It is very easy for our human interlocutor to answer a question like "Where is the pharmacy?" But for artificial intelligence it is extremely difficult. Elementary, Watson?
Israel Benjamin Galileo
"Of all the types of services on the Internet, this is the type of service we use the most frequently, even though it does not meet the real need for which we operate it." If you had to guess what the question was for this answer, you would probably answer "What is a search engine?"
This is the correct answer, because even though we use a search engine to find information, the service does not do it for us: it presents us with search results, leaving us the challenge of selecting the web pages that suit our purpose and locating the pages we need.
Because most of us are so used to this process, we forget that we would not agree to receive this kind of service from humans. Try to imagine a customer approaching a salesperson in a large store asking to find running shoes for a 12-year-old boy, and receiving in response a referral to several departments - to the children's clothing department, to the fitness equipment department, and to all the shoe departments.
The situation in the field is worse: a Google search for "running shoes for a 12-year-old boy" brought up the following results: a question about pain while running in a general orthopedics forum, a discussion in a recreational sports forum, a critique of the running shoe industry, and recommendations for TV shows. Indeed, most of us know that there are better ways to perform such a search, but why should we learn such methods, instead of expecting the computer to give us an exact answer as we demand from the seller in the store?
Don't search, find
In recent years, several websites have appeared that aim to answer questions in a reliable and relevant way. Most of the time these developments are successful only for a limited type of questions. For example, when ask.com asked "Who wrote 'The Catcher in the Rye'?" He immediately finds the answer.
On the other hand, when he was asked "When was 'The Catcher in the Rye' written?" He does not find an answer and displays normal search results, in which the sentences "'The Catcher in the Rye' is a book from 1951" and "'The Catcher in the Rye' was written in the XNUMXs".
Both sentences are correct - the first refers to the year of publication and the second to the period of writing - but they show that even in this simple challenge, one must deal with different possible meanings of the question.
Dealing with natural language (as opposed to computer languages) requires an understanding of meanings and meeting many challenges - knowing a large amount of information about the human world, value judgments, understanding metaphors and word games, and more.
Watson, without Holmes
The problems in understanding natural language are known and sometimes an ambitious goal definition with a clear success test is required to make significant progress. IBM did this when it built a special computer and developed software for it to compete with the world chess champion. With a similar ambition, IBM recently announced its intention to develop a computerized system that could compete against outstanding human players in the "King of Trivia" game.
Because most of us are so used to the search process, we forget that we wouldn't accept that kind of service from humans.
The project is called "Watson", after the founder of IBM, Thomas Watson. To prevent the computer from having an unfair advantage, it will not be connected to the Internet during the competition. In this way, the system is required to store large amounts of the information available on the Internet, and to organize the information so that it is available for quick analyzes and searches.
Bread and circuses
IBM has not yet announced when Watson will appear on the TV show, but project managers have already begun planning the event with the show's producers. To create interest, planners are considering selecting human competitors from among the greatest winners of the past.
More important questions for success in the competition concern the form of communication with the program director. It was decided that the questions would be transferred to the software in the form of text: the human competitors perceive the question by hearing the words and reading them on the program screens. For human competitors, these "achievements" are self-evident, but for artificial intelligence, turning visual and audio input into text is still a high demand.
But in the end, the entertainment should serve additional purposes, and contribute to progress towards the possibility of finding answers instead of looking for them.
One example of the difficulty of answering questions in the real world is the question "Where is X near where I am right now?" Where X is some classification, for example a pharmacy. It can be assumed that the location of the questioner is known, and that we have a list of sites with their classification and location, indeed, almost any navigation device can display answers to such questions, but as in the case of search engines, here too we have become accustomed to understating our requirements.
If we were to ask a passerby, we would expect much more helpful answers. For example, the human answer will be considered at the time of the question. After midnight a human guide will try to direct us to a place that he thinks will still be open, because he used the hypothesis that the questioner is interested in buying something at the pharmacy, and that for that the pharmacy must be open. The human guide will also consider the situation of the questioner - for example, is he a pedestrian or a car driver, and will decide accordingly what the word "near" means.
Smart filtering
Despite the efforts, it is difficult to find significant successes when the search topics are not limited to specific areas to which the software has been specially adapted. However, there are other approaches that offer improvements and help in filtering the results.
One of the approaches is an attempt to combine the search results in files dealing with different topics. For example, Google's "Wonder Wheel" feature displays for a search the English name of "The Catcher in the Rye", alongside the usual search results, the topics: Salinger, characters, works to be submitted, the movie, quotes, and more.
There are many tools that employ similar methods or other forms of improving the speed and accuracy of Internet searching, and that the pace of progress in this area is increasing as a result. Maybe one day we can expect computers to find an answer to our questions instead of searching thousands of pages where the answer might be hiding.
Israel Binyamini works at ClickSoftware developing advanced optimization methods.
The full article was published in Galileo magazine, September 2009
22 תגובות
I didn't download the file from MF but from Gauhack, in any case I didn't know that something like this could last so long,
Remembering of course the oil burning in Algiers that took a year to put out I think but it was local (albeit very intense)
I recommend going to Google Art and seeing.
Better a straight search engine than some other lady on the phone we have enough of those.
Those who do not know how to look for software that will compensate for their disability.
Forget the question...I missed a part of the article...he already thought about it...
to "Lamushon"
I know the project and it is really an interesting project.
But for it to really function like a thinking human brain
It also needs an input device, doesn't it?
Think that at some point I take your brain
and disconnects it from any means of input... (all including the body)
Except that I'm not sure there is enough information and enough
Resolution of the information that does exist to make the simulation more accurate
This is to the necessary degree, and there is a quite likely situation that a brain is
A system is sensitive to initial conditions, which will cause this
The whole project would be equivalent to trying to predict the weather
The air in California in 5 years...
Let's wait and see
Just a question, why only the last one, you already knew the other links?
Thanks for the last link
Moshon, there is one project that is exactly what they are trying to do, to create a complete and working simulation of an entire human brain on a computer - "from the bottom up", this project has, in my opinion, the best chances of reaching real artificial intelligence soon and not just "as if" -
1. Interview with Idan Segev, our representative in this amazing project -
http://www.youtube.com/watch?v=L0AR1cUlhTk
2. Another conversation with Idan Segev -
http://www.youtube.com/watch?v=Bz5IUaRr8No
3. From Calcalist -
http://www.calcalist.co.il/local/articles/0,7340,L-3359876,00.html
4. Fascinating article about the project, part A and part B -
http://seedmagazine.com/content/article/out_of_the_blue
http://seedmagazine.com/content/article/out_of_the_blue/P2
5. An amazing and particularly fascinating lecture by the head of the project, it is recommended to watch on full screen (clicking on the enlargement button under the video window)
http://ditwww.epfl.ch/cgi-perl/EPFLTV/home.pl?page=start_video&lang=2&connected=0&id=365&video_type=10&win_close=0
Additionally:
The set of all sets is really a set in the sense that it contains other sets
But it is not a group in the sense that the operations of a theory can be applied to it
The groups because it has no "background" = it is not contained in any group.
Good Day
In the context of religions - this is not the place, I believe everything
It should be in its place, and science does not involve religions!
But as a math enthusiast (including its philosophical side)
Group theory is at the foundation of mathematics and is
Yes, it includes such a super-bundle and it is, as you know, paradoxical
Or if you want illogical.
Therefore it is possible, if there is a correlation between mathematics and
Reality because reality itself is not entirely logical.
It is of course possible that this is not relevant at all to the above universe
and to science.
Good Day
Invisible = impossible
Sorry
Moshon as my personal logical fallacy - because my instinctive understanding would be to say that in a certain way it is a group, and in other ways it is not, again a logical fallacy - still a point (certain situation) where there is one existence and not in our first definition of existence-existence, it seems to me that it is possible even if not It seems to be proof - not related and not really helpful to the discussion, but many of the religions and sects rely on an overarching idea - that above the existence of this being or any other being in fact
For Ecuador:
If it is agreed that in any particular group everything cannot be given from within the group
The question is, can there be a point where everything can be seen...
Or if you like (in set theory): is the set of all sets that are defined
As a group that is not contained in any group is a group at all...
Good Day
First of all - a note to the procedure: You are right, sorry I didn't notice the jump
Thank you for the enlightening answer, so I can only understand that truth is also a non-finite relative concept, as a derivative of the creation of the very idea of truth by us, there is no single truth - a sentence that always seems to me to actually contradict itself, right in essence..?
It was noted that much depends on the point of view, as you mentioned, but my belief is that there is a dimension where it is possible to "see everything"
But there is no logical support for the matter apparently, thanks in any case
In honor of "Math LeMoshon":
Without a doubt you are absolutely right!
Nevertheless these empty systems started from a greater emptiness
If I understood correctly, only from a learning algorithm without any "database"
of words which is nice in itself (note that the grammar of the language has not been entered
artificial but developed through trial and error).
Personally, I'm in favor of the bottom-up approach...
In my opinion, their experience is not enough from below...
And it lacks essential things.
I would start with the intelligence of an animal...that is, define
in the algorithm certain parameters that will constitute the "good"/"bad" filter
Similar to instincts related to survival and reproduction
(an algorithm that will examine the usefulness of each action/speech/deed in relation to those parameters)
And of course a learning mechanism that relies on this filter that produces "feelings" (diagnoses
of good and bad in relation to those parameters).
Artificial intelligence will also not develop in my opinion without some parallel types
of "senses"
Good Day
Ecuador:
First of all - a note to the procedure: in my opinion it is not correct to move the discussion from thread to thread. This makes tracking him extremely difficult.
Physics and mathematics were created as part of our search for ways to understand the world and predict the happenings in it.
They deliver the goods so well that many philosophers have found it appropriate to raise the question "How is it possible that the world obeys so well the laws described by mathematical theories?" And they even brought certain people (like Tagmark http://en.wikipedia.org/wiki/Max_Tegmark ) to propose the proposition that there is nothing in the world but mathematics.
I don't know what the founders of science expected to find. In my opinion it is not at all true (historically) that they expected to find a final set of laws from which everything derives. The idea of the "theory of everything" is an idea from recent years and regarding it there is actually no requirement for the finality of the axiom system (and certainly there is not even disappointment if all the axioms that need to be added are technical axioms in the style of "the collection of axioms so far is consistent" which actually do not mean anything new and their necessity is only formal).
But even if it were historically true that the founders of mathematics and physics hoped to find a final set of axioms - their real goal was to discover the truth and if the truth is that there is no such final system then this is what it is and we will win with that.
All this does not belong to the question of whether there is a God, except that it assures us that the God of gaps will always have a place to hide in - that is - there will always be something that will allow the followers of God to say a sentence like "Okay. So you formulated laws 1,2,3,4,5,.... 57 But these laws do not allow you to conclude whether X is true or false, therefore God is responsible for X, whom you have already deprived of all other positions.
Michael - If so-called sweetening physics and these were created to confirm philosophical theories of man and the universe,
And at some point in the development of these sciences, a person showed that with this logic there is no final answer - that's why the journey began - I don't quite understand what this actually means - is it to be concluded from Godel's work that no single theory will include everything? And if so, this is added to the philosophical approaches that " There is no God" or "God is dead"??
Systems of the type you are talking about are extremely poor and really do not represent true artificial intelligence, not even close to it, try while talking to such software to ask it questions that refer to the topic you were talking about with it only a few sentences ago, and you will see how it starts to get confused and does not know what to answer and how to relate to it , it is impossible to have a serious conversation with the software because it always only knows how to refer to the last sentence or two you wrote to it and has no real understanding of the topic of the conversation... These programs are light years away from true AI, it is a shell empty of content.
point:
Since it does not seem to me that the human will be upgraded in the near term - it is more productive to define the problem as a software problem.
Only in this way will we be able to improve our dealings with you.
Objectively - incompatibility between two systems (in this case the software and the person) is not a problem of one or the other system but a problem of their combination.
Almost all problems are like this and we usually prefer to define the problem in terms of the solution.
When a certain food causes a person stomach poisoning - it is also possible to claim that this is a problem in the person and even confirm this claim by the fact that the bacteria feed on that food without any problem.
Still, the solution in this case would be to replace the food and therefore it is said that the food is spoiled.
http://www.a-i.com
It's weird that you think it's weird
really
Good Day
It's strange that you didn't mention the Israeli AI that just went out for a spin
In the network with "the god"...
Good Day
It seems to me that the problem is not with the search engines but with the specificity of narrow-minded people, relative to the breadth of the search engine results.