Computer crushes human 'Jeopardy!' champs (Update)

Under these conditions-competitive with humanity- does a computation equal a thought?
If the comp could see ,hear, taste, feel,smell, would the validity of its prowess be more, humm...'acceptable?
In a future, near or far, we give comps the ability to sleep/dream/feel-and have emotions- to hunger and thirst and make their own offspring: I can see; it would be MUCH easier to defeat them after undergoing a socialization! Watson is a brilliant tool BUT if he were 'smart' he would have NOT beat the humans to every buzzer so that no one would site his electronic or mechanical speed as an advantage...Watson needs to, in fact, 'grow up' to really win!

he would have NOT beat the humans to every buzzer so that no one would site his electronic or mechanical speed as an advantage.

That Watson comes up with the answer faster than they do is part of the point. He also has to physically buzz in the same way the players do.

This really is quite exciting! If any of you have ever read the Heechee series, it's one step closer to the Albert Einstein program :D. Ask a question and he shall answer! This would be such a great advantage for learning in the (hopefully near) future when a desktop computers power is great enough to process this kind of information.

Jaeherys: I would guess it will be about twenty years before a "desktop" or watch computer would be able to match this... Teachers, maybe you need to look for another occupation by then. why not let computers teach children? Yes, that brings up a few moral issues, I am sure.

Teachers, maybe you need to look for another occupation by then. why not let computers teach children?

Eh, why? Teaching isn't (or shouldn't be) just about regurgitating facts, otherwise you might as well read an encyclopedia.

This thing is just an advanced indexing tool. Of course it's faster than a human.

Not sure why this is supposed to be exciting... I eagerly await the day when a computer actually UNDERSTANDS the answer it is giving.

Not sure why this is supposed to be exciting... I eagerly await the day when a computer actually UNDERSTANDS the answer it is giving.

It's exciting because Watson needs to understand the question (which is in natural language) before it can answer.

Computers have always been good at regurgitating data, provided it is input in a structured form. Watson is the first, really good algorithm, for parsing natural language.

This was the most exciting episode of Jeopardy I've ever watched. I was giggling like a little girl every time Watson answered.

So what if it's an advanced indexing system? That's pretty much what our brains are. You have stored data which you need to access. If someone asks you who plays a certain song, you access your data based on the input query.

It made me think of the computer on the Enterprise. It will be awesome to ask a computer a question in the future and have a single answer returned instead of 10 million pages to sift through.

It's exciting because Watson needs to understand the question (which is in natural language) before it can answer.

Computers have always been good at regurgitating data, provided it is input in a structured form. Watson is the first, really good algorithm, for parsing natural language.

But it doesn't really "understand" it or even parse it. To my understanding, it works a bit like Google's translator - a big database of keywords and phrases linked up with scores of possible answers. It's a statistical analysis machine that is "smart" because it basically knows the questions and answers beforehand, even though they may not necessarily be written down in exactly the same way in its memory. It pieces together the question out of pieces that seem to match the input.

That's why it answered "Toronto" when the topic was US cities, and why Google sometimes gets simple things wrong like the difference between "is" and "is not"

I understand that Watson isn't perfect, but Toronto? If the category was US Cities, shouldn't it have automatically narrowed down the choices and took a more logical guess before the "answer" was even given?

I guess it's a good thing because IBM programmers have more work to do. This is a great start.

So what if it's an advanced indexing system? That's pretty much what our brains are. You have stored data which you need to access. If someone asks you who plays a certain song, you access your data based on the input query.

The difference is that your brain indexes the data based on what it means. If you ask it the question, "There's a white castle where a big man lives, controls the united tribes of the great land of the eagle.", Watson would probably be completely clueless because it has no keywords to match that exact phrase with "Washington D.C." unless programmed to do so. But then, it may just as well trip over "white castle" and "big man", and answer "What are hamburgers?" - a mistake that no thinking person would do.

It has syntax, but no semantics.

I understand that Watson isn't perfect, but Toronto? If the category was US Cities, shouldn't it have automatically narrowed down the choices and took a more logical guess before the "answer" was even given?

I suppose that's because there are several Torontos in the United States.

The exact question is an indicator to what went wrong.

"Its largest airport is named for a WWII hero. Its second largest, for a WWII battle"

"Toronto is the 2nd largest city in Jefferson County, Ohio, United States" - Wikipedia.

Then there's the Goodyear Aircraft Corporation which was based in Ohio, and many war heroes came from Ohio. Doing a related keywords search on WW2 hero, WW2 battle, airport, "second largest", US cities etc. easily produces a stronger signal on Toronto, Ohio, because the machine doesn't understand what part of the question is essential and what should be taken literally.

It has syntax, but no semantics.

Yet that seems enough to beat the human champions in this particular game (which has always been associated with contestants of above average intelligence). And pretty decisively at that.

That a computer would make different mistakes than humans (i.e. some that seem obvious to humans) isn't particularly startling. Conversely, if the humans had given more answers, we would certainly have seen things that are 'obvious' to Watson but aren't obvious to humans. So this isn't really an argument for human superiority.

Pretty exciting stuff.

I have to say that this is truly mind boggling. Since the AI field started in the 50'th we've been waiting for this! There was the expert systems and the whole neural network thing that initially looked promising. Sadly progress has been meager at best... until now. What a leap!

I have to say that this is truly mind boggling. Since the AI field started in the 50'th we've been waiting for this! There was the expert systems and the whole neural network thing that initially looked promising. Sadly progress has been meager at best... until now. What a leap!

I'm with you 100% in your enthusiasm and the significance of Watson's achievement, but I don't think this is a major step in AGI. It's an achievement in natural language processing, information indexing and relevance evaluation, but not general intelligence.

Yet that seems enough to beat the human champions in this particular game (which has always been associated with contestants of above average intelligence). And pretty decisively at that.

But that isn't really amazing. Throw in enough data and processors, and it was bound to do so by sheer brute force.

That's like pitching Usain Bolt against a motorcycle on a track. It isn't a running competition anymore.

Still: "natural language processing, information indexing and relevance evaluation" are pretty central to making 'intelligent choices'. and Watson seems to do perform this task extremely well.

What is still missing is the creativity aspect of intelligence by combining seemingly unrelated issues and noticing their common ground and the more elusive aspect of 'raw' creativity.

..and, of course, self awareness.

But as with all scientific endeavours: it's a process. Start with the (seemingly) easy things of parsing language and work from there on up.

But that isn't really amazing.

It is if you have ever worked in language processing. Searching for data is not the problem. Deciding WHAT you are searching for is tricky. A large database will not help you with that.

That is the main problem in the "only results matter" sort of take on AI, all the way down to the Turing Test which states that if a person can't distinguish a computer communicating from a person communicating, then the computer is "Intelligent".

But the computer doesn't have to be intelligent to pass off as a person. It only needs someone to program it with a huge number of logical permutations of conversations so that it would simply overwhelm the ability of the human to poke holes in its programming.

And that's the way these machines are built, because it's easier to just deliver the result than replicate the process. If you just want to beat Usain Bolt in a race, you don't build a running robot, you launch a friggin rocket and that's that.

It is if you have ever worked in language processing. Searching for data is not the problem. Deciding WHAT you are searching for is tricky. A large database will not help you with that.

Seems to work just fine for Google Translator. They operate on a huge dataset of written language where the translation is known, and then match pieces of phrases and individual words with the best match in the database. In the best case, they already have the entire sentence translated and all they have to do is query it.

It's a very very similiar problem.

But as with all scientific endeavours: it's a process. Start with the (seemingly) easy things of parsing language and work from there on up.

In my understanding, these machines aren't really trying to parse language, rather than simply trying to see if they already have a similiar string of data stored in memory.

To the machine, the input is "doo dah dah", and it finds it has "doo dah doh" stored in memory, so it determines that it's close enough and follows the link to the desired output result. Or perhaps it also has "doh dah dah" which is an equally good match, so it combines the results or tries to determine which of the results is more important based on some rules like "nouns before adjectives" etc.

Add a few more levels of complexity and some statistical analysis, and you have a Watson. It's not trying to find the relationships between "doo" and "dah" and "doh" because they're meaningless to the machine.

Still: "natural language processing, information indexing and relevance evaluation" are pretty central to making 'intelligent choices'. and Watson seems to do perform this task extremely well.

Yes, indeed.

But as with all scientific endeavours: it's a process. Start with the (seemingly) easy things of parsing language and work from there on up.

Yes, but I doubt that that is the intended purpose here, ie to design a general intelligence. A different project is much more likely to achieve that then Watson, namely, the Blue Brain project.

It would be interesting to see what Watson would think of a sentence where they verb nouns. I suspect it would throw it completely off kilter as it tries to search for items related to the nouns from its database without realizing that the real question is completely unrelated.

the Blue Brain project.

Blue Brain - despite its name - is not a project aimed at AI. It only models a cortex (e.g. it is used to figure out how lesions or chemical imbalances will affect the reaction of neurons). It's a neural simulator, not an evolving network. Even so, it's an awesome project.

To the machine, the input is "doo dah dah", and it finds it has "doo dah doh" stored in memory

I doubt it's that simple. Look at the narcolepsy question. There are no obvious keywords (besides 'disorder') in there.

I would agree that to us 'easier' questions are harder for Watson because the generality of the clues would make it harder for it to narrow down the search. Though then even a human would need more clues. There are plenty of sites where within 20 questions the database will figure out what you are thinking of with high accuracy.

Blue Brain - despite its name - is not a project aimed at AI.

Yes, I understand that, but I think it's eminently placed to spawn spin-off projects which do focus on AGI.

It only models a cortex

Only? That's a pretty darn big chunk of what makes us smart.

In fact, it's much more than what most people might think of as a 'simulation'. It reproduces all the physical functionality of neurons and how they interconnect, down to the minutest detail. These aren't simple abstractions that have been used in traditional neural network AI research.

If you can accurately simulate the brain's electro-chemical function enough to be clinically useful (which is the aim of Blue Brain), then you really aren't that far away from AI application.

It is if you have ever worked in language processing. Searching for data is not the problem. Deciding WHAT you are searching for is tricky. A large database will not help you with that.

Seems to work just fine for Google Translator. They operate on a huge dataset of written language where the translation is known, and then match pieces of phrases and individual words with the best match in the database. In the best case, they already have the entire sentence translated and all they have to do is query it.

I don't know how Google Translator works but it is (very) good only for single words. I'm working massively and daily with Google Translator for three languages and a friend is freelance translator and interpreter for Chinese and German. We agree on the superb usability of Google translator for single words. And we agree that the results for longer texts are getting worse with every additional word.

As they said in the video during the show, in the future, Watson will not but used alone. It could help give a fast and reliable 2nd opinion since usually, if Watson is wrong, it will be deadly wrong and plainly so. If some human had doubt on an answer, they could simply ask "Watson" and it could give its opinion too.

So in a couple years, I can have Watson on a walkie talkie button on my cell phone. Ask any question and get an answer. Or have my kid be able to ask her teddy bear a question and get an answer. Ask by voice, get a verbal reply. Is Watson intelligent? Perhaps. Is Watson a person? No. But program a bit of humor into Watson and we will probably regard 'him' as a friend. I can see people becoming emotionally attached to such programs. Inevitably, we will be assimilated, probably for the betterment of humanity.

We agree on the superb usability of Google translator for single words.

When I was working as a project manager in a translation company I found the opposite. Automated single word translations tend to be awful, since the translator lacks context (human translators are pretty bad under those circumstances, too, but they can improve if you give them even minimal context). It's hard to provide an automated translation system with context.

If you can accurately simulate the brain's electro-chemical function enough to be clinically useful (which is the aim of Blue Brain), then you really aren't that far away from AI application.

Intelligence requires making and breaking of connections. If you don't model that (which Blue Brain doesn't) then the best you can get is information retrieval and stimulus response. And even that is tricky since we can't yet interpret what a certain mindstate means (Blue Brain has no real sensors/actors).

Is Watson intelligent? Perhaps

No, it isn't. You could say it's intelligent within a restricted domain of expertise, but venture outside it, and it will fall in a heap. Another words, it doesn't exhibit general intelligence.

But program a bit of humor into Watson and we will probably regard 'him' as a friend.

No, it will take much more than that.

I can see people becoming emotionally attached to such programs.

Sure. People become emotionally attached to all kinds of things, even pet rocks. :)

Inevitably, we will be assimilated, probably for the betterment of humanity.

I don't think I'd fancy being assimilated (unless it was by 7 of 9).

Intelligence requires making and breaking of connections. If you don't model that (which Blue Brain doesn't)...

It doesn't, but it could, or rather, a different project that's based on similar principles (I know BB is funded purely for medical applications).

Blue Brain has no real sensors/actors.

Again, a different project could have these. But granted, that would add another level of complexity into the mix, but I think a true AGI will need at least audio and visual inputs, as well as autonomous mobility, otherwise it will probably go crazy, being trapped inside a blue box!

We agree on the superb usability of Google translator for single words.

When I was working as a project manager in a translation company I found the opposite. Automated single word translations tend to be awful, since the translator lacks context (human translators are pretty bad under those circumstances, too, but they can improve if you give them even minimal context). It's hard to provide an automated translation system with context.

You are right, but I was not speaking of automated translation.
Rather Google Translator is an excellent tool for the human (semi-/professional) translator who needs a quick list of target words in order to optimize the phrasing in the target language. It's more like using a synonym dictionary for the target language.

but I think a true AGI will need at least audio and visual inputs, as well as autonomous mobility, otherwise it will probably go crazy, being trapped inside a blue box!

I think for an AI you don't need all the specifics of Blue Brain. You need abstractions thereof. E.g. Blue Brain models a lot of stuff with ion channels, hormones and whatnot which aren't really relevant to the AI subject. In many respects it is overpowered (and in many more it is underpowered) when comparing it to an AI model. I'd say the two approaches share little with one another.

Biological brains (and Blue Brain) exist within a 3D setup. In this setup access to nutrients and possibly reachable neighboring neurons is restricted by the 3D topology.

In an nD setup (as could be easily modeled in a computer for AI models) those restrictions need not apply.

A lot of non technical people are on here making specualation seem like fact...

natural language processing is hard. no question. Is watson something to get excited about- YES. IT was asked questions in various abstract fashions and still arrived at the correct answer many times. The fact it wasn't perfect means nothing actually. It performed better than a human doing the same task -- that is one of the definitions of AI.

Google translate - this massive beast of a program uses A LOT of human processing power. Google created games and put them in the public space where a human matched a word to a picture -- this was done across a few languages and relationships were created based on confidence levels -- how many people agreed with what this person responded with.

this is all very abstract and challenging to program -- but it very closely mimicks how we percieve words - whether written or spoken.

No computer on the planet except Watson has shown capacity to do this.

in CS there are two general definitions of AI

1) The ability of a machine - machine meaning hardware or software or both - to perform a task that previously was done exclusivly by a human and performing better than that human.

2) a machine - same definition as above - that can pass a turing test. the test is if a human were to ask it a set of questions - by the end the human would not be able to determine that what gave the answers was a machine.

Google translate: i was not implying that humans are translating things on the fly but just introducing that Google stored a lot of data using very abstract ways to create translation software -- and picture identification software that many people are not aware of.

defining intelligence is very difficult ---

are humans intelligent -- well only compared to the other animals on the planet -- and then only in context - we have all seen videos of humans doing unintelligent actions.

is a dog intelligent -- is a cat -- is a pig ???

All three of those animals recognize themsleves in a mirror. And they all can do rudimentary counting. - they know if they are missing an offspring from a litter and will go look for it. which means the concept of less and more is understood by the animal. does this mean that they can do arithmetic - NO - but it shows that animals can attain a level of abstract thought.

is Watson intelligent -- well it all depends on context -- can it problem solve - appearently
-- can it do basic and complex math, well theres an app for that.

will it understand what it is doing -- well do you understand what you are doing at all times and do you think through all actions?? -

no you do not

but watson does.

he would have NOT beat the humans to every buzzer so that no one would site his electronic or mechanical speed as an advantage.

That Watson comes up with the answer faster than they do is part of the point. He also has to physically buzz in the same way the players do.

Here, I wil quote from the article:
"Five-time "Jeopardy!" champion Jeffrey Spoeri sympathized with Jennings and Rutter, and said the computer's speed to the buzzer seemed like an unfair advantage."
You say Watson had to answer"...the same way.." I ask, did you see Watson's hand come down on the buzzer? If Watson were 'smart' that is contextually and socially he would have told his inventors, "you really should give me a hand with which to ring the buzzer." DO I blame the machine? NO! Is it a great tool? YES!

I understand that Watson isn't perfect, but Toronto? If the category was US Cities, shouldn't it have automatically narrowed down the choices and took a more logical guess before the "answer" was even given?

I guess it's a good thing because IBM programmers have more work to do. This is a great start.

It is too bad we can't let all these critics play a game or two of Jeopardy and then analyze the answers that they give.

Did any of you people ever see Jennings or Rutters play? They are both amazing Jeopardy players. The fact that Watson beat them is just incredible. Its sad that people can't recognize what a monumental accomplishment this is.

I echo the fact that Watson is a smart indexing and voice recognition device. As Alex is speaking its parsing his words and doing index search. And hello people - they are targetting doctors and healthcare as major application area - so no surprise it kicks butt in those categories. Similarly anything to do with literature. You will know when it gets outside of its database - because it fails - as in the airports vis world war II links.

Yes its cool - yes it will advance knowledge sharing - but its not as fantastic as people are being led to believe. Nice IBM job on PR.

Some of you are criticizing the (alleged more over) inner workings of Watson's brain, but what do you know about how your own brain does it. There seems to be a "human" way (no specifics on the method), and a "machine" way which is an algorithm (or a combination of more of them). I'd not rush in to conclusions. You never know what will become of all of this. Nevermind. I think it's a next milestone in this area, and boy we've waited for it a long time. We're still not there, but at least something is happening, and it's fun too.

This kind of indexing, while only peripherally related to general intelligence, is fantastic none the less. Within 5-10 years you will see systems like this replacing search engines, giving you not just links to pages in response to queries, but actual answers where applicable. And you'll be able to query by simply talking. Probably to your phone. Tell me that isn't awesome? :)

You can already see fast search ahead in Google and Bing instant search features. Phone voice interface very limited in feedback abilities - maybe OK for support management - most people I think hate speaking to computers on the telephone!

This is incredible. This is a birth of something new and is still in its infancy. I will admit this is not going to be the path to a robot human any time soon we are very far away from that both software and hardware capabilities. Hardware will definitely have to be figured out as we are really to that point of being capped on how much more we can push without making bigger machines.

Software side this is a truly major accomplishment and this is just a show of new abilities we are making computers capable of doing.

As far as being fair I think they did very good considering the mass differences of a machine to a person. I think what people need to be really impressed with is the fact it took so much hardware and resources to match just a very small aspect of what the human mind can do. So by no means are computers even close to becoming human-like robots.

Hmmm - this thing is being feed text messages of questions - I just entered the three cited here into Google and it too found all the same answers!!! (fast forwarding past Jeopardy links) So all you need is a ranking mechanism and ability to select... and a large black box with flashing lights and an internet connection ; -)

mdbiker>
I did exactly the same test and agree with you.

Watson seems a gimmick and is just a Google/Bing/Yahoo data centre without the consumer front end overhead then squashed into a smaller box.

Even the internet connection could be redundant with the page/site caches and if it was a really close contest then Google could search personal email for information too!

Who would have ever thought of Ken Jennings as the underdog?

I echo the fact that Watson is a smart indexing and voice recognition device. As Alex is speaking its parsing his words and doing index search.

No, that's not quite right. Watson cannot see nor hear. The questions are fed to it electronically, like text messaging.

Watson seems a gimmick and is just a Google/Bing/Yahoo data centre without the consumer front end overhead then squashed into a smaller box.

No, it's much more significant than that. Yes, it seems a little less 'cool' that Watson isn't responding to spoken words, but the feat is no less special than if, say, a deaf person read the questions rather than hearing them.

I'd love to see this tech applied to something like Wolfram Alpha.

The difference is that your brain indexes the data based on what it means....

.... a mistake that no thinking person would do....

It has syntax, but no semantics.

I disagree. They said that when asked a question, Watson runs simultaneous algorithms for puns, rhymes, geography, etc. When I play Jeopardy, I consider that maybe my "geography" answers is too literal and I misunderstood the question and then I try to determine the answer using puns or word play. You assume our understanding of the meaning of the question is just given. The "meaning" has to be weighed against different factors and then we decide whether it's more probable that it's a joke or if it is not. Sometimes we are wrong, and that demonstrates that we don't always understand the meaning. I've seen players completely miss the point and give a ridiculous answer that didn't make sense.

..Sometimes we are wrong, and that demonstrates that we don't always understand the meaning. I've seen players completely miss the point and give a ridiculous answer that didn't make sense.

the next step is a computer that understands human idiosyncracies. what would watson do with a question that makes no sense? would it try to rephrase the question so it makes sense? would the next evolution in AI make a joke and insult the erroneous human? Show me an AI that knows how the human mind works. if watson can't hold a conversation, then it really is nothing more than a glorified indexing machine.

This thing is just an advanced indexing tool. Of course it's faster than a human.

It's more than an indexing system. It also is able to understand the question (most of the time). That is a fantastic accomplishment. And the way Jeopardy asks the question (or gives the answer) often isn't simple.

Teaching isn't (or shouldn't be) just about regurgitating facts, otherwise you might as well read an encyclopedia.

True though ai's wouldn't get impatient with students and would always be fair. Eventually AI will replace humans as teachers especially for the core subjects (english, science and maths)

DamiensS: "No, it's much more significant than that."

Yep sure is. Just shows how some smoke and mirrors can fool the ignorant!

Its a monkey in a rocket ship pushing buttons. Its a great feat in computations but the real issue here is not comparring it to humans but to a single task humans do. Memory. If the computer to stored a million times more information like humans do and then be able to do what a human does in 10 million other applications AND doing this at the same time it would be the firt time a portion of human existance would be somehwhat copied. Like i said. Its a monkey pushing buttons. Ho Humm.

This thing is just an advanced indexing tool. Of course it's faster than a human.

Not sure why this is supposed to be exciting... I eagerly await the day when a computer actually UNDERSTANDS the answer it is giving.

WHat do you mean by "understanding?" if what you mean is what it sees, hears, smells then understanding means not more than electrical signals. Does this sentence sound familiar from a famous movie?

This thing is just an advanced indexing tool. Of course it's faster than a human.

Not sure why this is supposed to be exciting... I eagerly await the day when a computer actually UNDERSTANDS the answer it is giving.

WHat do you mean by "understanding?" if what you mean is what it sees, hears, smells then understanding means not more than electrical signals.

I'll give you some "electrical signals":

ksero pos den ksero tipota

You "understand" all the symbols.
Google Translator "understands" that it is Indonesian, but Google is wrong.
Some people, however, smile when having read these signals or symbols. Why?

"I eagerly await the day when a computer actually UNDERSTANDS the answer it is giving." - Foofa

Can you precisely tell us what you mean by "understand" and why the current program doesn't
already "understand".

"Its a monkey in a rocket ship pushing buttons." - Tardboy

Or is it a neural network combined with other higher level analysis.

The fact is you don't have a clue to how it works, and you are the one just throwing out random ideas, like a monkey pushing buttons.

"if watson can't hold a conversation, then it really is nothing more than a glorified indexing machine." - Tard of Tards

Ya, just like a dog or a cat.

Watson has show that it is as good if not better than the top Jeopardy players in the world. Whats to say that Watson couldn't do scientific research as well as the top scientists in the world. They should put Watson to work on problems like fusion or an AIDS vaccine.

"if watson can't hold a conversation, then it really is nothing more than a glorified indexing machine." - Tard of Tards

Ya, just like a dog or a cat.

or just like Vendicar_Decarian but with little or nothing to index!

A symbolic coup for non-symbolic AI.

Watson has show that it is as good if not better than the top Jeopardy players in the world. Whats to say that Watson couldn't do scientific research as well as the top scientists in the world. They should put Watson to work on problems like fusion or an AIDS vaccine.

Watson is a comparitive search algorithim that operates based on the usage and context of words processed through a large set of rules and a massive bank of raw definition based data.

Watson cannot innovate, nor could he perform research. It is not comparable to human intellect in any manner, however, in the evolution of computing, Watson is impressive. We're approaching an understanding of the complexity fo the brain from the top down and the bottom up through research like Watson. The more complex a decision matrix becomes, the closer to intelligence it appears to be.

SH: You summed it very well.
I would also like to add compared to size the computer doesn't come close to the wonder of a human mind.
Maybe some day dear Watson, maybe some day!

SH: You summed it very well.
I would also like to add compared to size the computer doesn't come close to the wonder of a human mind.
Maybe some day dear Watson, maybe some day!

The NOVA PBS special on Watson was excellently descriptive. Showed the tuning process and structure of the arguments. It's beautifully simple and very similar to how one would teach a child.

For example: The question posed to Watson is "what fruit is also a color?" His answer "red apple". Which is obviously wrong to all of us, but that's because he didn't understand that "sometimes orange is a fruit and not a color". So they'd program that into the rule set and he'd start getting that question correct.

The performance on the Jeopardy test was incredible because of how Jeopardy questions are constructed. It was amazing to watch Watson reorg the rules in his system in order to grasp the correct answer for linguistically tricky queries.

But the computer doesn't have to be intelligent to pass off as a person. It only needs someone to program it with a huge number of logical permutations of conversations so that it would simply overwhelm the ability of the human to poke holes in its programming.

That maybe true in theory, given unlimited time for developers, but in reality it is impossible to write every possible "if" operator to any given situation. So, the only way to go is to actually write general purpose algorithms, which is pretty much AI.