Mad Dog 21/21: Talking Toklas
February 28, 2011 Hesh Wiener
Here is how the game show Jeopardy works. The presenter says that the answer is “9W.” The contestant must come up with the right question, “Do you spell your name with a V, Herr Wagner?” Here is another example: The answer is Linux. A winning response is, “What’s in the server?” or “What runs on the Watson server?” Jeopardy fans loved watching the computer win as it prepared to take away their jobs. And astute Power box users who bet their careers on IBM i or AIX formulated some special questions, such as “How much would we save if we ran Linux instead?”
For computer users wedded to proprietary software, “Why not Linux?” isn’t an idle question. IBM’s PR gang, which missed the point, was still aglow the week after the Jeopardy hack when editor-in-chief TPM raised the issue in an essay. Basically, he figured out that the Jeopardy job ported to Amazon‘s EC2 compute cloud, where Linux thrives, might cost only $1,152 per hour. Even if you used the system around the clock, that grand or so an hour for, say, a few weeks, is peanuts compared to the value of the TV ratings a show like Jeopardy can get. It’s also peanuts compared to something you have probably used (but not used exclusively) recently: a commercial aircraft.
A Boeing 777 might cost $600,000 to $900,000 per month on a dry lease, or $800 to $1200 an hour, if the Web postings I found are more or less accurate. In case you don’t lease aircraft regularly, I will remind you that a dry lease just gets you an aircraft. If you want the full monty, crew and all, insured and ready to fly, you need a wet lease for which you have to be prepared to pay five to 10 times the dry lease rate. That’s a lot more than Watson-in-the-cloud will cost . . . unless you go wild hiring geeks.
With or without a nice team of engineers, if a Watson class system can actually help a business, it might not be all that expensive. If it can help a business regularly and ends up with a full time job, the server capacity might cost no more than, say, one of the Bloomberg Falcons in the media company’s fleet of jetlets.
Okay, the cost of a Watson, or at least the computing power of a Watson, is within the reach of many organizations; it’s not just something special only IBM or another hundred-billion-a-year outfit can afford. But that still doesn’t really explain why IBM didn’t build its showcase on its commercial i software or its Unix variant. Big Blue’s choice of Linux is a powerful signal to any other company that wants to emulate the Watson concept of artificial intelligence (if that is what it actually is) that a Watson class system is within reach, technically and financially, and it can be built on just about any kind of equipment. Maybe some other company will build a Watson and ask it why IBM was happy to show off using Linux rather than i, AIX or z/OS. Maybe IBM should get Watson to critique its next dog and pony show before the crowd is in the tent.
Well, IBM isn’t alone. This seems to be the season for close-but-no-cigar stunt results. Just the other day I saw a formerly well-entrenched strongman botch an attempt to save his forty-year-old regime. Muammar Gaddafi went on TV hoping to persuade disgruntled Libyans that he would sweeten the conditions in his police state. In Libya, this was bigger than Jeopardy, which in any case was probably jammed at the time. For the occasion, Gaddafi dressed up like sweet Mary Poppins and carried an umbrella just to make sure his subjects got the entertaining reference. But then, like IBM going with Linux instead of its own software on Watson, he slipped. Goofy Gaddafi insulted and threatened his audience when he should have been handing out falafel or Yemeni qat and singing, A Spoonful of Sugar Helps the Medicine Go Down. Gaddafi could not have flubbed the Mary Poppins bit more if he talked about Chimneysweep’s Disease.
Weizenbaum grew to regret Eliza–or more accurately, the way people reacted to it. He was surprised his software toy fooled people, but it did and even if you know it’s silly you can play with it and it will fool you, too. The artificial intelligence in Eliza might have been weak, but the real intelligence of Weizenbaum was extraordinary. Still, brilliant as he was, it took him a whole book to try and debunk Eliza. Weizenbaum is dead. The book is out of print. Eliza, as I have just shown, lives on.
Eliza the software was named after Eliza Doolittle, a lower-class flower seller transformed into a middle or maybe upper class lady by Professor Henry Higgins through language training. This is the plot of George Bernard Shaw’s play Pygmalion. The original play had a bittersweet ending, but eventually it became a film and in that version Doolittle and Higgins end up happily ever after. The more cheerful alternative ending persisted as the play was turned into the musical My Fair Lady. Weizenbaum might well have preferred Shaw’s original ending. But whatever the scientist’s preferences in fictional plots, he never lost his concern about the ease with which his software got people to spill their heart out to a computer.
I wonder what Weizenbaum would have said about Watson and Jeopardy. Unlike Eliza, Watson’s quiz game program didn’t seem likely to inveigle trusting innocents into revealing their deepest feelings. It did, however, remind people just how entertaining chatterbots can be. Coupled with the kinds of apps that are popular on smartphones, chatterbots may change the way people interact with machines. If the children of Eliza catch on there will probably be some funny examples of people who are fooled by these things, and some tragic examples, too. It is the nature of the human condition to have more questions than answers, lots more.
The writer Gertrude Stein, whose life in Paris between the two great wars of the last century is a world in itself, correctly sensed that her end was at hand in 1946 as she was wheeled into surgery. With her was her longtime companion Alice B. Toklas. Just before going under anesthesia, Stein turned to Toklas and asked, “What is the answer?” Toklas had no answer; she was silent. So, after a pause, Stein spoke again. “In that case,” Stein said, “what is the question?”