This is what Sheila Ofek Kaufman, director of the natural language research department in the artificial intelligence division at the IBM research laboratory in Haifa, says in an interview with the scientist website, recounting the scientific progress of IBM Haifa researchers on the way to the development of the Debater
Extended coverage on the science website:
- The computer argues with the debating champions and ends up in a draw
- Over 30 scientific articles and breakthroughs by IBM Haifa researchers. Project Debater milestones
- My friends and rivals - IBM introduced Project Debater: an AI system for debate competitions
Sheila Ofek Kaufman, director of the natural language research department in the artificial intelligence division at IBM's research laboratory in Haifa, who was a partner in the development of Project Debater, tells the science website about the project, and about the debate that took place between the computer and professional experts yesterday in Givatayim.
Ofek Kaufman has worked for 14 years at IBM, and before that she studied computer science and an MBA at the Technion.
How do you turn a computer into a high-level expert in the English language, up to the point of formulating complex sentences?
"Language research is an existing field that is developing rapidly. There are quite established technologies in the field of analyzing parts of speech, sentence structure, entity recognition ie names of people and places. The significant leap we made during the development of Debater was in a deep understanding of the language. And beyond that, there is the ability to produce a text that is both long, also expresses a position, and coherent, a person understands the logical sequence of things and is also convincing. We were required to give the computer the same ability. All of this is required so that the system can formulate the first speech."
"An important element is the search in the databases. The system scans a database of 300 million documents, 10 billion sentences - and combines them together into a coherent and persuasive speech. It starts with a search as we know from the Internet, but it is required not only to find content that talks about the subject but also to identify arguments, to understand that some sentence is an argument. Just explaining what the concept is is not enough for a debate, but one needs to find argumentative content, that is, that a certain sentence contains a claim and not just a sentence that mentions the subject."
Then one must understand whether he claims to support the claim we are trying to discuss or whether he attacks it. Sometimes the sentence can also not mention any of the keywords. It is possible to understand this either from the context of a paragraph or from a semantic understanding that this is a higher concept, which relates to the subject we are discussing. This is in addition to the insight that an argument is part of a complete concept that the proven computer wants to bring up in the discussion, and the system needs to understand the position it is required to defend.
To summarize this part, Ofek-Kaufman says, the system produces a speech of several minutes, finds sources, quotes from studies and leading people in the field, checks whether they match or attack the topic and builds a consistent and organized argument from all of this, in which the sentences connect well.
and how does the debateThe answer to the human argument?
"Here we were required to develop technologies in the field of listening comprehension. Today, the computerized assistants and chatbots of various kinds know how to receive a one-sentence command, analyze it and give a short answer or carry out the request. Here the Debater has to listen to a four-minute speech by a person who sometimes speaks fast, and also understand the intonation."
"The first step was to use Speech to Text systems that turn the words of the human speaker into written text. Then analyze the text and try to understand what the gist of the person's words is, what are the higher concepts he is attacking, and then prepare the next speech that refers to what he said, sometimes the debater tries to attack the claims of his opponent but not always, sometimes he simply expands on the arguments he put forward himself in the previous speech. Another area we had to deal with is modeling the world of debates - understanding how a debate is conducted. This is what we learned while experimenting and working with the human debaters. We learned how they attack the subject and how to understand nuances."
After the arguments, my impression is that the Debater does not pass the Turing test, what are your plans?
Ofek Kaufman: "The development of the system is not over, we have reached a point where Debater is mature enough to present it and show the capability we have reached so far. Indeed, his ability to deliver the first speech is much better than his ability to refute the claims of the human debater, and there is a game of confidence here. We led you to safety so that the system will attack the opponent's claim when it has understood the opponent's claims and is confident that it can respond correctly."
"This is the point where there is a lot of work. The degree of confidence in understanding the speaker's words obviously depends on the material that the debater has on the specific topic. She listens to her opponent and checks whether she has enough material that allows her to refute or contradict what he said. If she doesn't have it, she simply reinforces her previous claims."
In conclusion, Ofek Kaufman says that the purpose of the project put forward by Dr. Noam Salonim and Dr. Ranit Aharonov, and thanks to which it was accepted by IBM worldwide, was precisely to break the boundaries of science, although in the more advanced stages of the competition it was necessary to explain how the system could be implemented and What will be its commercial aspect.
The center of gravity of the development was in Haifa.
4 תגובות
Neurolinguistics
As neurophysics replaced Newton's physics, so neurolinguistics replaced Chomsky's linguistics.
Newton believed that there exists in physical reality an imaginary idea called gravity, and Chomsky believed that humans naturally create languages with fixed syntax
Newton misled the physicists for 300 years, and Chomsky misled the linguists for 80 years.
Neurolinguistics is free and has no fixed syntax or certain grammar rules.
Neurolinguistics is simple and self-explanatory, and it is based on a person's natural knowledge.
Neurolinguistics develops in the identification of organs in the human body, which allow him to speak. (throat, mouth, tongue)
Throat, mouth, and tongue are a natural piano of 22 notes
A natural piano belongs to a person with natural knowledge
The man with natural knowledge invented a language for himself based on the sounds produced by the natural piano.
And this is how the process of creating a language is described.
A silent person with natural knowledge touches a block of ice, and miraculously a clear natural knowledge comes to him.
The silent person suddenly realized that he knew something.
There is no way to describe this news because man has not yet invented words. And even though the words had not yet been invented, the person who touched a block of ice suddenly knew a wonderful and clear new knowledge.
And here a miracle happened, and the silent man realized that he needed to give a name to this clear knowledge.
This silent person is a person with a natural knowledge inherent in him, and he knew that his natural piano can provide sounds, which can be used as the name of
The aforementioned natural knowledge.
This silent man chose a short name, which derives from two notes of his natural piano, the sound of K and the sound of R
And so it happened, that the combination of the two sounds (cold) became the name of a natural knowledge, which comes to man following contact with a block of ice.
This is where the creation of human language begins, which is the language of names of natural information.
All the conditions of creation are found in the silent person.
Ingenuity is inherent in the silent person, natural knowledge is inherent in the silent person, and the natural piano is part of the silent person's body.
Every person can learn a language of natural knowledge. This language requires the person who wants to learn to make an act of touching a block of ice, like the silent person,
After that, the learner will come to the two cold sounds of his natural piano, and then he will understand that this is the name of the natural knowledge that came to him, following contact with a block of ice.
The name of natural knowledge is chosen arbitrarily,
And any combination of sounds can correspond to this.
The combination of sounds (חם) was chosen to be the name of a natural knowledge, which miraculously comes to the person approaching the fire.
The combination of the sounds Hm was also chosen arbitrarily.
This is how human language is created, based on natural information that comes to them, following actions they do in physical reality.
such as (to touch, walk, jump, taste, smell) etc.
This language is a name language of natural information
coming to humans, following actual actions they do in physical reality.
Touch and natural knowledge comes, smell and another natural knowledge comes, see and natural knowledge comes
Otherwise, you taste and another natural knowledge comes, and more,
Now it remains to choose sounds from the natural piano, like
The sounds of Halak, or the sounds of Mahosfs, or the sounds of Metok, and these will be the names of some of the natural knowledge that came to humans, following the actions they did. in physical reality.
This is how a human language is created, which is a language of names of natural information. Every word in the human language is
Name of natural knowledge - coming to man, following an actual act he does in physical reality.
It is impossible to create human language with the help of the sounds emanating from the throat, because the sounds are just noise.
If the voices are names of natural information following an action in physical reality, then the language is created immediately, and it can be used.
All linguists in the last 100 years have used the noise of words coming from their throats to create human language and explain it.
All these linguists have failed. And as a result of this failure, the Nesbari Law appeared saying:
You cannot study a language through its words.
Neurolinguistics began to create a human language, which is a language of names of natural information.
The selection of names takes place with the help of the natural pianos of humans, which are able to produce many sounds such as sound A, sound B, sound C and so on.
Humans easily invented thousands of languages, because they
They knew - that language is invented by actions, not by words.
The linguists of the last 100 years did not know that language is invented by actions, and they spoke, and spoke,
And they spoke, and made a noise from their throats - which remained a noise.
This noise failed to invent a language.
Esbar's research revealed the secret of human language, which is the language of names of natural information - which come to man following an act he does,
Every such act brings natural knowledge to the person, and the person who creates the language gives a name to the natural knowledge.
Neural linguistics, versus Chomsky's linguistics.
Neurolinguistics is completely new, and it is completely different from Chomsky's linguistics.
Chomsky's linguistics is based on words, which are scribbles of meaningless lines, so the failure of this linguistics is predictable.
There is no possibility to understand doodles, with the help of other doodles, whose names are words.
Neurolinguistics is based on actual actions
that man does, and these actions bring man
Miraculously, natural news.
A person touching a block of ice receives a very clear natural knowledge, and the person gives it an arbitrary name.
He achieves the arbitrary name with the help of his natural piano. capable of making many sounds.
People already know the natural knowledge that the name of consists of the sound k and the sound r
This whole process takes place with the help of natural intelligence.
A robot touching a block of ice has no natural intelligence, and no natural knowledge.
The robot has artificial intelligence, and it can measure the temperature of the contact, but it does not have a clear natural knowledge of the contact with a block of ice.
A person has such clear natural knowledge.
There is a profound difference between artificial and natural intelligence. Natural intelligence belongs to a person who obtains natural knowledge as a result of an act he does.
Artificial intelligence is limited, and will never reach the natural intelligence of a human.
There are many expectations of artificial intelligence, and neural linguistics has lowered the level of expectations, to a simple mechanical level that works, but does not know.
Neurolinguistics is a song of praise for the knowledgeable person
A. Asbar
The era of scientific revolutions based on "natural knowledge" has arrived.
It is about a linguistic revolution that discovered the secret of human language.
And it is also about a geometric revolution, which discovered a new geometry.
And it is also about a physical revolution, which discovered a new physics and a new universe.
Computers will never be able to speak human language, because computers have no natural knowledge.
A. Asbar
https://nivbook.co.il/product/%D7%9E%D7%A1%D7%A2-%D7%94%D7%A7%D7%A1%D7%9D-%D7%A9%D7%9C-%D7%A2%D7%A6%D7%91%D7%A8-%D7%A2%D7%9C-%D7%9B%D7%A0%D7%A4%D7%99-%D7%94%D7%99%D7%93%D7%99%D7%A2%D7%94-%D7%94%D7%98%D7%91%D7%A2%D7%99%D7%AA/
In my opinion - there is no AI software today that can pass a Turing test, there are only people who don't know how to talk
with the programs in a sophisticated enough way to reveal the fact that they are artificial.
Sounds cool, like chess with a computer only as a debate.
I'll wait until it comes out and I'll be happy to try it