Marvin Minsky's “The Emotion Machine”: Chapter 5, “Simulations and Predictive Machines”

How to introduce “human” into technologies and how technologies help to understand and improve and scale “human”?

The stern Marvin Minsky will help us with this, who, with his ruthless mind, analyzes feelings, emotions, pain, love and consciousness.

§5-8. The concept of "Simulation"

“Reality leaves a lot to the imagination”
- John Lenon

All of us can recognize an arch made up of rectangular blocks.


But also we can all imagine an arch whose upper rectangular block will be replaced by a block with three sides.


How can a program or mind “imagine” things that are not in scope? We could do this by “imagining” changes in an object at any stage of perception!

Making changes at a low level: Basically, we could create a new picture by changing every point on the retina - but in practice, such changes would require huge calculations. Also, if you want to shift your point of view, you would need to “calculate” the entire image again. Worse, before you can make such calculations, certain parts of the brain must know exactly which picture should be described. But in order to do this you first need to present this picture at some upper level, but if the picture is described, then why should it be calculated?

Making changes at intermediate levels: What can change is not the image of the image itself, but part of a high-level description of the image. For example, at the level of the Regions Finding systems, it would be possible to change the name of the leading edge of the upper block of the arch from “rectangular edge” to “triangular edge”. However, this can cause problems at other levels, because the edges of the triangle will not have the same proportions as the edges of the areas next to it.


Later we will see that it is better to replace the whole block at the highest level of the Object Finding system.

Visualizer: I am having problems like the ones described above. When I try to imagine a triangular shape, I know where the three lines of the figure should appear, but I see them as foggy, luminous stripes whose fuzzy ends are often not found. When I try to correct this by “pushing” the line — it moves sharply with some constant speed that I cannot change, and when I tell this line to freeze, it still continues to move (although, oddly enough, it never goes far)

You tried to change the description of the object, but you could not save the necessary relationships between the parts of the object. Imagination is sometimes similar to vision, except that when you change the internal representation of an object, it cannot maintain its consistency. A real object cannot move at two different speeds at the same time, just like two real lines cannot intersect and not meet, but the imagination can allow such things.

Making changes at higher semantic levels: You can imagine changing the top block of the arch by changing the name of the shape of the top block itself, exactly changing the “rectangle” to “triangle” in the arch definition itself: “A triangular block supported by two vertical blocks”.

Now think how effective it is! To make such a change at the touch level, you will need to change the behavior of thousands of “pixels” (the data units that make up the picture), while you only need to change one word when you work at the abstract language level to represent any thing using only one or more compact characters. Of course, these symbols are useless until they are associated with structures that give more details or “meaning” to what you see.

Our Builder system could do similar tricks by making changes in the so-called “Semantic Networks”. For example, she could imagine an arch of three blocks by describing the relationship between the three blocks. [13] Then, in order for the Builder to present a triangular block at the top of the arch, he needs to change only one connection:


Making such changes at earlier levels would involve too much detail. If you used the replacement of recorded images similar to images, then it would be difficult for you to change any part of the already recorded image. But at the highest “semantic” level, you can more easily make meaningful changes. For example, when you describe a “lying block supported by two vertical blocks”, you don’t need to describe the vision of this block from the perspective of the observer. Therefore, a similar description applies equally to the following set of pictures:


If we replace the word “block” with the word “object”, then our network will be able to describe even more situations.

This shows how convenient the "abstract" descriptions. Sometimes the word “abstract” is used to mean an “intellectually complex” task, but in this case the word “abstract” has the exact opposite meaning: an abstract description is simpler because it suppresses various details of the description that are irrelevant. Of course, the descriptions should not be too abstract: when you ask someone for advice, and he gives an answer like "If you want something, then do what will give you this."

We discussed how we can present visual pictures by constructing “simulations” in our mind. We also do similar things in other areas. Perhaps some chefs present new tastes by directly changing the state of low-level sensory systems, and perhaps some composers likewise present the sounds of new instruments; however, such thinkers can also achieve similar results by making small changes to higher levels of presentation, thereby causing delight or disgust without building low-level details of the dishes or music presented.

Dramatic critic: I can clearly remember how I felt after participating in a certain performance, but I can’t remember any details about what this terrible game was about.

To discuss this, we must use the new word by combining the words “simulation” and “stimulus”. Simulation is a model perception of an object, which is caused by a change in the mental representation of the object. Thus, in the discussion of the Challenger in §4-7, we saw how one could use a simulation of defeat to evoke feelings of anger. In order to do this, it was enough to imagine a mockery on the face of our enemy without any features of this person, since we are in a position to make the simplest descriptions of phenomena using high-level abstractions.

Visualizer: When I think of my cat, her image is presented with so many details that I can imagine every hair on her body. Wouldn't it be our advantage to create a real image indistinguishable from the original? [14]

Perhaps when you first remember your cat, its surface has only a “woolly” structure, and only when you look at the picture more clearly do you begin to add more details to your mental representation. However, this can happen so quickly that you simply do not have time to notice this transition, and it begins to seem to you that you initially saw everything that way. Your experience may be one of the descriptions of illusions that we introduced in §4:

The illusion of immanence: When your questions are answered before you have time to ask them, it seems to you that you already knew about these answers.

The illusion of Immanence manifests itself not only when we present something; we never see “real images” because we do not perceive most of the small details until our mind asks for these details. In fact, recent experiments have suggested that our internal representation of visual images is rarely updated in real time. [15] In chapters §6 and §8 we describe a scheme called “Panalogy” that will help explain how our brain can isolate and work out the requested details in an incredibly fast manner.

§ 5-9. Predictive Cars

William James: “Try to imagine how you bend your finger, and at the same time hold it straight. After some time, he will begin to tingle and an imaginary change in position will be felt; however, the finger will remain in place because the thought “finger remains straight” also holds in your mind. Then drop these ideas and think only of some pure movement and voila! Your finger will take the required place without any effort on your part. ” - The Principles of Psychology, 1890, p. 527 [The Principles of Psychology, 1890, p527.]

Everyone can think of things without further action, just like Carol, who imagined moving blocks. But how does she manage to do this? You can now close your eyes, lean back and treat yourself to dreams and fantasies, or try to predict what may happen in the future.

Now we can create a machine that can do similar things, predicting the various outcomes of specific situations. Let's assume that it has rules like the following:


Then we will give our car, let's call it the Seer, a way to replace what she sees at a particular point in time, which is realized thanks to forecasts made based on the above rule. In this case, when the Seer is in situation A, and then considers action X, this will force the "seer" of the seer that he is in situation B.


I included the Suppression Band group for two reasons. The first is when the Seer represents future state B, we don’t want it to quickly replace the existing state A. Second, we still don’t want the Seer to perform step X, because he can still consider some other options for the future, before making a final decision . Thus, the Seer can use these suppressive stripes to abstract from the outside world, which will allow him to “stop and think” what action he should best do. [16]

By repeating such an operation, the Seer can use the predictive chains to simulate the actions taking place in the "virtual world". Of course, in order for the Seer to make such predictions, he must use the search methods described in §5-3 to simulate (and then compare) the effects of various actions before accepting one of them. This will require additional memory from him, as well as a large number of other mechanisms. Nevertheless, anyone who has played modern computer games can be convinced from personal experience how much the art of creating virtual worlds inside machines has advanced.

I expect that in the next few years we will find the structures described in the upper diagram inside our brain. How has our brain developed such abilities? The species of primates that preceded us should have had similar structures, thanks to which they could think through several future steps. But then, several million years ago, this system seems to have begun to develop rapidly, since the frontal lobes of our brain developed their current size and architectural complexity; apparently this was a key stage in the development of human intelligence.

Summary

This chapter described some structures and processes that can do the same things as a person. We noted a set of levels that we can use to gradually increase the ways of thinking.


However, we assumed a lot of detail about what is going on at each of these levels. Further, I will assume how this system works at each specific level, constantly responding to specific problems arising by switching to more appropriate Ways of Reflection. We will present this Mind Model using the simplest diagram:

Critical Selector Mind Model

In the remainder of this book, we will often switch between these two models of the mind, because each model gives the best answers to specific questions about ourselves. The Six model provides a better distinction between different levels of mental behavior, and the Critical-Selector model gives a better understanding of how we can solve complex problems.

Chapter 7 will combine these two models, because we often use different Critics and Selectors at each cognitive level.



However, no matter how this system is built, it will never seem complete until it knows the world in which it is located. In particular, she must be able to foresee some possible outcomes of actions, and she will not be able to do this until she has the appropriate types of knowledge. In humans, this knowledge is called "common sense" and reasoning. And although in everyday life this phrase means “things that most people find obvious,” the next chapter will demonstrate that “common sense” is surprisingly extremely complex.

---

Thanks for the translation, Stanislav Sukhanitsky. Who wants to help with the translation - write in a personal email or mail magisterludi2016@yandex.ru

about the author

Marvin Lee Minsky (born Marvin Lee Minsky; August 9, 1927 - January 24, 2016) is an 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 friends with the critic Harold Bloom of Yale University, who spoke of him as nothing more than "the sinister Marvin of 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 Douglas Adams cycle of hitchhikers in the galaxy and the movie 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 Minsky, the dog is named the main character in the movie Tron: Legacy. [ Wikipedia ]

---