Nova’s documentary on IBM’s Jeopardy machine: Questions raised

February 10, 2011
237 Views

I enjoyed watching the Nova special, Smartest Machine on Earth, about IBM’s Jeopardy computer. It was interesting to me that they chose to call Watson …’he’…. I think this is a fairly important editorial decision. In the book, we went the other way, referring to it as a thing.

I enjoyed watching the Nova special, Smartest Machine on Earth, about IBM’s Jeopardy computer. It was interesting to me that they chose to call Watson …’he’…. I think this is a fairly important editorial decision. In the book, we went the other way, referring to it as a thing.

This morning, I went to the Web site and found scores of people’s questions for the Jeopardy team. Since the responses to …’selected’… questions won’t be posted until Feb. 22, I’ll take a crack at a couple of them here.

Q: Human contestants often buzz in prior to fully processing a clue or determining a preferred response. To what degree does Watson buzz in before selecting his response? (Or, if he does not buzz in during his computation process, why not?) Matt R, Seattle, WA

A: Humans have an idea of what they know and can anticipate coming up with an answer that hasn’t yet surfaced. Watson does not have this ability. It must find candidate answers, pick one out, and calculate its confidence in it before buzzing. This is a key difference between the humans and the machine.

Q: Why did you create Watson to be sight and hearing impaired, and is that decision fair?Matt R, Seattle, WA

A: Here I think the Matt has it backwards. The real question is this: Did you consider equiping the computer with vision and speech recognition? How much would that have enhanced its ability in Jeopardy, and how much work would it have required?

Initially, they considered giving Watson speech recognition. It added another layer of cognitive complications (and contributed to scaring research teams from the project). Removing speech recognition simplified the job. Electronic words are much simpler for a computer. Vision would have required a major research effort and, as far as I know, was never seriously considered.

Q: Watson answered …’Richard Nixon’…; to a clue asking for a first lady. I know Watson selects among answer candidates based on probabilities, but why would it ignore words in the clue (…’first lady’…) that explicitly ruled out Richard Nixon as an answer?Dan R., Portland, OR

A: Gender is more complicated that it may appear. Certain clues with the words …quot;First Lady…quot;…nbsp; may…nbsp; refer to that woman’s husband. Since Watson is never 100% certain it understands a clue, it can never definitively rule out a response.

Q: Just saying …’Ask not ‘Can Machines Think?’, but ask ‘What qualifies as thinking?’ Okay?
Douglas Pankretz, Olympia WA

A: When discussing a machine that behaves in certain ways like a human, it’s very hard to divorce it from human processes. This was a struggle for me in the book. As Douglas points out, it could be argued that Watson does not think, know, or remember. I tried to avoid attributing those words to it in the book. It does, by contrast, process, calculate, recall, and estimate. You can analyze Watson’s information processing and debate whether it actually thinks. I’ve gotten into those discussions. But I don’t dwell on them in the book. Its …’thinking’… if that’s what you want to call it, is very different than ours.