Is this IBM's successor to Deep Blue?

[Father]

...

" Kasparov was busy throwing a tantrum, and might have mentioned anything he could to create suspicion. He should have just stood up, shook their hand, and said "I won't let that happen again." and he may well have won the third match, had that happened. But he didn't, he accused them of cheating, and any chance for a third and deciding match evaporated on the spot.

None of us that were active at the time had any doubt about the strength of the machine. Many of us played them multiple games, and watched them win almost every computer chess event from 1987 on, not to mention their performance against a wide range of strong human players in all sorts of tournaments and matches. The machine was a beast. Some today might well be as good or better. But in 1996 or so there was _nothing_ even remotely close to DB, in terms of other programs or hardware. Absolutely nothing."

I am absolutelly sure that Deep Blue sumarised a hard and wonderfull work of many smart people. But If Deep Blue had would have an active time playing after the match, probably, human beings could understood more about its real power. I think too, that Mister Kasparov lost in a pure an simple "Man Versus Machine battle". In my opinion the machine played all the time alone by itself. Simply he lost...

[bob]

...

" Kasparov was busy throwing a tantrum, and might have mentioned anything he could to create suspicion. He should have just stood up, shook their hand, and said "I won't let that happen again." and he may well have won the third match, had that happened. But he didn't, he accused them of cheating, and any chance for a third and deciding match evaporated on the spot.

None of us that were active at the time had any doubt about the strength of the machine. Many of us played them multiple games, and watched them win almost every computer chess event from 1987 on, not to mention their performance against a wide range of strong human players in all sorts of tournaments and matches. The machine was a beast. Some today might well be as good or better. But in 1996 or so there was _nothing_ even remotely close to DB, in terms of other programs or hardware. Absolutely nothing."

I am absolutelly sure that Deep Blue sumarised a hard and wonderfull work of many smart people. But If Deep Blue had would have an active time playing after the match, probably, human beings could understood more about its real power. I think too, that Mister Kasparov lost in a pure an simple "Man Versus Machine battle". In my opinion the machine played all the time alone by itself. Simply he lost...

I agree. Had IBM left DB intact for a year or so so that it could have played in the next WCCC, most of the past issues would no longer be the issues that they are today. But the problem with our wishes, is that for IBM it would have been an +enormous+ risk. Had they lost such a match, the shine would have been off the project. As it is, we are still talking about the machine today, 12 years after the match took place.

[bob]

Here's another example: "Mary saw the puppy in the window, she wanted it." Most humans would be able to understand that Mary wants the puppy, but how could an artificial intelligence understand that the puppy is more desirable than a window?

My personal favourite: "She saw the mountain flying over the Alps".

That sentence obviously means that the girl was flying over the Alps, and she saw the mountain on the ground. However, if all you do is resolve the grammer correctly, then we have a flying mountain! Humans are doing a lot of intelligent work to get to the intended meaning - but they do it so quickly that they barely notice that they are doing it. This might be an extreme example, but we're actually doing this kind of intelligent interpretation all day.

I'm not sure all of this is that hard. I've seen grammar parsing programs many years ago that could correctly parse the above to find the real subject, verb and direct object.

[mhull]

Here's another example: "Mary saw the puppy in the window, she wanted it." Most humans would be able to understand that Mary wants the puppy, but how could an artificial intelligence understand that the puppy is more desirable than a window?

My personal favourite: "She saw the mountain flying over the Alps".

That sentence obviously means that the girl was flying over the Alps, and she saw the mountain on the ground. However, if all you do is resolve the grammer correctly, then we have a flying mountain! Humans are doing a lot of intelligent work to get to the intended meaning - but they do it so quickly that they barely notice that they are doing it. This might be an extreme example, but we're actually doing this kind of intelligent interpretation all day.

I'm not sure all of this is that hard. I've seen grammar parsing programs many years ago that could correctly parse the above to find the real subject, verb and direct object.

But there are ambiguities that can only be solved by thinking in context. Knowing how far back relevant context extends is another intuitive judgment that would seem to defy programatic definition. Not to mention allusions, onomatopoeia, word-play (puns) -- simply intractable. AI just isn't, and won't be for eons to come. IMHO.

[bob]

I could write a program that will recognize the given sentence and return the answer (from the Encyclopedia database) in less than one millisecond.

The 2 main tricks here, from the top of my head, are:

* store all the data in RAM rather than on disc (some implementations of SQL have an option to do this, or you can write your own data storage model)

* think very carefully about the data indexing scheme, so that the smallest possible number of records have to be looked at

The best way to accomplish this is very close to your suggestion.

1. You use an in-memory database, and specifically a column store.
2. You create what is called an inverted index and also a bitmap index.

The lookups will be like lightning.

Here's the real question. Will this be done using traditional Jeopardy questions, or will they specifically write questions that a computer is going to have trouble with.

For example, in one Turing test experiment I took part in in 1984, the "tester" gave CB a position that was a very complex mate. It found it in a fraction of a second. I don't believe Turing envisioned that a computer could be "too good" and that that would/could be used to claim it failed the test.

If the questions are worded properly, the computer will never have a chance in a jeopardy contest. If they are worded as they usually are, the occasionally difficult to parse question will be screwed up by the computer, but it will still probably win, given enough preparation to work on the "in context" searching.