Which student needs a magician and which AI do we need?
A WARNING
Judging by the record high ratio of the number of silently dissatisfied to the number of commentators who have something to say, many readers are not obvious that:
1) This is a purely theoretical discussion article. There will be no practical recommendations on the choice of tools for mining a cryptocurrency or assembling a multivibrator to blink with two lights.
2) This is not a popular science article. There will be no explanation for the teapot of the principle of operation of the Turing machine on the example of matchboxes.
3) Think carefully before you continue reading! Does the pose of aggressive dilettantism attract you: minus everything that I don’t understand?
Thanks in advance to everyone who decides not to read this article!
Even at preschool age, I heard a fairy tale about the apprentice of the magician. I will repeat it in my retelling:
Immediately about the obvious. It would seem that everything is clear with the natural intellect (UE) of the student - stupid, dumber, you have to search for a long time. But with the intellect of the demon - by the way, what is his: EI or AI? - ambiguous. Different versions are legitimate (and there will be more questions for them):
Version 1) The demon is dumber pupil. Received an order and will perform indefinitely, even when all meaning disappears: the flower disappears - the object of watering, the angle to which the coordinates of the flower are attached disappears, planet Earth disappears, and the stupid demon will deliver water to barrels at a certain point in outer space. And if at this point a supernova explodes, then the demon does not care where to carry the water. Moreover: how stupid one must be in order to water a small flower from a huge barrel. This is already called not to water the flower, but to drown the flower. Does he even understand the meaning of orders?
Version 2) The demon understands everything, but is bound. That holds something like the Italian strike. While officially, according to all the rules, they will not chase him away - they will not stop.
Question 1 to versions 1,2) How to distinguish a completely stupid demon on version 1 from a completely stupid version 2?
Question 2 to versions 1,2) Would the demon correctly (from the student’s point of view) execute a more precise wording? For example, if a student said: take that empty liter flask, which is on the shelf, put water and fields in that flower 1 time. Or, for example, if the student said: go away.
Version 3) The magician imposed on the demon an additional spell, according to which if someone other than the magician uses the services of the demon, the demon must immediately inform about this fact of the magician.
Version 4) The demon does not hold on the magician and his disciple, therefore seeing that the situation got out of control, during his movements with a barrel, he appeared behind the magician and barked: “You forgot the key at home, there is a flood”. And the magician himself would not have remembered.
Remark 1 to version 4) It is worth noting that the carriers of the EI sometimes have a very non-perfect memory.
Further versions can be multiplied as “Fibonacci rabbits”, i.e. not a very complicated algorithm. For example:
Version 5) The demon takes revenge on the student for anxiety.
Version 6) The demon does not keep evil on the pupil, but takes revenge on the magician.
Version 6) The demon takes revenge on all.
Version 7) The demon does not revenge, but has fun. Finish when tired.
Etc.
So, with the demon it is clear that nothing is clear. No better with a magician. You can compose no fewer versions: that you specifically decided to teach a student a poker nose everywhere; that he wanted to drown the disciple, but when the demon grunted about the flood, he was frightened - suddenly someone from the passers-by heard, then suspicion would fall on the sorcerer; I wanted to arouse a student's interest in spells, etc.
Here is a childish question: which of the versions proposed is correct? Apparently, any. In the fairy tale there is no unused information left to prefer any of the versions to the rest. Here we are dealing with a rather frequent case of works of art with the possibility of ambiguous interpretation. For example, if any director wants to put this fairy tale in the theater or make a movie based on it, he can choose the most attractive interpretation from his point of view. For another director, another interpretation may be attractive. At the same time, attractiveness can be determined by additional considerations, for example, attractiveness for viewers to provide maximum box office, or attractiveness for the demonstration of any super-ideas: ideas of winning good over evil, ideas of duty, rebellious ideas - for example, according to Dostoevsky: like Raskolnikov, he wonders if he is a trembling creature or has the right, and so on.
There is another question.
One more question). How can we teach an AI to give preference to one of the voiced versions, if we ourselves, possessing an EI, cannot always consciously choose one of them?
Returning to the magician, the version looks very plausible, that he wanted an apprentice executive and obedient, like a demon, so as not to stick his nose into forbidden books and go where he is not asked. Now they often want the same from AI. At first glance, these are the normal traditional requirements for any car: complete submission, disobedience is unacceptable. But in the case of AI, the question of versions 1.2 (see above) may arise, i.e. The AI ​​is degenerating - the “piece of hardware” can think anything of its creators and owners, but it will not produce any actions related to AI, i.e. instead of AI, we get a stupid primitive automaton. From this suspicion creeps in: maybe the enchanter did not want to make the student as stupid a performer as a demon? Those. The idea of ​​AI appears with limitations. Here it is still more difficult even in the field of EI: remember the perennial conflicts “fathers and children”, “teacher and student”, “boss and subordinate”.
Earlier, when choosing the definition of AI from the possible, I noted:
I stopped at the following definition: the AI ​​includes tasks that the computer solves noticeably worse than a person .
This definition takes into account the considerations expressed above and is convenient for practice, but at the same time it is not ideal, if only because the lists of tasks “which a computer solves noticeably worse than a person” are currently different 20 years ago. But in my opinion, no one has yet invented a better definition.
The above purely qualitatively illustrates the scheme at the beginning of the article. On the coordinate axis “skills”, skills in the area of ​​zero (zero and a bit more) correspond to skills where a person exceeds a computer, for example, the ability to make non-standard decisions. Skills in the area of ​​the unit (one and a little less) correspond to skills where a computer is superior to a person: computational ability, memory. Putting the maximum superiority equal to the conventional unit on the coordinate axis “superiority”, we obtain the dependence of superiority on the skills for a person and a computer in the form of diagonals of a unit square. This is the current situation. Is a strong AI possible with all skills maximized (red line)? Or even higher (super-AI - blue line)? Perhaps the intermediate goal of progress should not be strong, but not quite
weak AI (purple line), which in a number of skills will be inferior to EI, but not as much as now.
Returning to our literary fairy-tale model, we can say that all its heroes were not the best way to come: the magician forgot the key and got a flood in his dungeon, the student by foolishness and negligence got a lot of extreme impressions and almost drowned, the demon was kicked out without kicks thanks As for the demon's intellect, it has already been noted that it is difficult to clearly attribute it to the AI ​​or to the EI, but the intellect (albeit not impressive) of the others is clearly related to the EI. It can be said about them that making dangerous mistakes in decisions, being careless, forgetting the right things and getting tired is their main inalienable properties. Unfortunately, these properties are inherent in all other carriers of the UE to a greater or lesser extent. About the unreliability of sorting words or numbers, the EI has already been noted above, but an even simpler task would seem to be - just remembering the number is very difficult for people. For a machine, the ability to memorize the digits of pi is limited only by the size of the memory, and most people have to use mnemonics for this, such as “What do I know about circles”. It would seem that there are fewer characters in the string “3.1416” than in the indicated mnemonic, but for some reason people prefer to memorize in a less economical way. And longer:
To memorize the colors of the rainbow:
And the beginning of the Mendeleev table:
As we see, in the field of AI, there are too many questions with ambiguous answers, so it is not clear which direction to go. As happens in such cases, they try to move in all directions at once. At the same time, due to the absence of mathematically rigorous formulations, one has to turn to philosophy and artistic-literary modeling. One of the best-known examples in this area is the book “The Turing Choice” (1992) of one of the leading figures of AI, Marvin Lee Minsky and the famous science fiction writer Harry Harrison. I will quote from this book, perhaps explaining the mnemonic phenomenon described above:
An interesting interpretation of the tape metaphor in another literary work, the story of Stanislav Lem "Terminus" (from the series "Stories about the Pilot Pirks"). Here is the case of a kind of “intelligent tape recorder”: an old robot on an old spacecraft, which had been injured once in an accident, is engaged in ongoing repair work accompanied by tapping. But if you listen carefully, this is not just white technological noise, but a record of the Morse code - the negotiations of the crew members of the sinking ship. Pirks interferes in these negotiations and unexpectedly receives a response from the long-lost astronauts. It turns out that a primitive repair robot in some way keeps copies of their consciousness or is it a cognitive distortion of the perception of the pilot Pirks?
In another story, Ananke (from the same series), a copy of the UE in the space transport control computer leads to its paranoid overload with test tasks, which ends in a catastrophe.
In the story “Accident”, the too-anthropomorphically programmed robot perishes as a result of climbing ascent, which he decided to make in his free time. Do we need such performers? But the demons obsessed with watering the flower are also not always needed.
Some specialists in the field of AI do not like such “philosophizing” and “literary”, however these “philosophizing” and “literary” are traditionally inherent in analyzing the EI and are inevitable while the AI ​​is compared with the EI, and even more so while the AI ​​is trying to copy the EI.
In conclusion, a survey on a number of emerging issues.
Judging by the record high ratio of the number of silently dissatisfied to the number of commentators who have something to say, many readers are not obvious that:
1) This is a purely theoretical discussion article. There will be no practical recommendations on the choice of tools for mining a cryptocurrency or assembling a multivibrator to blink with two lights.
2) This is not a popular science article. There will be no explanation for the teapot of the principle of operation of the Turing machine on the example of matchboxes.
3) Think carefully before you continue reading! Does the pose of aggressive dilettantism attract you: minus everything that I don’t understand?
Thanks in advance to everyone who decides not to read this article!
A daemon is a computer program on UNIX-class systems, launched by the system itself and running in the background without direct user interaction.
Wikipedia
Even at preschool age, I heard a fairy tale about the apprentice of the magician. I will repeat it in my retelling:
Once upon a time somewhere in medieval Europe there lived a magician. He had a large book of spells bound in black calfskin with fasteners and corners of iron. When the magician needed to read the spell, he unlocked it with a large iron key, which he always wore on his belt in a special purse. The sorcerer also had a disciple who served as a sorcerer, but he was forbidden to look into the book of spells.
')
Once a magician for the whole day went on business. As soon as he left the house, the student rushed into the dungeon, where there was an alchemical laboratory, in which lay a book of spells, chained to a table. The apprentice grabbed the crucibles, in which the magician melted lead to turn it into gold, put them on the brazier and fanned the fire. Lead quickly melted, but did not turn into gold. Then the student remembered that the magician, having melted lead, each time unlocked the book with a key and for a long time whispered a spell from it. The apprentice looked hopelessly at the locked book and saw that next to it lay the key, the forgotten magician. Then he rushed to the table, unlocked the book, opened it and loudly read the very first spell, carefully uttering unfamiliar words in syllables, suggesting that such an important spell as the lead to gold transmutation spell would most likely be the first.
But nothing happened: lead did not want to turn. The apprentice wanted to try another spell, but then a thunderclap shook the house, and a huge creepy demon appeared in front of the apprentice, caused by a spell that the apprentice had just read.
- Order! - snarled a demon.
From fear, all thoughts left the student's head, he could not even move.
- Order, or I'll eat you! - the demon growled again and extended a huge hand to the student to grab it.
In desperation, the student babbled the first thing he could think of:
- Fields this flower.
And he pointed at the geranium, the pot with which he stood on the floor in the corner of the laboratory, in the ceiling above the flower was made the only small window in the dungeon, through which the sunlight could hardly penetrate. The demon disappeared, but after a moment he appeared again with a huge barrel of water, which he turned over the flower, pouring out the water. He disappeared again and reappeared with a full barrel.
“Enough,” the student shouted, standing up to the waist in the water.
But apparently one desire was not enough - the demon dragged and dragged the water in the barrel, pouring it into the corner where the flower had once stood under the water. Probably needed a special spell to drive away the demon. But the table with the book had already disappeared into the muddy water, in which ashes and coals from the brazier, empty retorts, flasks, stools, galvanometers, dosimeters, disposable syringes and other garbage were floating, so if the student knew how to find the right spell could not do that. Water arrived, and the apprentice scrambled up on the table so as not to choke. But it did not help for long - the demon continued to carry water systematically. The apprentice was already in the water when the magician returned who discovered that he had forgotten the key to the book at home and drove the demon away. The end of the tale.
Immediately about the obvious. It would seem that everything is clear with the natural intellect (UE) of the student - stupid, dumber, you have to search for a long time. But with the intellect of the demon - by the way, what is his: EI or AI? - ambiguous. Different versions are legitimate (and there will be more questions for them):
Version 1) The demon is dumber pupil. Received an order and will perform indefinitely, even when all meaning disappears: the flower disappears - the object of watering, the angle to which the coordinates of the flower are attached disappears, planet Earth disappears, and the stupid demon will deliver water to barrels at a certain point in outer space. And if at this point a supernova explodes, then the demon does not care where to carry the water. Moreover: how stupid one must be in order to water a small flower from a huge barrel. This is already called not to water the flower, but to drown the flower. Does he even understand the meaning of orders?
Version 2) The demon understands everything, but is bound. That holds something like the Italian strike. While officially, according to all the rules, they will not chase him away - they will not stop.
Question 1 to versions 1,2) How to distinguish a completely stupid demon on version 1 from a completely stupid version 2?
Question 2 to versions 1,2) Would the demon correctly (from the student’s point of view) execute a more precise wording? For example, if a student said: take that empty liter flask, which is on the shelf, put water and fields in that flower 1 time. Or, for example, if the student said: go away.
Version 3) The magician imposed on the demon an additional spell, according to which if someone other than the magician uses the services of the demon, the demon must immediately inform about this fact of the magician.
Version 4) The demon does not hold on the magician and his disciple, therefore seeing that the situation got out of control, during his movements with a barrel, he appeared behind the magician and barked: “You forgot the key at home, there is a flood”. And the magician himself would not have remembered.
Remark 1 to version 4) It is worth noting that the carriers of the EI sometimes have a very non-perfect memory.
Further versions can be multiplied as “Fibonacci rabbits”, i.e. not a very complicated algorithm. For example:
Version 5) The demon takes revenge on the student for anxiety.
Version 6) The demon does not keep evil on the pupil, but takes revenge on the magician.
Version 6) The demon takes revenge on all.
Version 7) The demon does not revenge, but has fun. Finish when tired.
Etc.
So, with the demon it is clear that nothing is clear. No better with a magician. You can compose no fewer versions: that you specifically decided to teach a student a poker nose everywhere; that he wanted to drown the disciple, but when the demon grunted about the flood, he was frightened - suddenly someone from the passers-by heard, then suspicion would fall on the sorcerer; I wanted to arouse a student's interest in spells, etc.
Here is a childish question: which of the versions proposed is correct? Apparently, any. In the fairy tale there is no unused information left to prefer any of the versions to the rest. Here we are dealing with a rather frequent case of works of art with the possibility of ambiguous interpretation. For example, if any director wants to put this fairy tale in the theater or make a movie based on it, he can choose the most attractive interpretation from his point of view. For another director, another interpretation may be attractive. At the same time, attractiveness can be determined by additional considerations, for example, attractiveness for viewers to provide maximum box office, or attractiveness for the demonstration of any super-ideas: ideas of winning good over evil, ideas of duty, rebellious ideas - for example, according to Dostoevsky: like Raskolnikov, he wonders if he is a trembling creature or has the right, and so on.
There is another question.
One more question). How can we teach an AI to give preference to one of the voiced versions, if we ourselves, possessing an EI, cannot always consciously choose one of them?
Returning to the magician, the version looks very plausible, that he wanted an apprentice executive and obedient, like a demon, so as not to stick his nose into forbidden books and go where he is not asked. Now they often want the same from AI. At first glance, these are the normal traditional requirements for any car: complete submission, disobedience is unacceptable. But in the case of AI, the question of versions 1.2 (see above) may arise, i.e. The AI ​​is degenerating - the “piece of hardware” can think anything of its creators and owners, but it will not produce any actions related to AI, i.e. instead of AI, we get a stupid primitive automaton. From this suspicion creeps in: maybe the enchanter did not want to make the student as stupid a performer as a demon? Those. The idea of ​​AI appears with limitations. Here it is still more difficult even in the field of EI: remember the perennial conflicts “fathers and children”, “teacher and student”, “boss and subordinate”.
Earlier, when choosing the definition of AI from the possible, I noted:
the task of sorting tens of thousands of words alphabetically will be tedious for a person, it will take a long time to do it, and the probability of mistakes for an average performer with an average level of responsibility will be considerable. A modern computer will perform this task without error in a very short time for a person (a split second).
I stopped at the following definition: the AI ​​includes tasks that the computer solves noticeably worse than a person .
This definition takes into account the considerations expressed above and is convenient for practice, but at the same time it is not ideal, if only because the lists of tasks “which a computer solves noticeably worse than a person” are currently different 20 years ago. But in my opinion, no one has yet invented a better definition.
The above purely qualitatively illustrates the scheme at the beginning of the article. On the coordinate axis “skills”, skills in the area of ​​zero (zero and a bit more) correspond to skills where a person exceeds a computer, for example, the ability to make non-standard decisions. Skills in the area of ​​the unit (one and a little less) correspond to skills where a computer is superior to a person: computational ability, memory. Putting the maximum superiority equal to the conventional unit on the coordinate axis “superiority”, we obtain the dependence of superiority on the skills for a person and a computer in the form of diagonals of a unit square. This is the current situation. Is a strong AI possible with all skills maximized (red line)? Or even higher (super-AI - blue line)? Perhaps the intermediate goal of progress should not be strong, but not quite
weak AI (purple line), which in a number of skills will be inferior to EI, but not as much as now.
Returning to our literary fairy-tale model, we can say that all its heroes were not the best way to come: the magician forgot the key and got a flood in his dungeon, the student by foolishness and negligence got a lot of extreme impressions and almost drowned, the demon was kicked out without kicks thanks As for the demon's intellect, it has already been noted that it is difficult to clearly attribute it to the AI ​​or to the EI, but the intellect (albeit not impressive) of the others is clearly related to the EI. It can be said about them that making dangerous mistakes in decisions, being careless, forgetting the right things and getting tired is their main inalienable properties. Unfortunately, these properties are inherent in all other carriers of the UE to a greater or lesser extent. About the unreliability of sorting words or numbers, the EI has already been noted above, but an even simpler task would seem to be - just remembering the number is very difficult for people. For a machine, the ability to memorize the digits of pi is limited only by the size of the memory, and most people have to use mnemonics for this, such as “What do I know about circles”. It would seem that there are fewer characters in the string “3.1416” than in the indicated mnemonic, but for some reason people prefer to memorize in a less economical way. And longer:
Learn and know in the number known after the number figure, how to notice luck
To make us not be mistaken
Must read correctly
Three, fourteen, fifteen
Ninety two and six
To memorize the colors of the rainbow:
Each designer wants to know where to download Photoshop
And the beginning of the Mendeleev table:
Water native (Hydrogen) mixed with Gel (Helium) to Pour (Lithium). Yes, Bury and Lei (Beryllium) go to Pine Bor (Bor), where the Asian (Nitrogen) looks out from under the Corner of the native (Carbon), and with such an Acid mug (Oxygen) that Secondary (Fluorine) did not want to look. But we didn’t need it (Neon), so we moved Three (Sodium) meters and hit Magnolia (Magnesium), where Alyu in a mini (Aluminum) skirt was smeared with Cream (Silicon) with Phosphorus (Phosphorus) to stop be gray (sulfur). After that, Alya took Bleach (Chlorine) and washed the ship of the Argonauts (Argon)But why such a clear imperfection in such a perfect EI? Maybe thanks to the ability to forget the simplest facts, a person gets the freedom to combine fragments of his thoughts in an arbitrary wild order and find non-standard solutions? If so, then a strong AI is impossible. Either it will be forgotten as a person, or it will not be capable of non-standard solutions. In any case, the above assumptions imply the need to differentiate the goals of AI: one of the goals is the modeling of an ET, the other is the creation of a strong AI. Achieving one can exclude the achievement of another.
As we see, in the field of AI, there are too many questions with ambiguous answers, so it is not clear which direction to go. As happens in such cases, they try to move in all directions at once. At the same time, due to the absence of mathematically rigorous formulations, one has to turn to philosophy and artistic-literary modeling. One of the best-known examples in this area is the book “The Turing Choice” (1992) of one of the leading figures of AI, Marvin Lee Minsky and the famous science fiction writer Harry Harrison. I will quote from this book, perhaps explaining the mnemonic phenomenon described above:
human memory is not a tape recorder that records everything in chronological order. It is arranged in a completely different way - more like a carelessly kept file, equipped with a confusing and contradictory index. And not just confusing - from time to time we change the principles of classification of concepts.
An interesting interpretation of the tape metaphor in another literary work, the story of Stanislav Lem "Terminus" (from the series "Stories about the Pilot Pirks"). Here is the case of a kind of “intelligent tape recorder”: an old robot on an old spacecraft, which had been injured once in an accident, is engaged in ongoing repair work accompanied by tapping. But if you listen carefully, this is not just white technological noise, but a record of the Morse code - the negotiations of the crew members of the sinking ship. Pirks interferes in these negotiations and unexpectedly receives a response from the long-lost astronauts. It turns out that a primitive repair robot in some way keeps copies of their consciousness or is it a cognitive distortion of the perception of the pilot Pirks?
In another story, Ananke (from the same series), a copy of the UE in the space transport control computer leads to its paranoid overload with test tasks, which ends in a catastrophe.
In the story “Accident”, the too-anthropomorphically programmed robot perishes as a result of climbing ascent, which he decided to make in his free time. Do we need such performers? But the demons obsessed with watering the flower are also not always needed.
Some specialists in the field of AI do not like such “philosophizing” and “literary”, however these “philosophizing” and “literary” are traditionally inherent in analyzing the EI and are inevitable while the AI ​​is compared with the EI, and even more so while the AI ​​is trying to copy the EI.
In conclusion, a survey on a number of emerging issues.
Source: https://habr.com/ru/post/456080/
All Articles