Tag Archives: Watson

Did Watson Have an Empathy Algorithm?

17 Feb

I’ve read the different explanations from IBM about why Watson acted so quirkily at the end of the first game when it held a commanding lead–$36,681 compared to Brad Rutter’s $5,400 and Ken Jennings $2,400. In the final Jeopardy round’s category “U.S. Cities” the computer answered “Toronto” when the correct response was “Chicago.”

IBM has been quick to offer reasons about its most famous computer’s personality. And they make some sense in an obfuscation as techie jargon kind of way.

What I think happened is much simpler: an IBM programmer introduced an empathy algorithm into the software. That is, if Watson knew it was pounding its opponents into the intellectual trivia dust, it would back off; it would refrain from humiliating its opponents. Think of it as a variant of Isaac Asimov’s famous first rule of robots: Do not harm humans. What could be more harmful to smart people than to make them look stupid in public?

Three things make my empathy algorithm theory very possible. First, Watson blew the very basic Final Jeopardy category “U.S. Cities.” IBM lamely says the computer’s answer, Toronto, makes some sense in that there are U.S. cities with the same name. Maybe, but do any of the Torontos in Florida, Illinois, Kansas, Missouri, Ohio, South Dakota, and Texas have an airport, let alone two of them? Any named for World War II history? Didn’t think so.

Second, the $947 bet. If, despite the empathy algorithm, in the random chance that Watson was going to get the Final Jeopardy response correct, it could not chance adding brutishly to its insurmountable lead. So, it bet small. Logic alone would dictate a bet of around $15,000 to assure a two-game match victory. But Watson did the gentlemanly thing instead and bet politely.

Finally, an item in the Fast Company story is intriguing. Apparently some programmer took it into his or her head to let Watson make “non-zero” bets for things like Daily Doubles and Final Jeopardy. He or she thought those arbitrary bets would spice up the show. As, indeed, they did.

With that kind of freedom granted to Watson’s developers, I’m willing to surmise that one programmer thought it was wise to make Watson a good chap as well as a great player and so introduced an empathy algorithm.

Advertisements

IBM Wins Even If Watson Loses

15 Feb

There’s a scene in the hilarious movie Groundhog Day when Bill Murray’s bewitched character, Phil, is sitting in the parlor of the antique- and lace-ridden B&B where he’s staying. He and other guests are viewing Jeopardy on television. Phil already has seen countless Groundhog Days by then and has watched this show many, many times. As Alex Trebek rattles off the answers, Phil tosses off the correct question, one after the other. He even gives a correct response before Trebek finishes his prompt.

By this point, an older woman is gazing at him, not in awe at his intelligence, but in horror. Phil is not just smart. He’s scary smart like some kind of dangerous machine.

IBM’s Watson was in danger last night of becoming just another scary machine on the popular TV game show. After finding the Daily Double on its first query choice, a rarity for anyone who has watched Jeopardy over the years, Watson, like Phil, went on to give one correct response after another. This might not have been so impressive against the likes of, say, me, but Watson was whipping the two greatest Jeopardy players in history.

Then Watson slipped up. It answered “chic” when it should have said “class” and that opened the door for the humans to stage a comeback.

Good thing for IBM Watson flubbed. Its failure made the 2,800-server computer more human, which made for better television. Instead of millions of people turning off their TV sets, they’ll all be back tonight to see if Watson can regain its dominance or remain imperfect, human-like.

Sure, IBM’s engineers and marketers who are staging this week’s event want Watson to win. But the worst thing would be for the computer to trounce its flesh-and-blood competitors. It would fuel people’s underlying fears about soulless machines relentlessly pushing people aside as they continue their march to dominate humans. What viewers want and what IBM needs is for a close match, one that goes down to the wire with the outcome unknown until the very end.

Then, let the best man or machine win. That would be fun. And even if Watson lost at the end, IBM would still emerge victorious because it would fuel much more interest in the company’s high-performance computing systems. And, ultimately, that’s what the 100 year old firm wants: attention brought to the company’s accomplishments and what those achievements can do for others.