elements. From this, Watson can determine the type of question being asked and what sort of answer is expected. It then generates many different permutations of the question to give the greatest chance of coming up with the right answer. For each interpretation of the question, Watson searches (in parallel) its knowledge base to see what answers it can find, coming up with as many as possible.
For each of the possible answers, Watson goes on the hunt for evidence, both for and against. Based on this evidence, obviously incorrect answers are discarded and the remainder scored based on the quality and reliability of the evidence. Finally, Watson uses the experience it as gained in the past in answering similar questions to gauge the value of the different types of evidence and comes up with a final confidence rating for each answer and ranks them accordingly. The process is very similar to that used by Doctor Gregory House in the hit TV program. You write-up all the possible answers and cross them out as the evidence goes against them.
The first outing for Watson was to win the TV quiz game Jeopardy against two former champions. IBM admit that this was little more than a bit of fun and a publicity exercise. Watson is now being geared up for much more serious applications such as medical research and diagnosis support. The applications for such technology are legion and the team freely admit that there are far more valid use cases than they have the time to exploit right now.
For the first time in my life, I am genuinely inspired by an example of artificial intelligence and I will follow Watson’s progress with interest.
- Director GCHQ makes speech in tribute to Alan Turing (isykes.wordpress.com)
- Alan Turing’s cyber-legacy praised by GCHQ chief (unemploymentsucksblog.wordpress.com)
- Artificially intelligent game bots pass the Turing test on Turing’s centenary (sciencedaily.com)
- New Edition Of Monopoly Honors Gay Computer Genius Alan Turing (joemygod.blogspot.com)
- Supercomputer Genius Watson Is Headed for the Cloud [Supercomputers] (gizmodo.com)
- GCHQ chief expresses regret at treatment of Alan Turing (guardian.co.uk)