🤧 🤴🏽 👤 Why are there bugs in the brain? 😉 ✊ 👩🏿‍🤝‍👩🏼

When we use any program, we have two reasons for getting the wrong results:

Incorrect / insufficient input. If the program receives nonsense at the input, we will also receive nonsense at the output, even if everything worked perfectly inside. This concept is called Garbage In, Garbage Out . We cannot fix this - the problem is not in the program.
From the correct data, we got the wrong answer. Then we consider that there is a bug in the program and try to repair it.

The same can be said about us.

Sometimes the necessary information is missing and we cannot prevent the error.
For example, we give an estimate of the time to solve a list of tasks. In the process, it turns out that the list should have been 3 times larger. It is expected that our first assessment will be erroneous. Nothing depended on us here. Garbage In - Garbage Out.

But if we have already observed 20 times in a row how the number of tasks grows 3 times from the initial one, it is strange to assume that everything will go like clockwork for the 21st time.

If my program yielded a result 20 times in a row that did not coincide with reality, I would assume that something was wrong with it. Why do I need it if, when using it, I still get the wrong answer? It should take into account the fact that the number of tasks can grow. It must be repaired and adapted to such working conditions.

When I observe such an error for myself - I draw the same conclusion.

The situation described above is called the “Planning Mistake” . She is one of the items on the Cognitive Distortion list . There are many articles , books , resources on this subject.- but not one of them answered my simple question: “Where do the errors come from and how best to fix them?”. What is wrong with the brain that it predictably produces incorrect results? How to roll a patch on it? If this is not possible, then where to back up with crutches?

I began to study the basic principles of the brain to find out at what stage errors occur, what are their causes and how best to repair them.

What we work with

To begin with, I will make a reservation that the device of the brain is very different from PC. There is nothing approximately reminiscent of SSD, RAM and CPU - it is not there. The closest analogue is a video card or computing cluster.

Each neuron is a small computer. It accepts a bunch of input signals and, based on their totality, calculates when and how to give output signals. There are approximately 86 billion such pieces in the human brain, each with thousands of compounds (synapses), each connection with its own parameters ( more about the main features of the work of neurons ).

In short, if we imagine a video card with 3d cores of 86 billion in terms of cores, where each core is connected to thousands of others, we get a very crude model of the brain that works in a completely different way. Well, CPUs are not good at all.

By the way, this is the answer to the question why ANNs learn faster on a video card. It is not difficult to model a separate neuron, but we need to calculate a huge number of them. The solution is to count in parallel.

But what about memory?

And the memory in such an architecture is a very interesting thing. Instead of storing any “addresses” and reading blocks, the brain stores information on how to calculate the response to the signal. This is more like a Google request than reading from an SSD. Cons - you can not go to the address directly and quickly read the data. Pros - you can recall "the film in which a guy with a beard, in a red and gold metal suit, saved the world." ( More about memory )

You might have a question related to why we perceive our thoughts not as a bunch of processes taking place in parallel, but in sequence. This is due to the features of the prefrontal cortex and attention. I will write a separate article about them. So far, we do not need them, but we will need them to explain such a phenomenon as “ flow ”.

How do errors occur?

The problem is that an architecture sharpened to solve classification problems, process images and sounds, find dependencies and recover information from incomplete data will be ineffective in some other cases. I took apart the basic mechanisms, and tried to describe where and what could go wrong.

Strength of Communication and Memory

Bond strength effect

The states of most of the connections in the brain responsible for memory return to their original state for hours / days . When we encounter a problem that we have encountered for a long time - we have very little chance of extracting information about it. But, if we recently encountered such a problem - the connections are incredibly strong. So much so that we will see this problem everywhere, even if it occurs 1 time per million and is not worth that much attention. And if she accidentally meets a second time, then her priority will rise above the Tesla Roadster launched by SpaceX .

I think you have already associated this information with a planning error. Yes, we tend to consider only fresh experience. To look further - we need to make a lot of effort.

Want more examples?
How do you like the people who ask questions at the interview, the answers to which they learned a couple of days before?
Hype waves subsiding every couple of months and alternating with new ones?
And the desire to buy a bunch of unnecessary things, just because yesterday would they be welcome? And I don’t care that the situation of yesterday happened for the first time, and whether they will ever repeat it.

Correlation is not causality

The second fun consequence of link architecture is the lack of a difference between causality and correlation at low levels of the brain. In other words, if I often drop the keys in front of my door - the brain can go the path of least resistance and decide that my door has this property - to force me to drop the keys.

And when the brain makes a conclusion, he tries to remember it and maintain the strength of the connections.

False understanding

The two previous features give rise to an interesting, self-sustaining process:
If someone else drops a key near my door, this may be taken as a confirmation of my conclusions. Although in reality this is just an accident. Each time we observe something that matches the pattern, its priority increases and communication with it intensifies. As a result, we have a chance to start customizing everything around our “key and door theory”.

There is nothing wrong with brains. Yes, they are imprisoned for finding patterns, and yes - they sometimes get the wrong result. You just need to remember that some of our templates are wrong and be able to recognize it.
The situation is complicated by the fact that the brain has already built connections and does not give us an empty answer. We will not have the feeling that we "did not understand anything." But the model that results will be incorrect. Moreover, since we already have this model, the brain will choose it, and not try to build a new one.

How to fix it?

Mistakes associated with loose connections, such as planning errors, are fairly straightforward to fix. You need to make a habit of recording information on some stable medium and update your brain if necessary. For example, every time new tasks fall into the project and the time frame increases - write down. Next time, before planning, we can see how many times the list can become larger.

The error of correlation and causality can be corrected at the time of its occurrence. It is enough to try to build a causal relationship, once again dropping the key in front of the door. So you can understand that in fact - the reason is that you get the key from a small pocket, and not at all the magical properties of the door.

The problem is that when these errors turn into a false understanding - thinking that something is going wrong is quite difficult. We are already accustomed to assume that the door causes the key to fall - why should it be further analyzed? Does the key in front of the door fall? Is falling. Does the circuit work? Works. Why touch her?

To detect that my brain heated me somewhere, and slipped the illusion of understanding, I use this technique:
I try to describe what I had to understand when translating into another system of terms. In fact - using analogies. For example, if I can describe what Reactive Programming is using terms from mathematics, or by the example of how water flows, I most likely understood it (this concept is perfectly described by teapots connected by pipes - I explained this to a girl).

The second option is less pleasant. Our template should break, as a result of the fact that we sat down in a puddle. For example, we brought all our friends to the “magic door”, but no one’s key fell. And with us too. There are ways to try to leave the template, for example: “You did not believe enough in the magic of the door!” Therefore, she did not work! ” - we will talk about such a mechanism further.

I have prepared a list of cognitive distortions that work according to the scheme described above. It is quite large, but you do not need to remember it. If you have nothing to do and want to read something, open it. There are many interesting examples of errors, the correction of which allows you to slightly improve your life. Inside, instructions on how to use it better.
Link, in case you want to bookmark it.

Cognitive Distortion List (Memory and Communication):

.
.
, - .
, .

. — . — , , , . — .

.
-.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
().
.
.
.
.
.

It would be just fine if the list ended there, but we have another big layer of problems.

Deferred learning and ignoring

For training - some kind of feedback is needed. Its source may be someone who adjusts the process - then this is training with a teacher. Or we can learn from some information, independently evaluating the results - this is training without a teacher.

If we try to make a program that distinguishes cats from dogs, we make her guess what picture she was given to enter. When a program is mistaken, we try to make changes to its structure (for example, to weaken connections for active “neurons” in the ANN). When all is well, we must fix the current state (for example, strengthen active connections).
If we give an answer about the correctness of recognition ourselves - we are a teacher. If we built the program so that every time it makes an assumption, it independently checks it - we have training without a teacher.

The brain is an example of a system that can learn on its own.

External feedback

Let's go back many years when brains solved simple problems. For example: "How to get a coconut from a palm tree." Suppose we came to the conclusion that we need to shake the palm, the coconut will fall and we will have breakfast. We shake a palm tree, coconut falls on our heads. Now, instead of breakfast, we have a huge lump on our head and it hurts.

We received feedback from our actions. The sensation of a coconut blow on the head was directly related to the decision to shake the palm. Next time we will try to come up with something else.

Now, back to our time. All week we worked on Important Tasks. We managed to complete everything by Friday. In the evening we went to a bar, then had a cool weekend, went out into the countryside, and worked out our favorite hobby.

Arriving at work on Monday, we see a colleague who tells us from the threshold: “What the hell are you doing?”

Bam! An imaginary coconut has just fallen on us. The mechanism is about the same: we were going to calmly go to work, have tea, sit down at tasks. Instead, some *** knocks us all plans, and is going to run into us! Here is the radish, huh? We must besiege this guy to drink tea, and normally proceed to the tasks. Doesn’t he understand that we have a deadline on our nose, now is not the time to suffer from such garbage at all? You need to work!
We are responding with something sharp so that they are behind us.

At that moment in that guy’s head: “Oh my God, he is an impassable idiot. In his task, which he made on Tuesday, a gross mistake that will cost us dearly. But he doesn’t want to hear about it! ” - Note, now our colleague was hurt by a mental coconut. - “I’ll tell this upstart that because of his unwillingness to listen to me he will ruin the entire project, maybe even listen?”

Coconut exchange can last quite a while.

How does this situation differ from the shaking of a tree familiar to the brain?

Firstly, the brain doesn’t understand what the coconut flew into it for now. Imagine that the universe suddenly started working in a different way and now a blow to the head occurs a week later after we unsuccessfully shook the tree. Yes, we have already 5 times to forget that he was shaking.

When the brain doesn’t understand, he tries to find a suitable explanation. ( You can read more in this article ) And what is the most suitable explanation for the coconut that another person threw at you? That's right - this person is *** and is inclined to throw coconuts in everyone.

In order to think about his good motives, to remember that a person always always behaved rationally and it was pleasant to work with him, to come to the conclusion that something serious had to happen so that a coconut flew into you - you need to make a significant effort . Unlike the idea that we have a “coco thrower”, this logical chain is completely unobvious.

Aw, the brain made the wrong conclusion, by the same mechanism that led us to the erroneous "key and door theory." Instead of analyzing the whole chain and understanding the reason, we are satisfied with the correlation.

And now, based on its “magnificent” idea, the brain will do one more logical thing: it will try to solve the problem of coconuts flying into it. What is the most obvious solution? That's right, - you need to get rid of the "coco thrower". Well, if “getting rid of” would mean simply ignoring and going into the sunset. But this is NOT fair! The coconut was thrown at us first, and for an even count, you need to return it to the thrower. He himself must admit that he was wrong starting to throw coconuts at us.

But our opponent sees the situation exactly the same. Therefore, the cycle of mental coco-throwing can sometimes end with completely physical bumps from a very material assault.

The funniest thing in this whole situation is that usually the “winner” is the one who threw the last coconut. Although, in fact, the more my colleague and I were in the cycle, the more cones we got together.

Inside feedback

And now another interesting fact - sometimes the source of discomfort is our own memory and our beloved brain.

Suppose we parted with a colleague without clarifying anything. Now it is unpleasant for us to recall this. The brain has already concluded that our colleague is an evil coco-thrower, so the decision to “try to calmly talk and figure it out” looks completely unattractive. But with an unpleasant sensation, something needs to be done.

Wow, we have one great strategy that worked with a colleague - to ignore. Or switch to something else. Great way out, right? You just need to behave as if nothing had happened. I’ll think about something nice and make a cup of good tea. And that’s all - somehow it will resolve itself.

Remember, at the beginning of the text I said that there are two types of errors? One of them is information processing errors. Another is errors caused by its distortion or lack. And just the brain managed to plunge into a mistake of the second type. Now, instead of the full picture, he sees only the most pleasant piece of it, completely ignoring the options in which "does not resolve."

I dived into this puddle with my head about 7 years ago and splashed in it with pleasure for another 4 years. The problem here is also that when the ignored is "resolved" - you pay attention to it and consider that your strategy is effective. But if the problem remains, then you just continue to ignore it and do not take it into account in the analysis. Very nice puddle.

With such a strategy, the problem remains. We just create a blind spot, cover the lack of a window with curtains. We throw ourselves with painkillers, looking at the blushing and swollen finger. If someone accidentally pokes into it or patience overflows, it breaks us like a dam.

Sometimes everything really goes by itself. But sometimes - you can find yourself in a huge hopeless ass. The author was in it and does not recommend visiting.
Therefore, I look with a smile at the companies trying to introduce a cult of smoothing sharp corners and positive thinking. Too fragile a house of cards to ignore problems.

It seems that it is worth to come to the conclusion that sticking a blinking red light on the “problem” is an extremely bad strategy, but ...

Painkillers

But this is not so. And in this puddle I sat for another 3 years.

The bottom line is this: our mental bruise with coconut is not just some kind of "subjective garbage." This is a consequence of objectively occurring processes in an objectively existing brain. His “pain” is as objective as the pain of a burned finger or pulpitis tooth. The brain at the physical level attacks counter signals with already established connections and tries to reduce their strength by additionally cutting carpet bombing with neurotransmitters. (I can explain in the comments or in a separate article why I think so)

When a tooth hurts, we are in no hurry to refuse anesthesia. And we don’t blame other people for asking for an injection of ultracain before removing the teeth. We have no desire to say to the person to whom the eight is removed - “Get a rag! This is just a tooth! Nobody has died from this yet. ” (Although, probably, if half of the people in the world didn’t feel pain while removing the tooth, and the other half would ask for pain medication, the second would quickly pass for “rags, drug addicts, unable to cope with anything without their ~~antidepressants~~ painkillers.”)

When sucks - we really think differently. This is not the most suitable state of the brain to build chains more difficult than "shake the palm-> coconut fell-> it hurts."

It is curious that this process can lead us into a cycle:
The brain focused on the discomfort. Because of this, we forget that the brain during troubles does not work in the “long-term planning” mode. As a result of this, the thought does not come to him that one needs to calm down and return to the “planning mode”. Instead, the brain chooses a strategy that leads to a new portion of discomfort. The brain focuses on the discomfort and moves further away from the normal mode of operation.
(Then I will explain why so many things related to the brain have cyclic and recursive patterns).

Altogether, ignoring incoming information is a bad strategy, because it can lead to a mistake. Not to ignore is also bad, because a brain that is sick with coconuts thinks very poorly, and this leads us to an error. Hopeless situation? Well, not quite…

Exposure Compensation

I apologize in advance for the most hackneyed plot twist that you can imagine, but the only strategy I came up with is to find a balance. The golden mean between the perception of information and its ignoring.

Both of these tools are useful. If too much falls on us, we cannot handle it. If there is too little information, we will be mistaken because of its lack. We need to learn to seize the moment when our brain begins to think worse due to coconut blows, and do not go into the world of rainbow ponies, where coconuts do not hit anyone.

Using the analogy with painkillers - throw them in to get to the doctor - okay. Throw them until the burn on our finger causes blood poisoning - no.

Where is this balance located? You know, if I were a philosopher, I would now start saying something meaningful, cleverly sounding and absolutely useless, like: "Each of us is unique, and everyone must find their own path to balance ... blah blah blah." Usually all snowstorms of this kind should be translated like this: "I do not know." But, since I am not a philosopher, I do not like such an ending. I have an answer to that too.

How to fix?

Fail fast.
If we know that errors result from architecture and will happen, the best we can do is identify them as early as possible and prevent them from becoming ERRORS. The longer the error remains undetected, the more consequences it has, both outside the brain and inside it.

In the situation with our colleague, you could just take a hit and ask him to start the story from the beginning, and not from his final conclusions. If this is not enough strength - ask to postpone the conversation and calmly continue it after half an hour.

In the first case, we defined a situation pattern and broke it using ignore. In the second, they left the loop in the first step, again using ignore. And in both cases, we did not completely forget about the problem that they tried to tell us about. These decisions are many times better than a skirmish, which could otherwise happen. We save time, nerves and at the same time solve the problem.

But in order to act this way, we need to learn how to define patterns of such situations. The brain must understand that he turned onto one of his favorite paths, predict the development of events and tell us the right strategy. How to write a pattern to the brain about its own patterns?

There is this article for this.

I tried to start from the very beginning, with neurons and connections, and predict what features of information processing will be in such an architecture. I was hoping that some simple rules existed, much simpler than a huge list of cognitive biases. And given that in 10 minutes I described the mechanisms of occurrence of about forty-two of them, I think I guessed right.

Fading and linking - make fresh information more accessible.
Correlation is defined as causality - because of this, “gaps” arise in models.
False understanding - because of it, we cannot find the gap that has arisen.
Ignoring - distorts perceived information. In theory, this mechanism should filter out interfering factors; in practice, you can get carried away and filter out something useful for solving the problem.

Four factors, a combination of which generates almost all known cognitive biases. And 4 reasons are remembered much easier than 42 consequences.

Here is the rest of the list:

Cognitive bias involving ignoring

-

« »

Are there any bugs in the brain?

The brain is an amazing thing. So far, we have not created a mechanism that can solve the same problems as he does. Can this be considered architecture errors? Flaws?
I would not be so categorical. The fact that the brain is sometimes not suitable for solving the problem that has fallen on it does not mean that it is something bad. I will not be judged on the effectiveness of the microscope by the way it clogs nails.

Yes, due to the peculiarities of architecture, our brain does not always give the right answer. But he is also able to recognize his mistake and try to correct it. And so far, this is the best we have.

Most likely in this article there are errors, the nuances that I missed and the places where I overdid it with an explanation. If you notice them, please write.

It is possible that in my line of reasoning there is a flaw that will destroy it to the ground. If you find him, I will gladly hear about him.

Yes, it will ruin 2 years of my work and hit me harder than any coconut. But this is the price I pay for the opportunity to become better. And for now, I am pleased with the exchange rate.

That which can be destroyed by the truth should be. - PC Hodgell, Seeker's Mask

PS
If you do not understand something, or vice versa, if you would like to hear a more “technical / mathematical” explanation for a moment, it will be great to see your comment. I will think about it and answer it with pleasure.
If what I described is at variance with your observations, or vice versa your experience is consistent with my conclusions, I will also gladly listen to such a feedback.
After all, Habr is the place where comments are no less interesting than the article itself.

Why are there bugs in the brain?