Artificial Intelligence II: Watson, Tay and driverless cars
Last month I talked about computers that have become the world’s strongest players in checkers, chess and a game called Go. This month I would like to talk about some other computer programs that are reaching levels of human-like abilities in other domains.
A number of you may have seen the computer program Watson from IBM defeat two strong human players on Jeopardy! in 2011 to win a one-million dollar prize. Ken Jennings, one of the humans defeated by Watson, suggests that the program is taking the same approach that he does in answering these trivia puzzles, analyzing many possible questions suggested by the answer provided. Watson is now being used to help doctors to make lung cancer treatment decisions in an American hospital. If a computer program could read X-ray images better than a human doctor, would you trust it?
Tay is a “chatterbot” computer program released by Microsoft earlier this year to hold conversations on Twitter with 18- to 24-year olds. The program was designed to increase AI knowledge about natural conversation in a controlled format. It was shut down after only 16 hours because some human online users hijacked this corporate software, teaching it to make inappropriate sexist and racist responses. This incident suggests that it may be easier to write computer programs for games with specific rules like Go than for real-world, unconstrained conversations.
Some of the large computer companies are currently creating software and hardware that can drive cars without needing human drivers. These cars are already being tested on the road in a variety of limited formats. I think it is fair to say that in many contexts they are already better drivers than many humans; they don’t get distracted by non-driving tasks. If they could significantly reduce the number of road accidents, would we be pleased or upset about the accidents that remained? This technology is an advance on the autopilot function that has been used successfully in planes for years. The combination of some road rules software, short-range radar technology (already in use to warn about cars in our blind spot) and GPS software has the potential to revolutionize our driving experience.
Clearly, before widespread adoption of driverless cars, there are significant issues that need addressing. What happens if a dog wanders into the road, for example, and the choice is between swerving and damaging the car or hitting the dog? We have all read stories about people following GPS software blindly and driving into a lake or other strange places. What happens in these rare incidents? But all these incidents happen with human drivers. Should computer programs be compared to the many times I have gotten lost because I would not follow a human’s instructions?
In the image of
Computers already do many things better than we can. If I have a question or need some information, I “google” (a verb now) some key terms, and it is rare when the first or second search result does not lead me to the facts I need. If I need a journal article for my writing I go to PubMed (from the American National Library of Medicine) to find it rather than looking in my folders of reprints. If I go on Amazon to buy a book, the site can make suggestions based on my past purchases and current searches – a feature that has led me to some very interesting books that I am not sure I would have found otherwise. These tools available on my “smart” phone, along with many useful apps, make our lives easier but also more dependent on technology.
Naturally, in discussing artificial intelligence, what we mean by intelligence becomes an important and ongoing question, with no simple answer. One issue raised by artificial intelligence and smart apps for Christians is this: what does it mean to be human, to be made in the image and likeness of God? Would an intelligent computer have these characteristics? I will speak to this Imago Dei issue in my next column.