What is important for creating an AI?
After reading many articles on artificial intelligence, I decided to write my own. I have been watching the development of AI, or rather, artificial neural networks for quite a long time. Heard about projects to emulate human brains, cats, etc. I can not express my disappointment on this topic. In fact, the intelligence power of artificial neural networks is exaggerated and embellished by journalists and developers. For example, I saw the following article: “Google's neural network learned to identify cats in images.” This article is so inflated and embellished that you really think: “Eureka! Here it is our AI! Finally!". Many, listening to such statements on television, with their mouths open say something like: “Look at what the technique has reached!”. This just makes me smile and disappointment. And this is why ... Google itself said that among the images of the “fed-up” neural networks, most are images of cats. If we take into account the property of the neural network to learn, then no intelligence and the supernatural are not visible here, it was natural that the neural network would learn to find cats, agree. Okay, quite a lyrical indent, let's get down to the real reasoning.
I myself have little experience with neural networks and, rather, not with their creation, but with their research and experiments. I myself thought a lot about the creation of AI, there was a time when I myself got the bait of embellished stories about ANS, which is why my interest in them was born. In my plans for the future to create my own INS, after I understand how to create a really smart one. For this, I made a small list of questions on this topic that should help in creating the AI.
I read an article about the importance of glial cells and, perhaps, to create AI, you need to emulate them in the ANN. In my childhood I watched a lot of a parrot, and was amazed by its intellect. He sometimes got up to achieve his goal, that I would never have thought of that. I don’t want to clutter up the article and I’ll just say that anyone who is interested can find a bunch of videos on YouTube confirming that the parrots "have brains." To my great regret, there is almost no information about the brain of parrots on the Internet, except for the general superficial one. The intelligence of a parrot is compared to the intelligence of a three-year-old child. Looking at all this, I decided to simulate the brain of a parrot, due to the fact that there, approximately one million neurons, it will be easier than to simulate a human brain. But what struck me, I quote: “Everything can be, but I must say, anatomists have long noticed that if we consider the ratio of body weight to brain mass, then parrots have a brain that is relatively large, and interestingly, it lacks glial cells , which are supporting, and therefore in a smaller space in parrots more neurons fit. ” I can put an end to this by saying that we don’t need to emulate them - glial cells are not important for building AI.
I came across an article that neurons are not an adder, but actually react to a combination of signals. To make it clearer: let's say we have some trained neuron with two inputs (dendrites) X1, X2 and one exit (axon) Y, which works on the combination X1 = 1, X2 = 0 and issues Y = 1, and all others combinations Y = 0. I became interested in the idea, but upon detailed study I found out that any part of ANN has such properties, i.e. group of neurons. I agree that using such a neuron you can build an interesting ANN, but we can say that this is not important either, just the group of ordinary, for us, artificial neuron-adders will play the role of such a neuron.
')
Once I solved a difficult task. It was necessary to develop and integrate into the system developed by me a new module, besides to invent it. I will not go into details, which I don’t remember, but nothing worked out for me. And I just suffered, I was so angry that I did everything in defiance. As a result, I found a solution to the problem and at the same time realized the great importance of emotions for the intellect. They make you think differently, get new solutions, set new goals and much more. Emotions are very closely related to the intellect. The higher the intellect of the creature, the more diverse the emotions of them. Compare the creatures according to their intelligence and pay attention to their emotions and you will understand. What is the conclusion: a creature without emotions can not be smart.
I think that this is understandable to all, because the creators of the ANN are trying to copy the structure of the brain in order to create AI. From the very beginning, when the system is stupid and does not know anything, it should already be initially programmed to something. I'm talking about the analogue of unconditioned reflexes, as well as instincts, which from the very beginning set goals for the intellect. Of these goals, other more complex ones emerge. For example, a person needs to eat, to eat, you need to get food, to get food you need to work, to work you need to be able to do something, to be able to have something to learn. Recall a child who is interested in everything, he is studying everything, exploring, this is already “born” in him from birth. How exactly this program is embedded in the brain, as I understand it, is unknown.
During the life of the neurons of the brain constantly break and create connections (synapses). I want to note that, alas, this is not an accident. The mechanism of this phenomenon is unknown. After all, we don’t have a mess in our heads. Neurons didn’t go anywhere, but the general architecture remained as it was (see anatomy). Many who scream: "Yes, everything is clear there!". Yes, it is not clear! - I will say. Take the human brain and apply the laws of the Kohonen network to it ... No, I do not in any way ridicule his network, it sometimes shows phenomenal results, but it’s not even funny to me when I imagine what will happen to the brain. So it is not clear how new neuron connections appear. I think it’s understandable how they are broken - the extremely low value of the synapse weights, but I am sure that this is not the only reason and there are exceptions. Ultimately, the overall architecture of the brain remains unchanged.
In favor of the fact that the architecture of the brain is important, we can mention the fact that different people have the same part of the brain responsible for the same functions. Verdict: The architecture of the brain is very important. And we are unlikely to get a strong AI as a result of creating an arbitrary ANN.
It is known that neurons die without stimuli, but this is not the point. It will not be possible to create an AI, simply fill it with a lot of information. The most important thing in AI is an understanding of the information it has learned. How many do not show the neural network photos of the cat, she will not understand what it is (or who it is), except how it will begin to recognize it. The memorization of the material does not give understanding. The intellect will never understand what is so cold or hot (let it be scientifically understood that it is temperature, but, for example, it will never understand what love is without feeling it, maybe I ran very far, but still). Returning to the cat, the AI ​​should see the cat with its “eyes” and understand what it is (or who it is). In fact, it is not necessary for AI to be in the real world. It can be placed in the virtual world, but to accurately simulate the real one is impossible. And moving into the real world of AI will look stupid.
It is still unknown what it is and how it works. It is not known how we all see, hear, feel, understand and realize. (My imagination draws a small ball (or something similar), to which all parts of the brain are connected, the subconscious tells me that there must be a part of the brain, which we ourselves are, and all the rest are just means of receiving and processing information. But this is just my imagination.)
Here you can add about my assumption that inside the brain there is something that stimulates neurons from the inside. Here, take, for example, ins. She "thinks" only when an external signal is given to her, and we think even after the cessation of the signal. Suppose we are shown a photo of a cat (I apologize for the irony, forgive me Google), but even after the cessation of the signal (the photo was hidden), we can think about it and analyze it. The second example: we were asked something (ie, said a few words), but we can state the whole story, despite the fact that we are no longer “influenced”.
It follows that in our brain, neuron irritation continues. Something stimulates neurons (this is how I describe thinking) before getting the desired result, this is consciousness ... I think to begin experiments in this area. That's just not clear exactly how to stimulate neurons and when to assume that the problem is solved and you need to stop the stimulation. Here, again, the question of the architecture of the ANN (brain) spontaneously arises, namely where exactly to stimulate neurons, to get thinking, how exactly to construct an ANN, how to manage the stimulation of neurons.
I have it all! Do not judge strictly, this is my first article on the site. I hope someone will push on new thoughts or someone will push me on them.
I myself have little experience with neural networks and, rather, not with their creation, but with their research and experiments. I myself thought a lot about the creation of AI, there was a time when I myself got the bait of embellished stories about ANS, which is why my interest in them was born. In my plans for the future to create my own INS, after I understand how to create a really smart one. For this, I made a small list of questions on this topic that should help in creating the AI.
Glial cells
I read an article about the importance of glial cells and, perhaps, to create AI, you need to emulate them in the ANN. In my childhood I watched a lot of a parrot, and was amazed by its intellect. He sometimes got up to achieve his goal, that I would never have thought of that. I don’t want to clutter up the article and I’ll just say that anyone who is interested can find a bunch of videos on YouTube confirming that the parrots "have brains." To my great regret, there is almost no information about the brain of parrots on the Internet, except for the general superficial one. The intelligence of a parrot is compared to the intelligence of a three-year-old child. Looking at all this, I decided to simulate the brain of a parrot, due to the fact that there, approximately one million neurons, it will be easier than to simulate a human brain. But what struck me, I quote: “Everything can be, but I must say, anatomists have long noticed that if we consider the ratio of body weight to brain mass, then parrots have a brain that is relatively large, and interestingly, it lacks glial cells , which are supporting, and therefore in a smaller space in parrots more neurons fit. ” I can put an end to this by saying that we don’t need to emulate them - glial cells are not important for building AI.
Do neurons react to a certain combination of signals or do they simply sum the signals?
I came across an article that neurons are not an adder, but actually react to a combination of signals. To make it clearer: let's say we have some trained neuron with two inputs (dendrites) X1, X2 and one exit (axon) Y, which works on the combination X1 = 1, X2 = 0 and issues Y = 1, and all others combinations Y = 0. I became interested in the idea, but upon detailed study I found out that any part of ANN has such properties, i.e. group of neurons. I agree that using such a neuron you can build an interesting ANN, but we can say that this is not important either, just the group of ordinary, for us, artificial neuron-adders will play the role of such a neuron.
')
Are emotions important for creating AI?
Once I solved a difficult task. It was necessary to develop and integrate into the system developed by me a new module, besides to invent it. I will not go into details, which I don’t remember, but nothing worked out for me. And I just suffered, I was so angry that I did everything in defiance. As a result, I found a solution to the problem and at the same time realized the great importance of emotions for the intellect. They make you think differently, get new solutions, set new goals and much more. Emotions are very closely related to the intellect. The higher the intellect of the creature, the more diverse the emotions of them. Compare the creatures according to their intelligence and pay attention to their emotions and you will understand. What is the conclusion: a creature without emotions can not be smart.
The importance of brain architecture (or ANN) for AI
I think that this is understandable to all, because the creators of the ANN are trying to copy the structure of the brain in order to create AI. From the very beginning, when the system is stupid and does not know anything, it should already be initially programmed to something. I'm talking about the analogue of unconditioned reflexes, as well as instincts, which from the very beginning set goals for the intellect. Of these goals, other more complex ones emerge. For example, a person needs to eat, to eat, you need to get food, to get food you need to work, to work you need to be able to do something, to be able to have something to learn. Recall a child who is interested in everything, he is studying everything, exploring, this is already “born” in him from birth. How exactly this program is embedded in the brain, as I understand it, is unknown.
During the life of the neurons of the brain constantly break and create connections (synapses). I want to note that, alas, this is not an accident. The mechanism of this phenomenon is unknown. After all, we don’t have a mess in our heads. Neurons didn’t go anywhere, but the general architecture remained as it was (see anatomy). Many who scream: "Yes, everything is clear there!". Yes, it is not clear! - I will say. Take the human brain and apply the laws of the Kohonen network to it ... No, I do not in any way ridicule his network, it sometimes shows phenomenal results, but it’s not even funny to me when I imagine what will happen to the brain. So it is not clear how new neuron connections appear. I think it’s understandable how they are broken - the extremely low value of the synapse weights, but I am sure that this is not the only reason and there are exceptions. Ultimately, the overall architecture of the brain remains unchanged.
In favor of the fact that the architecture of the brain is important, we can mention the fact that different people have the same part of the brain responsible for the same functions. Verdict: The architecture of the brain is very important. And we are unlikely to get a strong AI as a result of creating an arbitrary ANN.
The importance of AI communication with the outside world
It is known that neurons die without stimuli, but this is not the point. It will not be possible to create an AI, simply fill it with a lot of information. The most important thing in AI is an understanding of the information it has learned. How many do not show the neural network photos of the cat, she will not understand what it is (or who it is), except how it will begin to recognize it. The memorization of the material does not give understanding. The intellect will never understand what is so cold or hot (let it be scientifically understood that it is temperature, but, for example, it will never understand what love is without feeling it, maybe I ran very far, but still). Returning to the cat, the AI ​​should see the cat with its “eyes” and understand what it is (or who it is). In fact, it is not necessary for AI to be in the real world. It can be placed in the virtual world, but to accurately simulate the real one is impossible. And moving into the real world of AI will look stupid.
Consciousness
It is still unknown what it is and how it works. It is not known how we all see, hear, feel, understand and realize. (My imagination draws a small ball (or something similar), to which all parts of the brain are connected, the subconscious tells me that there must be a part of the brain, which we ourselves are, and all the rest are just means of receiving and processing information. But this is just my imagination.)
Here you can add about my assumption that inside the brain there is something that stimulates neurons from the inside. Here, take, for example, ins. She "thinks" only when an external signal is given to her, and we think even after the cessation of the signal. Suppose we are shown a photo of a cat (I apologize for the irony, forgive me Google), but even after the cessation of the signal (the photo was hidden), we can think about it and analyze it. The second example: we were asked something (ie, said a few words), but we can state the whole story, despite the fact that we are no longer “influenced”.
It follows that in our brain, neuron irritation continues. Something stimulates neurons (this is how I describe thinking) before getting the desired result, this is consciousness ... I think to begin experiments in this area. That's just not clear exactly how to stimulate neurons and when to assume that the problem is solved and you need to stop the stimulation. Here, again, the question of the architecture of the ANN (brain) spontaneously arises, namely where exactly to stimulate neurons, to get thinking, how exactly to construct an ANN, how to manage the stimulation of neurons.
I have it all! Do not judge strictly, this is my first article on the site. I hope someone will push on new thoughts or someone will push me on them.
Source: https://habr.com/ru/post/183994/
All Articles