|
On February 17 2011 06:29 SpoR wrote:
People get so defensive about machines/computers besting them. The machine isn't smarter, or more intelligent. It just brute forces a giant knowledge data base and does it really really fast. If humans had access to all that plus our own brains we would win uncontested every time.
Ultimately, that is the point.. For us to say look how much we can do with computers, we can apply these things in other areas and have a better world.
No, there is more at stake here than most people like to admit. Look into the concept of a mind and what the artificial intelligence people are saying, there's a pretty significant implication here with Watson.
Also that "If" statement carries a very big qualifier considering the fact that we don't even understand how our own minds work.
|
On February 16 2011 19:57 BlackJack wrote:Show nested quote +On February 16 2011 19:39 igotmyown wrote:
For matches against the machine, if a contestent buzzes in within some small time period (like .25 seconds) and the computer does as well, someone should be randomly selected to answer. Yeah these matches are kind of stupid. They didn't prove they could build a computer that can beat a human at jeopardy, all they've proven is that they've built a robot that can click a buzzer faster than a human..
But that's how you're supposed to beat a human at Jeopardy, because the human pretty much knows all the answers anyway. Jennings won all those games not because he knew more than his opponents, but because he revolutionized the way people would buzz in.
I think it's already impressive that Watson got so many answers correct. That he could buzz in before humans is pretty much a given.
Also, the way the system works is that after Alex finishes reading the question, a light goes on signaling the contestants can buzz in. Watson is simply reacting to that signal (which is fed into him because he doesn't have eyes). The humans, on the other hand, don't react to the light; what they do is anticipate by formulating a pattern between the end of the question and the light. Obviously, you'd rather have super-human reaction time than rely on anticipation, but they're only human.
|
This is actually the coolest shit... Sick
|
This works well because you can find relevancy between words. The computer is not doing any "thinking" but is correlating information in the answer to a word. It's similar to the 20 question bot online. I went to the AAAI conference 2 years ago and this is just a small application in the world of AI. Not really a huge advancement, more like a giant toy. When computers can beat humans at game not involving perfect or large amounts of information, then that's when AI has advanced.
|
On February 17 2011 09:37 darmousseh wrote:
This works well because you can find relevancy between words. The computer is not doing any "thinking" but is correlating information in the answer to a word. It's similar to the 20 question bot online. I went to the AAAI conference 2 years ago and this is just a small application in the world of AI. Not really a huge advancement, more like a giant toy. When computers can beat humans at game not involving perfect or large amounts of information, then that's when AI has advanced.
like what?
This project is not really about AI at all, its about computers interpreting natural language, and in that field its absolutely enormous.
|
I watched this just now (rerun?). Wow, is Watson insane.
|
Wow I just watched all 3 shows. I am really impressed by its performance. I'd really like to see the source materials it used for each answer.
I guess I'll go search for some of the papers released during the development of watson
edit: again used his instead of its, really need to stop humanizing it
|
On February 17 2011 12:46 sob3k wrote:Show nested quote +On February 17 2011 09:37 darmousseh wrote:
This works well because you can find relevancy between words. The computer is not doing any "thinking" but is correlating information in the answer to a word. It's similar to the 20 question bot online. I went to the AAAI conference 2 years ago and this is just a small application in the world of AI. Not really a huge advancement, more like a giant toy. When computers can beat humans at game not involving perfect or large amounts of information, then that's when AI has advanced. like what? This project is not really about AI at all, its about computers interpreting natural language, and in that field its absolutely enormous.
Interpretting natural language is something we would consider requiring usage of our intellect. So I would argue this is still quite clearly in the realm of AI.
|
|
|
Let me make myself more clear:
What we want to see is Watson THINK faster than human, not click faster.
If you can only click AFTER the host read the question, this scenario is quite likely:
Host start to read the question
Human1 realize the answer, but cannot click
Watson realize the answer, later, but cannot click
Host finish reading the question
Watson clicks, and he wins
Clearly, human SHOULD win in the above scenario because he came up with the answer faster but was not allowed to click, however, Watson will win every time in the above scenario.
Either way, either IBM cheating or there is an inherent flaw build into Jeoporty's rule system that changed the game from "comming up with answer fastest" to "clicking fastest"
IMO this whole tournament needs to be re-run since it is just a clicking war atm which doesn't really prove much. It only proved it is possible for the machine to find the answer in the time period of reading the question, it did not prove the machine could do so FASTER than human.
edit: if anyone is going to quote me to make a reply, please also send me a pm notifying you done so since I'd like to discuss this issue further if people are responding to it.
|
United States12244 Posts
On February 17 2011 07:05 NEOtheONE wrote:So Ken Jennings had an epic quote today in his final Jeopardy response "(I for one welcome our new computer overlords)." + Show Spoiler + Watson wins with a two day total of over $77,000
hahaha yesssss
+ Show Spoiler +
|
On February 17 2011 13:50 evanthebouncy! wrote:
Let me make myself more clear:
What we want to see is Watson THINK faster than human, not click faster.
If you can only click AFTER the host read the question, this scenario is quite likely:
Host start to read the question
Human1 realize the answer, but cannot click
Watson realize the answer, later, but cannot click
Host finish reading the question
Watson clicks, and he wins
Clearly, human SHOULD win in the above scenario because he came up with the answer faster but was not allowed to click, however, Watson will win every time in the above scenario.
Either way, either IBM cheating or there is an inherent flaw build into Jeoporty's rule system that changed the game from "comming up with answer fastest" to "clicking fastest"
IMO this whole tournament needs to be re-run since it is just a clicking war atm which doesn't really prove much. It only proved it is possible for the machine to find the answer in the time period of reading the question, it did not prove the machine could do so FASTER than human.
edit: if anyone is going to quote me to make a reply, please also send me a pm notifying you done so since I'd like to discuss this issue further if people are responding to it.
Yes, I'd like to see this computer be used in NAQT setting:
http://www.naqt.com/
You can buzz in at any point while the question is being read. If Watson still wins at that, then it's legit.
|
Jeopardy between just human contestants is for the most part a clicking contest.
|
On February 17 2011 13:50 evanthebouncy! wrote:
Let me make myself more clear:
What we want to see is Watson THINK faster than human, not click faster.
If you can only click AFTER the host read the question, this scenario is quite likely:
Host start to read the question
Human1 realize the answer, but cannot click
Watson realize the answer, later, but cannot click
Host finish reading the question
Watson clicks, and he wins
Clearly, human SHOULD win in the above scenario because he came up with the answer faster but was not allowed to click, however, Watson will win every time in the above scenario.
Either way, either IBM cheating or there is an inherent flaw build into Jeoporty's rule system that changed the game from "comming up with answer fastest" to "clicking fastest"
IMO this whole tournament needs to be re-run since it is just a clicking war atm which doesn't really prove much. It only proved it is possible for the machine to find the answer in the time period of reading the question, it did not prove the machine could do so FASTER than human.
edit: if anyone is going to quote me to make a reply, please also send me a pm notifying you done so since I'd like to discuss this issue further if people are responding to it.
Honestly people are blowing the whole game show part of this out of proportion. The fact that it can win more money than Ken Jennings is completely beside the point of the demonstration, which is to show that we have the capabilities to create a computer which can interpret even the most convoluted natural language and respond appropriately in a reasonable timeframe.
Whether or not it wins or is just somewhat competitive, or wins due to reaction time or fast thinking doesn't really matter past a really shallow publicity level.
On February 17 2011 14:15 DTK-m2 wrote:Show nested quote +On February 17 2011 13:50 evanthebouncy! wrote:
Let me make myself more clear:
What we want to see is Watson THINK faster than human, not click faster.
If you can only click AFTER the host read the question, this scenario is quite likely:
Host start to read the question
Human1 realize the answer, but cannot click
Watson realize the answer, later, but cannot click
Host finish reading the question
Watson clicks, and he wins
Clearly, human SHOULD win in the above scenario because he came up with the answer faster but was not allowed to click, however, Watson will win every time in the above scenario.
Either way, either IBM cheating or there is an inherent flaw build into Jeoporty's rule system that changed the game from "comming up with answer fastest" to "clicking fastest"
IMO this whole tournament needs to be re-run since it is just a clicking war atm which doesn't really prove much. It only proved it is possible for the machine to find the answer in the time period of reading the question, it did not prove the machine could do so FASTER than human.
edit: if anyone is going to quote me to make a reply, please also send me a pm notifying you done so since I'd like to discuss this issue further if people are responding to it. Yes, I'd like to see this computer be used in NAQT setting: http://www.naqt.com/You can buzz in at any point while the question is being read. If Watson still wins at that, then it's legit.
Thats taking the exercise to an entirely different level of predicting and filling in entirely missing parts of the query. Not that thats not probably a future goal, but its not what Watson was designed to do.
TLDR:
The point here is not to create a robot that can win quiz shows, its to create a machine that can respond correctly to natural language. The quiz show is just a fun demonstration.
|
On February 17 2011 14:15 DTK-m2 wrote:Show nested quote +On February 17 2011 13:50 evanthebouncy! wrote:
Let me make myself more clear:
What we want to see is Watson THINK faster than human, not click faster.
If you can only click AFTER the host read the question, this scenario is quite likely:
Host start to read the question
Human1 realize the answer, but cannot click
Watson realize the answer, later, but cannot click
Host finish reading the question
Watson clicks, and he wins
Clearly, human SHOULD win in the above scenario because he came up with the answer faster but was not allowed to click, however, Watson will win every time in the above scenario.
Either way, either IBM cheating or there is an inherent flaw build into Jeoporty's rule system that changed the game from "comming up with answer fastest" to "clicking fastest"
IMO this whole tournament needs to be re-run since it is just a clicking war atm which doesn't really prove much. It only proved it is possible for the machine to find the answer in the time period of reading the question, it did not prove the machine could do so FASTER than human.
edit: if anyone is going to quote me to make a reply, please also send me a pm notifying you done so since I'd like to discuss this issue further if people are responding to it. Yes, I'd like to see this computer be used in NAQT setting: http://www.naqt.com/You can buzz in at any point while the question is being read. If Watson still wins at that, then it's legit.
considering good NAQT questions uniquely define the answer in the first clause (by definition), then yes, this would quell all the criticisms.
|
On February 17 2011 14:27 sob3k wrote:
Thats taking the exercise to an entirely different level of predicting and filling in entirely missing parts of the query. Not that thats not probably a future goal, but its not what Watson was designed to do.
no it isnt. NAQT questions are "pyramid-style" -- the first clauses of the clue contain ridiculously obscure references to the answer but the clauses uniquely identify the answer -- that is, there is only one possible answer given the first clauses. from there, you get less and less obscure until the prime ("... for ten points, ....") where the most obvious clause in the clue is given.
so you can (and many good players often do) answer correctly from only a few words in the clue.
|
On February 17 2011 14:31 annul wrote:Show nested quote +On February 17 2011 14:27 sob3k wrote:
Thats taking the exercise to an entirely different level of predicting and filling in entirely missing parts of the query. Not that thats not probably a future goal, but its not what Watson was designed to do. no it isnt. NAQT questions are "pyramid-style" -- the first clauses of the clue contain ridiculously obscure references to the answer but the clauses uniquely identify the answer -- that is, there is only one possible answer given the first clauses. from there, you get less and less obscure until the prime ("... for ten points, ....") where the most obvious clause in the clue is given. so you can (and many good players often do) answer correctly from only a few words in the clue.
but if you can buzz in before the clauses are finished then it involves predicting missing information
|
On February 17 2011 14:27 sob3k wrote:Show nested quote +On February 17 2011 13:50 evanthebouncy! wrote:
Let me make myself more clear:
What we want to see is Watson THINK faster than human, not click faster.
If you can only click AFTER the host read the question, this scenario is quite likely:
Host start to read the question
Human1 realize the answer, but cannot click
Watson realize the answer, later, but cannot click
Host finish reading the question
Watson clicks, and he wins
Clearly, human SHOULD win in the above scenario because he came up with the answer faster but was not allowed to click, however, Watson will win every time in the above scenario.
Either way, either IBM cheating or there is an inherent flaw build into Jeoporty's rule system that changed the game from "comming up with answer fastest" to "clicking fastest"
IMO this whole tournament needs to be re-run since it is just a clicking war atm which doesn't really prove much. It only proved it is possible for the machine to find the answer in the time period of reading the question, it did not prove the machine could do so FASTER than human.
edit: if anyone is going to quote me to make a reply, please also send me a pm notifying you done so since I'd like to discuss this issue further if people are responding to it. Honestly people are blowing the whole game show part of this out of proportion. The fact that it can win more money than Ken Jennings is completely beside the point of the demonstration, which is to show that we have the capabilities to create a computer which can interpret even the most convoluted natural language and respond appropriately in a reasonable timeframe. Whether or not it wins or is just somewhat competitive, or wins due to reaction time or fast thinking doesn't really matter past a really shallow publicity level. Show nested quote +On February 17 2011 14:15 DTK-m2 wrote:On February 17 2011 13:50 evanthebouncy! wrote:
Let me make myself more clear:
What we want to see is Watson THINK faster than human, not click faster.
If you can only click AFTER the host read the question, this scenario is quite likely:
Host start to read the question
Human1 realize the answer, but cannot click
Watson realize the answer, later, but cannot click
Host finish reading the question
Watson clicks, and he wins
Clearly, human SHOULD win in the above scenario because he came up with the answer faster but was not allowed to click, however, Watson will win every time in the above scenario.
Either way, either IBM cheating or there is an inherent flaw build into Jeoporty's rule system that changed the game from "comming up with answer fastest" to "clicking fastest"
IMO this whole tournament needs to be re-run since it is just a clicking war atm which doesn't really prove much. It only proved it is possible for the machine to find the answer in the time period of reading the question, it did not prove the machine could do so FASTER than human.
edit: if anyone is going to quote me to make a reply, please also send me a pm notifying you done so since I'd like to discuss this issue further if people are responding to it. Yes, I'd like to see this computer be used in NAQT setting: http://www.naqt.com/You can buzz in at any point while the question is being read. If Watson still wins at that, then it's legit. Thats taking the exercise to an entirely different level of predicting and filling in entirely missing parts of the query. Not that thats not probably a future goal, but its not what Watson was designed to do. TLDR:
The point here is not to create a robot that can win quiz shows, its to create a machine that can respond correctly to natural language. The quiz show is just a fun demonstration.
First of all the queries are by no means the most convoluted form of language. The queries, ahem, questions, are basically facts inquiring more facts. This is probably most relevant to a google search.
Take some question that Watson got wrong for example:
1) toronto is in the U.S.
2) voldemort killed blah blah blah and it's hard to name him
These questions' responses clearly demonstrate Watson is not parsing the language as semantics but is still on the pattern matching level. So he's a far way from responding correctly to natural language.
So, Watson is doing "fairly good pattern matching in a reasonable time".
Second of all, IBM built Watson knowing Jeoparty is a good project, because it is just a little bit more than pattern matching, yet the amount of natural language complexity is still limited. So, by all means this is a great project, but it's not that much revolutionary. People in the NLP community have been able to do this for awhile now, the only constraint is they might need more time.
Third of all, IBM obviously had PR in mind when they built the machine, look at how much PR they are getting from this, it is a smart move.
Fourthly, what I am proposing is not to have watson infers half the question. Maybe I said it bad but you didn't get what I meant. What I meant is you still give everything at the start, but people are allowed to buzz in at any time instead of having a halt period where no-one could buzz until the end of reading the question and some light turn on.
Finally, I like to emphasize that the goal of this WATSON project is to show a machine can parse, and pattern match a given query fast. We've already seen that it is fast, but given the current format of this competition, there are much more to be desired, since so far it only shows WATSON can answer reasonably fast (before the host finish reading) and beats human at clicking every time.
footnote:
I'm a student in computer science and I have dealt with AI before and understand some natural language processing. So rest assured I'm not pulling stuff out of nowhere.
|
How long a question is seemed to effect Watson i noticed on shorter questions that Watson wasn't so dominating it guess it lacked time to process an answer.
|
On February 17 2011 14:38 sob3k wrote:Show nested quote +On February 17 2011 14:31 annul wrote:On February 17 2011 14:27 sob3k wrote:
Thats taking the exercise to an entirely different level of predicting and filling in entirely missing parts of the query. Not that thats not probably a future goal, but its not what Watson was designed to do. no it isnt. NAQT questions are "pyramid-style" -- the first clauses of the clue contain ridiculously obscure references to the answer but the clauses uniquely identify the answer -- that is, there is only one possible answer given the first clauses. from there, you get less and less obscure until the prime ("... for ten points, ....") where the most obvious clause in the clue is given. so you can (and many good players often do) answer correctly from only a few words in the clue. but if you can buzz in before the clauses are finished then it involves predicting missing information
You are still missing the point here. If you watched the first episode the host said:
When the question is displayed on the screen, an entire text file is sent to Watson containing the full information of that question.
So no, there will be no predicting.
|
|
|
|
|
|
EPT![[image loading]](https://lh3.googleusercontent.com/__ApcKUzLrio/TVyMBfxl9JI/AAAAAAAAB_8/VQmv42X7ArA/lol.jpg)
