From Watson's DD betting to sheer dominance most of the game to FJ guess, awesome all around

Is it just me or does Watson have a clear buzzing-in advantage? It seemed like both human competitors were trying to buzz in as soon as they could at the beginning of the episode for what were some (relatively) easy questions, yet Watson was called on like 15 straight times. Apologies is this has been covered.

One could argue that the major advantage Jennings and Rutter had over other human Jeopardy contestants during their runs was that they were superior/experienced with the buzzer.

I get the feeling from watching Jennings/Rutter over the last couple of days that they know just as many of the questions (or at least very close to) as Watson does. You can particularly tell with Ken, who at times visibly shows his frustration with not being first to buzz in.

So what has this competition really become if all three contestants can answer 85+% of the questions correctly, just a show of who can buzz in the quickest? Of course a machine designed specifically to buzz in immediately will have faster reflexes than even the best humans. That doesn't surprise me at all.

Why couldn't they build Watson to engage Alex in some witty banter halfway through the Jeopardy round? I would be a lot more impressed if it could have a conversation with Alex during the break. The developers even alluded to this during one of the background video segments, that creating a computer than both understands and speaks natural language is one of their ultimate goals. That, to me at least, would be a lot more impressive than just being able to buzz in faster than Ken/Brad, which I believe is what this challenge has become.

A_K,
Agreed, and well said IMO.

Watson doesn't have any voice recognition capabilities currently afaik. The questions come in as text files.

I think many people are overthinking the purpose behind all this. IBM sought out to create a machine that can understand natural language and answer questions based on Jeopardy format. What is really amazing is that they built a machine that understands (most) questions posed to it in natural language and is able to discern the appropriate answer rather than providing you with links to where a possible answer might be.

Since the format of the show is what it is, there is no way to level the playing field by adding a time buffer to when Watson is able to buzz in. The real neat part is that the machine 'knows' the answer at the same percentage as the humans (i.e about 85% of the time as was written above).

From my understanding of Jeopardy rules, you can only buzz in when Alex finishes reading the question....so that's sets the baseline and everyone starts at the same starting point. The machine also uses a buzzer in the same way that the humans do. The humans have advantage because they can read faster than Alex reads the question so there is no significant (most of the time) lag.

There was a question today where all three contestants did not buzz in (the humans did not know or were unsure, and Watson's confidence was not a the appropriate level). That in itself was very interesting.

If you did not know that Watson is a machine and did not see the show (rather just listened to it), Watson would pass the Turing test: http://en.wikipedia.org/wiki/Turing_test

That's the accomplishment here.

Totally agree with this.

It's the next step in google. Instead of giving you thousands of documents with varying degrees of relevance to your query it goes through the documents to put an answer to you. It's a huge step.

Disagree.

This isn't a conversation.

Conversation or no, the point is that you would need to be able to decide that it's definitely not a person. The 1920s question was an epic fail, as no human who got any right at all would do that.

By the standards in the post above yours, I passed the Turing Test some years ago with a chat bot that talked to a programmer (so I didn't have to) for 8 months spitting back ~100 phrases partially at random with some delays and "BRB"s thrown in there for good measure.

Its far away from passing the Turing test. Technically even the replicant in Blade Runner couldn't do that.

That said, Watson is an immense and marketable application.

I imagine that for example, human call centers/tech support to be rendered fairly obsolete pretty soon, search engines drastically improved. And thats just the tip of the iceberg.

I can tell you that there's a mint to be made for IBM in just adapting this thing for legal precedent searches like Quicklaw and Westlaw (o/t: although I'm still hoping both those will be replaced in due course by an expanded Google scholar).

Anybody watching IBM stock this week? I don't follow markets or even know how really, but I'd be curious if IBM is ticking up as a result of the showcase.

i could've programmed something to filter for only US cities when asked a question about airports in US cities. its amazing that he rarely gets answers wrong in a "good try" manner. he is unbelievably bad when he's bad. imagine his wrong diagnosis of a disease. you'll come in with a questionable tumor and hell tell you you have cable tv

the thing about US cities or decades is that it isn't programmed to tackle those topics, or any topics in a particular way.

No one sat there and said "decades, ok that's these", "presidents, that's these guys", "cities, the list is here", .... . Instead that just bulldozed tons of into into the thing and programmed all purpose algorithms for question answering.

Really that makes the machine more impressive as AI (though less effective as Jeopardy contestant).

Yeah, its difficult to handicap it though.

I think the best solution to the buzz-in problem is this:

Give each player (or Watson) a button they can press if they wish to ring in, including while Alex is reading. Then choose one person who has rung in for that question at random to answer.

Seems like a good thing to me. If you have this thing as an ER assistant you'd rather have him tell you to try cable tv than to name the wrong treatment procedure.

Buzzing advantage is making the show less fun than it should be.