Will we ever ... pass the Turing test for computers?
Welcome dear!
The more we reflect on the future of our services, the more we want to dream of space colonization and artificial intelligence. More and more steps are being made in the direction of Mars, and small steps, such as Apple’s Siri interface, are being made in the direction of AI.
Therefore, undoubtedly, the entire IT industry will move precisely in this key, although in our area of ​​hosting conservatism and reliability should come first.
')
In a word, we will try to dilute “specific” articles with articles filled with a few “dreams”.
What would you say if the computer could think like a person?
In 1950, Alan Turing, the father of computer science, offered a simple test.
Step one: develop a computer program that can imitate human conversation. (What was a feat, considering how primitive computers were in the middle of the 20th century).
Step two: put the computer behind the veil or somehow hide it from view.
Step three: invite the person to talk with the computer via text messages. Step four: ask the person whether the invisible interlocutor is the same as he is human or machine.
If he or she mistakenly takes a machine-generated conversation for a conversation with a person, then the conclusion: the computer, according to Turing, can be said to "think."
It sounds more like a room game than a serious mental experiment, and many regard it. However, the "Turing test" continued to be used in artificial intelligence research. What even gave rise to the annual contest called the Lebner Prize, which has been held since 1991, where judges hold short conversations with artificial intelligence programs and people, and then have to decide who is who. (The most "human" award is given).
Turing predicted that the computer would successfully complete his experiment before 2000, but to date no program has passed this test - even the recent winner of the Lebner Prize 2012 Chip Vivant. This is partly due to the conditions of the test itself - for example, how long should a computer participate in a conversation before a person decides on his identity? 5 minutes? Three hours, three o'clock? Turing never said - but also because it was much more difficult to perfectly simulate human conversation than anyone expected. So, what will it cost to build a machine that can pass the Turing test?
One thing we know for sure: the “head-on” solution cannot succeed. In the early years of research in artificial intelligence, “thinking” was understood as a simple combination of symbols using the rules of discrete mathematics. “In the 1960s, the idea arose of reducing“ the whole world ”to simple things and actions that you can call: a book, a table, talk, run,” says Robert French, a cognitive psychologist at the French National Center for Scientific Research. “All words in the dictionary are symbols that refer to our world. Therefore, if you arrange them all in a certain way, roughly speaking, thinking should appear. ”
But no. This approach, called “symbolic AI,” is crumbling like a house of cards, because subjected to ambiguity. In the end, no rule will explain how to properly answer the usual question “What's up?” (If you answer “Up is the opposite of down,” assume that you failed the Turing test). A tightly interconnected database may contain “smart” information, but this does not mean that it is smart. Brian Christian, author of The Most Human Human: What Artificial Intelligence Tells Us About Being Alive, says: “When we read a book, we don’t think that the book has ideas.”
You can imitate human communication if you abandon logic and be anonymous. With such communication, each answer is loosely connected in meaning with the previous message. And this communication model can easily be programmed. The so-called "chat bots" were common in the 60s. Eliza is one of the world's first chat bots, who essentially imitated a psychotherapist and responded to users using the vocabulary of their messages. 3 times received the Loebner Prize in the 2000s with a similar program more complex bot "Alice". These bots still can hardly claim artificial intelligence. What is funny, it is their almost vacuousness similar to human communication. “Alice” may have surpassed other programs, but she failed to deceive the living judges during the Turing test.
Given that the mind is largely unconscious and consists of associative processes of perception, there is an interaction between the individual, who asks the question: "what does it look like?" And the world around it. This “sub-cognitive” information could include the memory of a bike breakdown and an abrasion on the knee, or a piece of sandwich on the beach, and a feeling of sand crunch between your teeth. It also includes more abstract things, like the answer to the following question: “Is the name Flugly better for a charming actress or a teddy bear?”
Robert French claims that while this is nonsense, almost any English-speaking person would choose a teddy bear. Why? “The computer does not have a history of embodied experience of soft teddy bears, contemplation of pretty actresses, or even sounds of the English language,” says the French scientist. “All these things allow people to answer such questions consistently that a computer is not given.” Therefore, any free program has an Achilles heel when it comes to taking the Turing test.
But that may change soon. French refers to experiments on the phenomena of life, for example, the efforts of the researcher Deb Roy (Massachusetts Institute of Technology, USA) to videotape every moment of awakening his infant as a possible embodiment of this problem. “What if the computer had the same information, sound and visual impressions that a person had for many, many years?” French says: “We can now collect this data. If a computer analyzes them and correlates them correctly, is it unreasonable to assume that it will answer questions like the question of the name Flugly just like a person? ”
French does not think that continuous analysis of such a massive data set will be possible soon. “But at some point in time, we will achieve this,” he says. If a computer program passes the Turing test - what will it mean in a practical sense? We will consider the device reasonable? Or, we’ll just add “able to communicate” to the ever-growing list of interesting things computers can do, such as “beat a man in chess” (like IBM Dark Blue did in a match with Garry Kasparov in 1997).
Well-known programmer Edsger Dijkstra said that “the question of whether machines can be thought of is almost as relevant as the question of whether submarines can swim.” Leaving the “mind” aside, you can be sure that the computer that can pass the Turing test does exactly one thing very well: it speaks to people like a human being. Which means that the Turing test may simply be replaced by other questions, those that are not so difficult to answer. "A computer can imitate a person," says Brian Christian. "But to which one?"
The more we reflect on the future of our services, the more we want to dream of space colonization and artificial intelligence. More and more steps are being made in the direction of Mars, and small steps, such as Apple’s Siri interface, are being made in the direction of AI.
Therefore, undoubtedly, the entire IT industry will move precisely in this key, although in our area of ​​hosting conservatism and reliability should come first.
')
In a word, we will try to dilute “specific” articles with articles filled with a few “dreams”.
What would you say if the computer could think like a person?
In 1950, Alan Turing, the father of computer science, offered a simple test.
Step one: develop a computer program that can imitate human conversation. (What was a feat, considering how primitive computers were in the middle of the 20th century).
Step two: put the computer behind the veil or somehow hide it from view.
Step three: invite the person to talk with the computer via text messages. Step four: ask the person whether the invisible interlocutor is the same as he is human or machine.
If he or she mistakenly takes a machine-generated conversation for a conversation with a person, then the conclusion: the computer, according to Turing, can be said to "think."
It sounds more like a room game than a serious mental experiment, and many regard it. However, the "Turing test" continued to be used in artificial intelligence research. What even gave rise to the annual contest called the Lebner Prize, which has been held since 1991, where judges hold short conversations with artificial intelligence programs and people, and then have to decide who is who. (The most "human" award is given).
Turing predicted that the computer would successfully complete his experiment before 2000, but to date no program has passed this test - even the recent winner of the Lebner Prize 2012 Chip Vivant. This is partly due to the conditions of the test itself - for example, how long should a computer participate in a conversation before a person decides on his identity? 5 minutes? Three hours, three o'clock? Turing never said - but also because it was much more difficult to perfectly simulate human conversation than anyone expected. So, what will it cost to build a machine that can pass the Turing test?
Watch your tongue
One thing we know for sure: the “head-on” solution cannot succeed. In the early years of research in artificial intelligence, “thinking” was understood as a simple combination of symbols using the rules of discrete mathematics. “In the 1960s, the idea arose of reducing“ the whole world ”to simple things and actions that you can call: a book, a table, talk, run,” says Robert French, a cognitive psychologist at the French National Center for Scientific Research. “All words in the dictionary are symbols that refer to our world. Therefore, if you arrange them all in a certain way, roughly speaking, thinking should appear. ”
But no. This approach, called “symbolic AI,” is crumbling like a house of cards, because subjected to ambiguity. In the end, no rule will explain how to properly answer the usual question “What's up?” (If you answer “Up is the opposite of down,” assume that you failed the Turing test). A tightly interconnected database may contain “smart” information, but this does not mean that it is smart. Brian Christian, author of The Most Human Human: What Artificial Intelligence Tells Us About Being Alive, says: “When we read a book, we don’t think that the book has ideas.”
You can imitate human communication if you abandon logic and be anonymous. With such communication, each answer is loosely connected in meaning with the previous message. And this communication model can easily be programmed. The so-called "chat bots" were common in the 60s. Eliza is one of the world's first chat bots, who essentially imitated a psychotherapist and responded to users using the vocabulary of their messages. 3 times received the Loebner Prize in the 2000s with a similar program more complex bot "Alice". These bots still can hardly claim artificial intelligence. What is funny, it is their almost vacuousness similar to human communication. “Alice” may have surpassed other programs, but she failed to deceive the living judges during the Turing test.
Photographic memory
Given that the mind is largely unconscious and consists of associative processes of perception, there is an interaction between the individual, who asks the question: "what does it look like?" And the world around it. This “sub-cognitive” information could include the memory of a bike breakdown and an abrasion on the knee, or a piece of sandwich on the beach, and a feeling of sand crunch between your teeth. It also includes more abstract things, like the answer to the following question: “Is the name Flugly better for a charming actress or a teddy bear?”
Robert French claims that while this is nonsense, almost any English-speaking person would choose a teddy bear. Why? “The computer does not have a history of embodied experience of soft teddy bears, contemplation of pretty actresses, or even sounds of the English language,” says the French scientist. “All these things allow people to answer such questions consistently that a computer is not given.” Therefore, any free program has an Achilles heel when it comes to taking the Turing test.
But that may change soon. French refers to experiments on the phenomena of life, for example, the efforts of the researcher Deb Roy (Massachusetts Institute of Technology, USA) to videotape every moment of awakening his infant as a possible embodiment of this problem. “What if the computer had the same information, sound and visual impressions that a person had for many, many years?” French says: “We can now collect this data. If a computer analyzes them and correlates them correctly, is it unreasonable to assume that it will answer questions like the question of the name Flugly just like a person? ”
French does not think that continuous analysis of such a massive data set will be possible soon. “But at some point in time, we will achieve this,” he says. If a computer program passes the Turing test - what will it mean in a practical sense? We will consider the device reasonable? Or, we’ll just add “able to communicate” to the ever-growing list of interesting things computers can do, such as “beat a man in chess” (like IBM Dark Blue did in a match with Garry Kasparov in 1997).
Well-known programmer Edsger Dijkstra said that “the question of whether machines can be thought of is almost as relevant as the question of whether submarines can swim.” Leaving the “mind” aside, you can be sure that the computer that can pass the Turing test does exactly one thing very well: it speaks to people like a human being. Which means that the Turing test may simply be replaced by other questions, those that are not so difficult to answer. "A computer can imitate a person," says Brian Christian. "But to which one?"
Source: https://habr.com/ru/post/145167/
All Articles