Marvin Minsky "The Emotion Machine": Chapter 5 Thinking
How to bring "human" technology and how technology helps to understand and improve and scale the "human"?
This will help us harsh Marvin Minsky, who with his merciless mind analyzes feelings, emotions, pain, love and consciousness.
')
When Joan chose what to do - run across the road or run back, she had to choose one of the following rules:
If you're on the roadway, Then run away from her
If you are on the roadway, Then quickly run across it
However, in order for Joan to make these decisions, she needs some kind of mechanism that would predict and compare possible options for resolving this situation. What helps Joan make such predictions? The simplest way is to carry with you a collection of three-part rules “If -> That -> In such a case”, in which each If describes a specific situation, each That describes a specific action, and each Then describes a possible result of the work done.
If you are on the roadway, Then run back, In this case, try to cross the road a little later
If you are on the roadway, Then move on to it, In this case, you will arrive much earlier
If you are on the roadway, Then go, In this case, you can get wounds
But what if several rules can be applied to a particular situation? In this case, you can select the necessary rule after comparing the results that they predict:
Thus, these threefold rules allow us to build experiments in our head, before we begin to act in full danger of the world around us; we can mentally “imagine the result before making the jump” and choose the most attractive alternatives. For example, suppose Carol plays with cubes and talks about the construction of a three-block arch:
Right now she has three blocks, which are as follows:
So, she imagines a plan for building this arch: initially she will need a place to erect the base of the arch, which she can find using the following rule: if the block is lying and if she puts it on its side, then it will occupy less space on the ground.
Then she can put two short blocks on the ribs, making sure that they are at the correct distance from each other, and then finally put a long block on them flat. We can present a sequence of actions, describing them as a sequential change of frames in the video.
To imagine a four-step sequence of actions, Carol needs to have quite a lot of skills. First, her visual systems must describe the shape and location of these blocks, with some of their parts may be out of sight, and she will need the ability to plan which block to move and where to put it. Then, whenever she blocks, she has to program her fingers to grab these blocks, then move the blocks to the right place, and finally release them, controlling, so that her hand does not collide with her body, or face, or another unit that is already in place. She also needs to control the speed in order to be able to put the block on top of two blocks, trying not to drop the supporting blocks.
Carol: None of these actions seemed problematic to me. I just imagined the arch in my mind and understood how each of the blocks should stand. Then I just had to put two blocks perpendicularly up (when I was convinced that they were at the right distance from each other), and then put a long block on top of these two. In the end, I did things like that. Perhaps I remembered my experience and simply repeated it.
But how could Carol "imagine" how the figure would look after moving the block before it directly moves the block?
Programmer: We know how to teach a computer to do such things; we call it "physical simulation." For example, at each stage of the design of a new aircraft, our programs can accurately predict the force acting on each point of the aircraft when it moves in airspace. In fact, we can do it so well that we will build a plane from scratch on a computer, and then we will recreate it in reality - it will fly.
No human brain can do such complex calculations, but we can still make useful predictions using the rules "If -> That -> In this case." For example, when Carol planned to build this arch, she could imagine a step in which she places a long block on top of short blocks:
Of course, this action will end in failure, because the top block will fall. However, after Carol gains more experience, she will also learn to predict that the top unit in this situation will fall down.
Please note, you can also use these rules “in reverse order” to explain what happened in the previous steps! Thus, if you see a fallen block (A), you can guess that the previous state was (B):
Student: I wonder if using these rules would be practical? It seems to me that in order to fulfill these plans, Carol would need a huge amount of rules “If -> That -> In this case”. For if each of the three blocks would have thousands of different forms, then Carol would need to use thousands of different rules.
In fact, if we make the rules “If -> That -> In such a case” too specific, then they will apply only to a few situations. This means that our rules should not take into account too much detail, but take into account more abstract ideas. Thus, the rule applicable to a physical object should describe this object in a certain non-visual form, which would not change if a particular object changes its visual form. It is naive that most of us tend to believe that we “present” specific scenes as well-defined pictures. However, in Section 5-8, we show that all such models are illusions, because these images do not behave like real pictures.
Consider that in the physical world, when you think about capturing or lifting a block, you do not take into account its weight, and predict that if you release it, it will fall. In the economic world, if you pay for the purchase, then you will have the purchased item, otherwise you will have to give it back. In the world of communication, when you say something, then the audience can remember it, but this fact will happen with much greater probability if you say that this particular information is important.
Every adult knows a lot of such things and sees them as obvious, sensible knowledge, but every child needs to learn these things for years, until he understands how they operate in different worlds. For example, if you move an object in the physical world, then the object will change its place of position, but if you give any information to your friend, then it will already be in two places at once. We will consider these things in more detail in Chapter §6. [1].
By linking two or more rules “If -> That -> In such a case” into a chain, we can imagine what will happen in a few completed actions, and thus predict what will happen in the future, but only if we are able match previous and next If. For example, if you are in situation P and want to be in situation Q, you may already know the necessary rule: If P -> Do something A -> In this case, Q. But what if you don’t know this rule? In this case, you can search for such a rule in memory that would connect P and Q through some intermediate chain S.
If P -> Do A -> In this case, S and then If S -> Do In -> In this case, Q
Then, if you cannot find the necessary two-part chain, then you can start looking for longer action chains. It is quite clear that most of our reasoning is based on finding such “chains of reasoning” and as soon as you learn how to use this process, you will be able to resolve more complex problems and look at several steps forward. For example, you often think:
If I ask Charles to give me a ride to the store, he can answer me with a “yes” or “no”. If he answers “yes”, everything will be fine, but if he answers “no”, then I will offer him some kind of reward and this will probably help him to change the decision.
However, when you need to anticipate events for a greater number of steps, the search for such chains may become too complicated - because the complexity of such calculations grows exponentially, like a densely branching tree. Thus, if each branch leads only to two alternatives, and you need to predict the next 20 steps, you will have to look through about a million of these paths, because so many branches will give a sequence of 20 choices.
However, there is a trick that could make the search more streamlined. For if there is a 20-step path from A to B, then there should be a path only from 10 steps, from each end! Thus, if we start the search at both ends at once, in this case we can meet in the middle at the point M.
On the left side of this tree will be about 1000 branches, if this is also true for the part on the right, in which case the search will be several hundred times smaller. And then, if you somehow find out where point M should be, then you can speed up the search even further by dividing each side into a sequence of 5 steps.
If this works, then your search will be several thousand times smaller! However, none of the above techniques seem to work in practice, because they assume that each “reverse” search will encounter only two branches, and this is not always the case. However, if M's guess was incorrect, you can still try other similar guesses, and even you will not be lucky 50 times in a row, using this technique you would still spend far less time and effort if you were to go through all the options!
This example demonstrates why having plans is good. If you can predict some of the “steps” in the way of solving an extremely difficult problem - then you will be able to consistently solve much less complex problems! Therefore, every attempt to "separation and rule" can make the difficult task simple. In the early years of artificial intelligence, when most programs were based on "trial and error," many researchers tried to find technical methods that allowed them to reduce the number of attempts to achieve a result. Now, however, it seems that the more important task is to find precisely these “steps”, because it is thanks to them that our common sense proceeds. Chapter 6 claims that our most powerful ways of resolving problems arise from inventing good analogies, with which we know how to work, to the problem to be solved.
Whenever we solve a problem, we use a variety of thinking techniques. Some of them have their own names, such as planning and rationale, but most do not have their own names. Some of these methods seem to be formally described and neat, while others are more “intuitive.”
For example, we constantly use prediction chains in a way that resembles the following: If A implies B, and B implies C, then we can say with absolute certainty that A implies C. And if all of the premises are true, just like our logical thinking - in In this case, this conclusion will be true, and we will never make a tragic mistake.
However, in real life, it seems most of the assumptions are erroneous, because the "rules" on which they are based always have exceptions. This means that there is a difference between the rigid methods of logic and the chains of various forms of common sense used daily.
We all know that the physical chain is as strong as its weakest link. But mental chains are even more fragile, because each new link weakens the next!
So using logic is something like walking on a board. She assumes that every single step is correct, while common sense should be provided with more support — it is necessary to bring new evidence for it after several steps have passed. And the weakness of the chain increases with the growth of its length - as each additional link increases the chance of breaking it. That is why, when people state their arguments, they often interrupt themselves to enter a lot of evidence or analogies in their messages - they feel the need to support the present thinking step before moving on to the next step of thinking.
Building long chains of reasoning is one of the ways to predict the future situation - in Chapter 7 a large number of other means will be discussed. I suspect that when we encounter problems in everyday life, we tend to use several different methods of prediction, clearly realizing that each of them has its drawbacks. But since each of these methods has different disadvantages, we can combine these methods to use their joint strengths.
Everyone has many ways to make short-term plans, compare available scenarios and apply everyday methods of reasoning - and we can do it so easily that we can hardly think about these processes. However, when these processes do not fulfill their functions, and we need to change our methods of short-term predictions - then we start thinking about what we are doing - and this is called reflexive thinking .
For the translation, thanks to Stanislav Sukhanitsky. Who wants to help with the translation - write in a personal or mail magisterludi2016@yandex.ru
Marvin Lee Minsky (Eng. Marvin Lee Minsky; August 9, 1927 - January 24, 2016) - American scientist in the field of artificial intelligence, co-founder of the Laboratory of artificial intelligence at the Massachusetts Institute of Technology. [ Wikipedia ]
Interesting Facts:
This will help us harsh Marvin Minsky, who with his merciless mind analyzes feelings, emotions, pain, love and consciousness.
')
§5-3. Thinking
When Joan chose what to do - run across the road or run back, she had to choose one of the following rules:
If you're on the roadway, Then run away from her
If you are on the roadway, Then quickly run across it
However, in order for Joan to make these decisions, she needs some kind of mechanism that would predict and compare possible options for resolving this situation. What helps Joan make such predictions? The simplest way is to carry with you a collection of three-part rules “If -> That -> In such a case”, in which each If describes a specific situation, each That describes a specific action, and each Then describes a possible result of the work done.
If you are on the roadway, Then run back, In this case, try to cross the road a little later
If you are on the roadway, Then move on to it, In this case, you will arrive much earlier
If you are on the roadway, Then go, In this case, you can get wounds
But what if several rules can be applied to a particular situation? In this case, you can select the necessary rule after comparing the results that they predict:
Thus, these threefold rules allow us to build experiments in our head, before we begin to act in full danger of the world around us; we can mentally “imagine the result before making the jump” and choose the most attractive alternatives. For example, suppose Carol plays with cubes and talks about the construction of a three-block arch:
Right now she has three blocks, which are as follows:
So, she imagines a plan for building this arch: initially she will need a place to erect the base of the arch, which she can find using the following rule: if the block is lying and if she puts it on its side, then it will occupy less space on the ground.
Then she can put two short blocks on the ribs, making sure that they are at the correct distance from each other, and then finally put a long block on them flat. We can present a sequence of actions, describing them as a sequential change of frames in the video.
To imagine a four-step sequence of actions, Carol needs to have quite a lot of skills. First, her visual systems must describe the shape and location of these blocks, with some of their parts may be out of sight, and she will need the ability to plan which block to move and where to put it. Then, whenever she blocks, she has to program her fingers to grab these blocks, then move the blocks to the right place, and finally release them, controlling, so that her hand does not collide with her body, or face, or another unit that is already in place. She also needs to control the speed in order to be able to put the block on top of two blocks, trying not to drop the supporting blocks.
Carol: None of these actions seemed problematic to me. I just imagined the arch in my mind and understood how each of the blocks should stand. Then I just had to put two blocks perpendicularly up (when I was convinced that they were at the right distance from each other), and then put a long block on top of these two. In the end, I did things like that. Perhaps I remembered my experience and simply repeated it.
But how could Carol "imagine" how the figure would look after moving the block before it directly moves the block?
Programmer: We know how to teach a computer to do such things; we call it "physical simulation." For example, at each stage of the design of a new aircraft, our programs can accurately predict the force acting on each point of the aircraft when it moves in airspace. In fact, we can do it so well that we will build a plane from scratch on a computer, and then we will recreate it in reality - it will fly.
No human brain can do such complex calculations, but we can still make useful predictions using the rules "If -> That -> In this case." For example, when Carol planned to build this arch, she could imagine a step in which she places a long block on top of short blocks:
Of course, this action will end in failure, because the top block will fall. However, after Carol gains more experience, she will also learn to predict that the top unit in this situation will fall down.
Please note, you can also use these rules “in reverse order” to explain what happened in the previous steps! Thus, if you see a fallen block (A), you can guess that the previous state was (B):
Student: I wonder if using these rules would be practical? It seems to me that in order to fulfill these plans, Carol would need a huge amount of rules “If -> That -> In this case”. For if each of the three blocks would have thousands of different forms, then Carol would need to use thousands of different rules.
In fact, if we make the rules “If -> That -> In such a case” too specific, then they will apply only to a few situations. This means that our rules should not take into account too much detail, but take into account more abstract ideas. Thus, the rule applicable to a physical object should describe this object in a certain non-visual form, which would not change if a particular object changes its visual form. It is naive that most of us tend to believe that we “present” specific scenes as well-defined pictures. However, in Section 5-8, we show that all such models are illusions, because these images do not behave like real pictures.
Consider that in the physical world, when you think about capturing or lifting a block, you do not take into account its weight, and predict that if you release it, it will fall. In the economic world, if you pay for the purchase, then you will have the purchased item, otherwise you will have to give it back. In the world of communication, when you say something, then the audience can remember it, but this fact will happen with much greater probability if you say that this particular information is important.
Every adult knows a lot of such things and sees them as obvious, sensible knowledge, but every child needs to learn these things for years, until he understands how they operate in different worlds. For example, if you move an object in the physical world, then the object will change its place of position, but if you give any information to your friend, then it will already be in two places at once. We will consider these things in more detail in Chapter §6. [1].
Planning and Search
By linking two or more rules “If -> That -> In such a case” into a chain, we can imagine what will happen in a few completed actions, and thus predict what will happen in the future, but only if we are able match previous and next If. For example, if you are in situation P and want to be in situation Q, you may already know the necessary rule: If P -> Do something A -> In this case, Q. But what if you don’t know this rule? In this case, you can search for such a rule in memory that would connect P and Q through some intermediate chain S.
If P -> Do A -> In this case, S and then If S -> Do In -> In this case, Q
Then, if you cannot find the necessary two-part chain, then you can start looking for longer action chains. It is quite clear that most of our reasoning is based on finding such “chains of reasoning” and as soon as you learn how to use this process, you will be able to resolve more complex problems and look at several steps forward. For example, you often think:
If I ask Charles to give me a ride to the store, he can answer me with a “yes” or “no”. If he answers “yes”, everything will be fine, but if he answers “no”, then I will offer him some kind of reward and this will probably help him to change the decision.
However, when you need to anticipate events for a greater number of steps, the search for such chains may become too complicated - because the complexity of such calculations grows exponentially, like a densely branching tree. Thus, if each branch leads only to two alternatives, and you need to predict the next 20 steps, you will have to look through about a million of these paths, because so many branches will give a sequence of 20 choices.
However, there is a trick that could make the search more streamlined. For if there is a 20-step path from A to B, then there should be a path only from 10 steps, from each end! Thus, if we start the search at both ends at once, in this case we can meet in the middle at the point M.
On the left side of this tree will be about 1000 branches, if this is also true for the part on the right, in which case the search will be several hundred times smaller. And then, if you somehow find out where point M should be, then you can speed up the search even further by dividing each side into a sequence of 5 steps.
If this works, then your search will be several thousand times smaller! However, none of the above techniques seem to work in practice, because they assume that each “reverse” search will encounter only two branches, and this is not always the case. However, if M's guess was incorrect, you can still try other similar guesses, and even you will not be lucky 50 times in a row, using this technique you would still spend far less time and effort if you were to go through all the options!
This example demonstrates why having plans is good. If you can predict some of the “steps” in the way of solving an extremely difficult problem - then you will be able to consistently solve much less complex problems! Therefore, every attempt to "separation and rule" can make the difficult task simple. In the early years of artificial intelligence, when most programs were based on "trial and error," many researchers tried to find technical methods that allowed them to reduce the number of attempts to achieve a result. Now, however, it seems that the more important task is to find precisely these “steps”, because it is thanks to them that our common sense proceeds. Chapter 6 claims that our most powerful ways of resolving problems arise from inventing good analogies, with which we know how to work, to the problem to be solved.
Cause and Reliability
Whenever we solve a problem, we use a variety of thinking techniques. Some of them have their own names, such as planning and rationale, but most do not have their own names. Some of these methods seem to be formally described and neat, while others are more “intuitive.”
For example, we constantly use prediction chains in a way that resembles the following: If A implies B, and B implies C, then we can say with absolute certainty that A implies C. And if all of the premises are true, just like our logical thinking - in In this case, this conclusion will be true, and we will never make a tragic mistake.
However, in real life, it seems most of the assumptions are erroneous, because the "rules" on which they are based always have exceptions. This means that there is a difference between the rigid methods of logic and the chains of various forms of common sense used daily.
We all know that the physical chain is as strong as its weakest link. But mental chains are even more fragile, because each new link weakens the next!
So using logic is something like walking on a board. She assumes that every single step is correct, while common sense should be provided with more support — it is necessary to bring new evidence for it after several steps have passed. And the weakness of the chain increases with the growth of its length - as each additional link increases the chance of breaking it. That is why, when people state their arguments, they often interrupt themselves to enter a lot of evidence or analogies in their messages - they feel the need to support the present thinking step before moving on to the next step of thinking.
Building long chains of reasoning is one of the ways to predict the future situation - in Chapter 7 a large number of other means will be discussed. I suspect that when we encounter problems in everyday life, we tend to use several different methods of prediction, clearly realizing that each of them has its drawbacks. But since each of these methods has different disadvantages, we can combine these methods to use their joint strengths.
Everyone has many ways to make short-term plans, compare available scenarios and apply everyday methods of reasoning - and we can do it so easily that we can hardly think about these processes. However, when these processes do not fulfill their functions, and we need to change our methods of short-term predictions - then we start thinking about what we are doing - and this is called reflexive thinking .
For the translation, thanks to Stanislav Sukhanitsky. Who wants to help with the translation - write in a personal or mail magisterludi2016@yandex.ru
Table of Contents of The Emotion Machine
Introduction
Chapter 5. LEVELS OF MENTAL ACTIVITIES
Chapter 6. COMMON SENSE
Chapter 7. Thinking.
Chapter 8. Resourcefulness.
Chapter 9. The Self.
Chapter 1. Falling in Love
Chapter 2. ATTACHMENTS AND GOALS
Chapter 3. FROM PAIN TO SUFFERING
Chapter 4. CONSCIOUSNESSChapter 5. LEVELS OF MENTAL ACTIVITIES
Chapter 6. COMMON SENSE
Chapter 7. Thinking.
Chapter 8. Resourcefulness.
Chapter 9. The Self.
about the author
Marvin Lee Minsky (Eng. Marvin Lee Minsky; August 9, 1927 - January 24, 2016) - American scientist in the field of artificial intelligence, co-founder of the Laboratory of artificial intelligence at the Massachusetts Institute of Technology. [ Wikipedia ]
Interesting Facts:
- Minsky was a friend of critic Harold Bloom from Yale University (Yale University), who spoke of him as “sinister Marvin Minsky”.
- Isaac Asimov described Minsky as one of two people who are smarter than himself; the second, in his opinion, was Karl Sagan.
- Marvin is a robot with artificial intelligence from the cycle of Douglas Adams novels Hitchhiker's Guide to the Galaxy and Hitchhiker's Guide to the Galaxy (film).
- Minsky has a contract to freeze his brain after death in order to be “resurrected” in the future.
- In honor of Minsk named the dog of the protagonist in the movie Tron: Legacy. [ Wikipedia ]
About #philtech
#philtech (technology + philanthropy) is an open, publicly described technology that aligns the standard of living of as many people as possible by creating transparent platforms for interaction and access to data and knowledge. And satisfying the principles of filteha:
1. Opened and replicable, not competitive proprietary.
2. Built on the principles of self-organization and horizontal interaction.
3. Sustainable and prospective-oriented, and not pursuing local benefits.
4. Built on [open] data, not traditions and beliefs.
5. Non-violent and non-manipulative.
6. Inclusive, and not working for one group of people at the expense of others.
Philtech's social technology startups accelerator is a program of intensive development of early-stage projects aimed at leveling access to information, resources and opportunities. The second stream: March – June 2018.
Chat in Telegram
A community of people developing filtech projects or simply interested in the topic of technologies for the social sector.
#philtech news
Telegram channel with news about projects in the #philtech ideology and links to useful materials.
Subscribe to the weekly newsletter
#philtech (technology + philanthropy) is an open, publicly described technology that aligns the standard of living of as many people as possible by creating transparent platforms for interaction and access to data and knowledge. And satisfying the principles of filteha:
1. Opened and replicable, not competitive proprietary.
2. Built on the principles of self-organization and horizontal interaction.
3. Sustainable and prospective-oriented, and not pursuing local benefits.
4. Built on [open] data, not traditions and beliefs.
5. Non-violent and non-manipulative.
6. Inclusive, and not working for one group of people at the expense of others.
Philtech's social technology startups accelerator is a program of intensive development of early-stage projects aimed at leveling access to information, resources and opportunities. The second stream: March – June 2018.
Chat in Telegram
A community of people developing filtech projects or simply interested in the topic of technologies for the social sector.
#philtech news
Telegram channel with news about projects in the #philtech ideology and links to useful materials.
Subscribe to the weekly newsletter
Source: https://habr.com/ru/post/359050/
All Articles