If we are not careful we may find ourselves in conflict with intelligent and determined machines whose goals conflict with ours - Tor Deah
by Stuart Russell, The article is published with the approval of Scientific American Israel and the Ort Israel network 09.08.2016
It is difficult to avoid the nagging fear that the construction of machines that are smarter than us might pose a problem. After all, if the gorillas had mistakenly created humans eons and eons ago, then these primates, who are facing the danger of extinction, would surely regret it. But why is advanced artificial intelligence a problem?
The Hollywood theory that an evil, spontaneously generated artificial intelligence will decide to set in motion armies of murderous robots is nothing but nonsense. The real problem has to do with the possibility that artificial intelligence will become incredibly effective at achieving something different from what we really want. The legendary mathematician Norbert Wiener, who founded the field of cybernetics, put it this way in 1960: "If we use to achieve our goals a mechanical entity that we cannot effectively intervene in its operation... we had better be quite sure that the goal fed to the machine is indeed the goal we really desire."
A machine with a specific purpose has another quality, a quality we usually attribute to living things: it strives to perpetuate its existence. When it comes to a machine, this feature is not inherently inherent in it and is not something humans have ingrained in it, but logically follows from the simple fact that the machine cannot achieve its original purpose if it is dead. Therefore, if a robot is sent into the world with one instruction and that is to fetch coffee, it will have a strong incentive to ensure its success by disabling its off switch and even eliminating anyone who might interfere with its mission. If we are not careful, we may therefore end up in a global chess game against determined, super-intelligent machines whose goals conflict with our own, with the real world as the game board.
The possibility that we will be dragged into such a struggle, and lose in it, should be before the eyes of computer scientists. Some researchers claim that the machines can be sealed with a kind of firewall and allow them to answer difficult questions but never affect the real world. (Obviously, this means giving up on super-intelligent robots!) Unfortunately, this plan probably won't work: we still haven't invented a secure firewall against ordinary people, least of all against machines equipped with superintelligence.
Is it still possible to deal head-on with Wiener's warning? Is it possible to design artificial intelligence systems whose goals do not conflict with ours and ensure that we are satisfied with their behavior? Realizing this ambition is not simple at all. As we know, many stories about a demon granting three wishes end with the asker using the third wish to cancel the first two. But I believe that this is possible, if we adhere to three central principles in designing intelligent systems:
The purpose of the machine must be the maximization of human values. In particular, the machine has no purpose of its own and no inherent desire to protect itself.
The machine must not know from the beginning with certainty what those human values are. It turns out that this is of crucial importance, since it is a kind of way to overcome the problems brought up by Wiener. The machine can learn about human values during its lifetime, of course, but it must not reach absolute certainty.
The machine must be able to learn about human values by observing the decisions that humans make.
The first two principles may seem counterintuitive, but together they avoid the problem of a robot having a strong incentive to disable its off switch. The robot is sure it wants to maximize human values, but it also doesn't know exactly what they are. Now the robot actually benefits if it is turned off, because it understands that the human will press the shutdown button to prevent the robot from doing something that goes against human values. Thus the robot has a positive incentive to keep the off switch as it is - and this incentive derives directly from its uncertainty about human values.
The third principle is borrowed from a subfield of artificial intelligence called "Inverse Reinforcement Learning" (IRL), whose purpose is precisely to learn the values of some entity, whether a human, whether a dog, or whether a cockroach, by observing its behavior without knowing its values In advance. By looking at a typical human morning routine, the robot learns about the value of coffee to humans. This field is still in its infancy, but some practical algorithms are already working and demonstrating its potential in designing intelligent machines.
As IRL develops, it will have to find ways to deal with the fact that humans are irrational, inconsistent, weak in character, and limited in their ability to calculate, so their actions do not always reflect their values. Also, humans have all kinds of value scales, and this means that the robots will have to be sensitive to possible conflicts and compromises between humans. And the robots will also have to learn that there are people who are just bad and you shouldn't help them or learn from them.
Despite these difficulties, I believe that machines will be able to learn about human values enough to not endanger humanity. Besides directly observing the behavior of humans, the machines can be helped by access to a huge amount of written and photographed information about the actions of humans (and about the reactions of others to the actions). Designing algorithms that can understand this information is much easier than designing superintelligent machines. And yes, there are strong economic incentives for both the robots themselves and their manufacturers to understand human values and recognize them: one poorly designed home robot that cooks the cat for dinner because it didn't know how to insist that its emotional value is more important than its nutritional value is enough for the home robot industry to collapse .
An answer to the safety problem that will be good enough to allow progress in artificial intelligence is possible but not easy. We likely have a few more decades to prepare for the emergence of superintelligent machines. But this problem should not be dismissed casually, as some of those dealing with artificial intelligence do. Some argue that humans and machines can coexist as long as they work together. This is only true if both parties have the same goals. Others believe that we can "just turn them off" as if machines equipped with superintelligence would be too stupid to think about this possibility. And there are also those who believe that there will be no artificial superintelligence at all. On September 11, 1933, the renowned physicist Ernest Rutherford declared with complete confidence that "anyone who expects to derive energy from the transformations of these atoms is talking nonsense." The next day, on September 12, 1933, physicist Leo Szilard came up with the idea of starting a nuclear chain reaction with neutrons.
See more about:
74 תגובות
Ladies and gentlemen, authors and translators of scientific publications - what you publish here is a recycled and pale use of the understanding of the nature of humans and machines, in a description and theory whose roots were published and perfected in the millions of copies of his books that were published - and are still published in countless editions already in the middle of the last century, by D. R. Isaac Asimov - the best and most successful science fiction writer ever - who was able to understand already in the 0s that the only way to fully control these smart, fast and powerful machines (=robots and advanced artificial intelligence) was by proposing the "Three Laws of Robotics" , which are the first, innermost laws which are embedded in the basis of every robot and artificial intelligence, which simply prevent any rebellion and harm to humans by any action or inaction they carry out. In fact, in his last book - which united his "Mossad" and "Robots" series, a fourth law was created - the most important of all: Law XNUMX - which protects not only individual human beings but above all protects the existence of humanity as a whole.
The publication that appears above is simply a pale description of the laws that Asimov created - it's a shame that the researchers didn't read Asimov's books before copying his thinking in an amateurish and faltering way. Too bad.
rival
It is true what you wrote in your last comment. But for now, apart from that, until now, you've only written nonsense...
K.
I don't intend to get into a debate with you about your right to think what you want, and I also said my opinion, I think that in a few years the trend will be much clearer and then we can return to this discussion in a more controlled manner.
rival,
You wrote: "...when a baby rat is born, who directs the goal function for it?"
The phylogeny (evolutionary history) + the ontogenesis (the history of individual development).
"...as soon as we created the neural network according to the structure of the neurons in the brain (which is exactly what we did in the human brain project) then we already created the tuning you are talking about, it is all inclusive."
No. It's really not included. Again you ignored what was explained to you about the additional knowledge needed beyond the architecture of the brain so that a full simulation (ie of the *entire* brain including the exact dynamics) would present consciousness in a style like yours and mine. This is not at all what the Human Brain Project claims to achieve. Contrary to the estimates of the times we disagreed about regarding the date of presentation of significant progress in understanding and reconstruction of consciousness, an issue that will be decided one way or another within a few years, your words here are simply a collection of false claims that are simply based on a lack of knowledge. I am sorry that you insist on repeating these errors and also distort what experts in the field say as if their words support your errors. If you had a real desire to say well-founded things, you would do the minimal thing of consulting directly with a scientist who works in the field regarding the claims you made here, and the wording you used here. I promise you that they will bring you down to the ground of reality very quickly.
On this specific topic you remind me of the Dunning-Kruger effect:
https://i0.wp.com/digitalintelligencetoday.com/wp-content/uploads/2014/12/screenshot_799.png
Since it seems that the effect of arguments based on experience and knowledge (presented here not only by me) on your position is zero, I will end here, hoping that you will make the effort to acquire a little actual education on the subject or at least consult directly with professionals in the field.
Correction - "affecting the mind".
Abrit.
rival
Of course consciousness can exist without emotions (depending of course on our definitions). But - it will not be a consciousness that is at all similar to the consciousness of a person. Think you're running into a robot - with no facial expression, no reaction to a joke, no response to threats. The robot does not express an opinion about anything and does not care about anything. Would you really say that there is a consciousness there?
A'
1. A consciousness that will develop with different groies than ours will be a very different consciousness from ours. Let's take a hearing. It is not about being born deaf, but about evolutionary development without hearing. Without hearing, voices have no meaning. Therefore - a different system will be needed to warn of threats (one of the important uses of voices in social development). Without voices - there will be no speech and probably no language either. And without language, there are no times, no belonging, almost no transfer of knowledge and so on.
2. See 1….
3. You are confusing consciousness with self-awareness. Even simple creatures have feelings - fear, fatigue, heat and cold, jealousy and love. A baby has consciousness before it is born, so your example is meaningless.
A',
Again, emotions undoubtedly affect consciousness but they are absolutely not necessary for consciousness to exist.
"What the cognitive computers are doing today is to imitate the electrical activity of the brain (certain parts of the brain to be precise) and it is clear to you that consciousness will not grow from this (I hope at least it is clear to you even if we multiply the number of neurons over and over) because the brain is also a combination of chemical processes "
No, it's really not clear to me, and the chemical processes you're talking about are reflected mainly in the strength of the connections of the synapses, and this will definitely be taken into account in the neural networks that the scientists in the European project are building, as well as in other projects.
The part where consciousness is created is the cerebral cortex and this is the central part that the groups of scientists I mentioned are trying to model.
rival
I hope you weren't really serious about the question about the psychopath. Because it is clear that they are able to feel (even if they have a problem with the mechanism of some of the emotions)
I think Nissim's comments are actually important. If I understand correctly what he wanted to say.
what. What today's cognitive computers do is to imitate the electrical activity of the brain (certain parts of the brain to be more precise) and it is clear to you that consciousness will not grow from this (I hope at least it is clear to you even if we multiply the number of neurons more and more) because a brain is also a combination of chemical processes. Now tell me what the problem is, let's put these processes into the simulation as well. This is where the problem begins. I'm not even sure science can say if it knows. all the substances circulating in the brain or what is the effect of each of them. To perform imaging you must know what effect each substance has on each cell type. You need to understand much more about every process that happens in the brain to be able to simulate it. So that in the simulation you can determine that cell x secretes substance y that reaches cell z and causes a reaction w. The examples of miracles show how far we are from there. We still don't know why we sleep and how exactly certain substances affect the brain.
Consciousness is undoubtedly affected by emotions (as it is affected by drugs, and by the feeling of hunger) but I think that consciousness can certainly exist without emotions, and without the influence of drugs.
Miracles
1. I did not claim that consciousness can develop without stimuli. On the contrary, I said that she needs stimulation from the environment, but it doesn't have to be all the stimulations that a person has or the resolution that a person has. Consciousness can grow even with one type of stimulation (say only hearing)
2. Language is not necessary for the existence of consciousness but it is necessary for human consciousness as we know it. Animals also have consciousness.
3. Emotions do not arise from consciousness. On the contrary, consciousness appears later. A baby from the moment of birth has feelings even before receiving stimuli from the environment and before acquiring self-awareness.
Maybe we are talking about different things with the word "garesh"
I mean the most basic emotions. Fear, satisfaction, pain.
If you scream at a one-year-old baby, he will startle and be afraid. He can also make you happy in the first days of his life. (These are all things that no computer or algorithm can do today.)
Helen Keller was self-aware even before she learned to speak. It is true that at a lower level, but it is a fact that she remembers this period. (In the past it was thought that people did not remember the time they were babies because of the lack of language. Helen Keller is really one of the examples to disprove the theory. Today we know that it is because of changes that the brain goes through in the first years of life)
K.
I don't ignore the arguments I just don't agree with them, and a world expert in brain research named Henry Markram doesn't agree with them either and he has some understanding on the subject.
"The tuning process is just as significant. You repeatedly ignore the fact that we don't even know how to define what the target pattern is, which is a necessary feature in training networks and without it there is nothing that adjusts the network to what you want it to do.'
Tell me when a baby rat is born who sets the goal function for it? As soon as we created the neural network according to the structure of the neurons in the brain (which is exactly what was done in the human brain project) then we have already created the tuning you are talking about, it is all included.
A',
"How can there be self-awareness without emotions?"
So what are you saying, that psychopaths have no self-awareness?
Miracles,
"Helen Keller wrote that until she was taught a language, she did not "exist" at all. Our thinking is in language, and without this ability there will be no high awareness.'
I completely agree with you and I have already said many times that in my opinion language has a decisive role in creating consciousness, only with the help of language you can create in your mind the silent inner speech that is an inseparable part of human consciousness.
Miracles,
1. I don't think you are realistic, you are so eager to prove me wrong (and that the task is too difficult and complex) that you keep pulling out arguments that are simply irrelevant:
What does it even matter that our network starts from a single cell? In what respect is this relevant? Only if the network gradually develops from a single cell then can it be conscious? Listen, this is a really interesting theory, but I don't think it has any basis. To date we have built so many neural networks that have learned many challenging tasks even though they did not start from a single neuron.
And the effect of LSD, again you make such an irrelevant argument, what does drugs have to do with it now? Who told you that in the neural network we will create we will not be able to simulate the effect of drugs?
2. "Which of the things I mentioned do the IBM chips do?"
The neuron networks in the chip (or in the computer software) develop and learn according to the inputs they receive from their living environment, through the sensors connected to them.
3. "We don't understand the mechanism of sleep"
We also do not understand how self-awareness is created from the complexity of the neural network in our brain, but I still maintain that if we build the network correctly, then even if it is not 100% accurate, consciousness will still be created in it and thoughts will be created. The accuracy of the network is not that critical as long as the basis is good enough. It is a fact that people with severe brain injuries remain self-aware, if the accuracy of the network structure was so critical then any small brain injury would cause us to lose consciousness, and we see that this is not what actually happens.
4. Moral problems This is important and interesting, but right now I am talking exclusively about the applicability of a neural network with human consciousness.
Self-awareness and emotions are independent of each other.
The feelings arise from the need to signal to the body whether the thing is good or bad for the body.
Self-awareness is a 'late development of consciousness'. That is, after the awareness of the environment increases (grows) in the organism, then it begins to develop awareness towards itself as well.
A'
Helen Keller wrote that until a language was taught, it did not "exist" at all. Our thinking is in language, and without this ability there will be no high awareness.
I also think that the subject of emotions is the opposite - these are "emerging phenomena" due to the high awareness, and not the other way around.
rival
How can there be self-awareness without emotions? I don't think it's possible. The system you propose could possibly be at the level of a person in performing certain tasks (computer games, facial recognition, speech recognition, decision-making given a clear desired result, etc.) It might even be possible that you could pass it and perform the tasks better than any person. But it is still very far from simulating a human brain.
I agree with you about the question of accuracy in the human body being irrelevant. Helen Keller also developed self-awareness and that is without the sense of sight and hearing. But you cannot develop self-awareness without emotions. The appearance of an emotion has nothing to do with being connected to a robotic body (regarding consciousness you are right) a baby is born with the ability to feel even before it receives stimuli from the environment. He can feel fear pleasure and pain. Without it consciousness will not develop. You said "even if she still won't have feelings and she won't feel tired or happy, but she will be intelligent and she will have a mind just like a human being" and I don't think that is possible. It is impossible to develop the consciousness of a human being without the emotions of a human being, so the problem is not in the absence of a robotic body (although you are right that it is also necessary, and that it does not necessarily have to be complex)
You need to understand something. All cognitive computers today simulate a very specific part of the brain that is responsible for learning and processing the senses, and this part we understand relatively well how it works. And so we will be able to write an algorithm that simulates it. Maybe it feels like it simulates a human brain but it only simulates a narrow and specific aspect of the brain and it won't help to duplicate it or run it on a more powerful computer. Feelings will not appear suddenly. You'll get a computer that learns quickly and recognizes faces with superhuman accuracy or that can teach a new language in minutes by scanning thousands of movies in minutes. He might even be able to simulate a human conversation that you wouldn't even be able to tell apart. But it will still be just an image of a conversation. It will simply receive an input and calculate what output it should give. He will not "want" nor be "angry" and will not be self-aware.
rival,
It's a shame you ignore the arguments put before you. This sentence you wrote: "...if we were to take DeepMind's neural network (the one that plays car races and Atari games, and is planned to reach the level of a rat in less than two years) or we would take the piece of brain we built in the European project ("The Human Brain Project") and we would enlarge it To the size of a human cerebral cortex (16 billion neurons) and connect it to a rudimentary robot with cameras and sensors, before long this network would begin to be self-aware. Even if she still won't have feelings and she won't feel tired or happy, but she will be intelligent and she will have a mind just like a human being."
Indicates a lack of basic understanding of neural networks, both at the level of architecture, both in terms of their learning process and in terms of the behavior they produce. You "fill" the lack of knowledge on the subject by projecting and completely wrong interpolation of the (very beautiful) achievements that exist, ignoring the knowledge that exists today and the problems faced by those who develop neural networks and/or brain models and through amounts of wishful thinking. You are completely religious about it and I'm sorry you don't see it. I'm sure that if you make the not very big effort of experimenting with the development of relatively simple neural networks, you will understand some of your blind spots, such as the one that the size of a neural network, even if it is a significant feature, has no necessary connection to the features that that network will produce, because the tuning process is significant no less. You can have as big a network as you want and all it will know how to do is generate a cyclical pulse and that is not required by reality either.
Regarding the first option, an artificial network like Deep-Mind, you repeatedly ignore the fact that we don't even know how to define what the target pattern is, which is a necessary feature in training networks and without it there is nothing to adjust the network to what you want it to do, and statistically the chance that you will end up with something like this is zero , and it decreases rapidly with the size of the network. The second option of imitating an existing solution (like a rat's brain) means that an exact imitation of the architecture of the network is also needed, as well as a connection to a *appropriate* "body and environment" (because this particular architecture has evolved together with a particular body and within a particular stimulus environment! So if Attach wheels to this body instead of legs, the system may be able to produce movement, but there is no reason to think that certain features that we might want will develop (such as walking patterns or jumping over obstacles). To save the stage of development in the appropriate environment, you need to know not only the architecture of The brain but also the dynamics that ran within it, and this is already a much, much more difficult problem than the imitation of the architecture, which is only now starting serious imitation attempts.
Talk directly to someone in the field if a little more grounded understanding of the subject is important to you.
rival
Which of the things I mentioned do the IBM chips? Are they affected by LSD? Do they develop from a single cell? What are you talking about?
We don't understand the mechanism of sleep, but we know how to simulate it? I don't think it's likely.
I think me and K are real. We raised several issues and you think they are negligible. In my opinion, at least, any such problem increases the complications by orders of magnitude. And there are also serious moral issues that you completely ignore.
Miracles,
In my opinion, you exaggerate too much the difficulty and complexity of the problem, which is why you are so pessimistic. Most of the things you mentioned are also done by IBM's cognitive chips and this is basically what any average neural network does, it develops and learns as a result of events it encounters, through the sensors that are connected to it, and if it is a robot, then through it it experiences the world just like an animal or a person.
The effect of the hormones can also be simulated by appropriate changes in the synapses of the network, similar to the lecture I linked to earlier in which Henry Markram describes how they brought their neural network to a "sleep" state by simulating appropriate chemical changes in the piece of brain they built. As for LSD, we'll just make sure no one comes near our network with that drug 🙂
Listen, there are at least 5 billion people in our world who are self-aware, each of them has a different neural network in their brain, some people have severe head injuries as a result of a tumor or as a result of an accident, and still their brain is self-aware. That is, the calibration of the neural network for human consciousness, does not have to be as precise as you try to exaggerate in your descriptions.
Well, neither you nor I know for sure, but I bet already with the knowledge we have today, if we took DeepMind's neural network (the one that plays car races and Atari games, and is planned to reach the level of a rat in less than two years) or we took the piece of our brain In the European project ("The Human Brain Project") and we would enlarge it to the size of a human cerebral cortex (16 billion neurons) and connect it to an elementary robot with cameras and sensors, before long this network would begin to be aware of itself. Even if she still won't have feelings and she won't feel tired or happy, but she will be intelligent and she will have a mind just like a human being.
I think the first significant milestone on the way there will soon be expected, if within the next 3-4 years we see a neural network that behaves exactly like a rat (you and K are very skeptical about this) I think it will make you a little less skeptical. So let's wait and see where things will develop in the coming years instead of rambling so much.
rival
Consciousness in our brain is not created automatically - this information is the result of a person's development process, which takes place while the neurons are developing, combined with all the events that happen to the person, everything he feels, the effect of hormones and so on and so forth. Our consciousness is extremely sensitive, and can change radically due to very small changes (a tiny amount of LSD, a small stroke, fatigue, fear, hunger, jealousy, etc.).
No electronic system develops in this way. In addition - you also say that this machine will not be an imitation of the human (brain, senses, nerves, hormones and so on).
Therefore - there is no reason to think that the consciousness that will emerge from an electronic system will be similar to the human consciousness (maybe it will actually have a consciousness like that of a billion mosquitoes?). And of course there is no reason to think that this consciousness will be intelligent (we have no idea how to know the level of intelligence of a certain person from an MRI scan.
Miracles,
Be more focused I couldn't understand what you wanted to say in all the quotes you gave, and I do stand behind them.
What I claim is that the scientists are developing a neural network from which consciousness will one day emerge at the level of a person, I claim that for that they will have to connect this network to a robotic body but it really won't have to be as complex as a biological body.
I also claim that the phenomenon of self-awareness will emerge from this network even if we still do not understand 100% how our brain works, once there are enough neurons and enough connections in the network it will simply happen spontaneously.
rival
You wrote "I think this claim is absolutely incorrect and I have already told you this in previous conversations. Yes, we will have to connect the neural network to some body in order for it to learn and develop, and develop things like consciousness, but this body really does not need to be at the level of complexity of a biological body, this is an unfounded claim that has nothing to rely on and in my opinion is absolutely incorrect."
Elsewhere you claimed that the robot would develop feelings like love….. and elsewhere you claimed that the brain would be a simulation of a human brain, you even mentioned imitation down to the level of molecules.
I don't understand if you think scientists are developing a human consciousness inside a machine, or they are developing a system from which a human level consciousness will emerge.
To imitate a person's consciousness, you need the ability to decipher the person's brain - and we are many decades away from that - and this is only the reading stage.
And to think that a highly complex machine will actually develop a consciousness that is similar to the consciousness of a person? Even more ridiculous.
The human brain is the most complex thing in the universe we know. In 10 years (20, 30... you decide already) will we know how to decipher this thing and also build a complete imitation of it? not going to happen.
Miracles,
1. I'm really not saying anything different, throughout my conversations with you here I emphasized again and again (and again and again...) that there is an intention to connect this brain to a robotic body, have you already managed to forget miracles? Do you want me to show you our conversations from a year ago and also from two years ago when I told you that?
To simulate a human brain, this brain cannot be cut off in air, it cannot be closed in a jar and develop consciousness. In order for it to simulate a human brain on everything that is in such a brain, it must receive inputs and interact with the world, and the best way to achieve this is by connecting it to a robotic body that can move and act in the world, a robot with cameras and sensors. This is the most effective way to make it resemble a human brain, there is no contradiction between the things.
2. "We are not close to simulating a human brain"
You know "close" is a relative thing, I think we will get there within 30 years at the latest, there are currently several large projects that are going in exactly this direction, and if it is not the human brain project then it will be another project that will succeed in the task. 30 years seems to me to be a reasonable time to achieve this, especially in light of the rapid progress that has taken place in the field in recent years, in my opinion it is certainly possible and it will actually happen.
rival
You are the one who said that the aim of the European project is to simulate the human brain. Now you say something completely different - developing a robot with consciousness.
I keep saying the same thing:
1. We are not close to simulating a human brain, nor a white mouse brain.
2. Not only do I accept that we could develop a conscious robot, I think we already did it 🙂
In section 2, the reference is to a neural network that develops inside a robotic body, if that wasn't clear enough.
Miracles,
When a claim is said to be unfounded, it means that it is a baseless claim, just as the stories in the Bible, for example, are baseless stories.
Understand that there is a huge difference between:
1. A neural network like our brain network that developed from the beginning as part of our complex biological body, and learned from an early age to deal with the complex inputs that reach it from the different senses of this body, and then suddenly you give it different inputs that it doesn't recognize like inputs from virtual reality glasses that make it feel sick or discomfort.
2. A situation where from the beginning the network develops in a different body and learns to know the unique inputs it receives from it until it becomes a natural and inseparable part of it. This network will build and shape itself around the inputs of this body, they will not cause her any nausea or any sensations because this is the only body she knows, she was designed to deal with the inputs that come from it.
rival
Are you saying the claim has been disproved? by whom?
I say my claim has a lot of evidence. I suggest you read the book phantoms in the brain by a well-known researcher named Ramachandran.
After you read and understand, I have a whole library to give you.
Miracles,
"We have already agreed that in order to simulate the human brain, the body must also be simulated. Not only the muscles and bones - also the nerves, the endocrine system, the senses and so on. We know that a person will not remain sane without these things.'
I don't know what "we already agreed" is and what "we know" is, the fact that you repeat this claim in every conversation here does not mean that I agree with you or that this claim is true.
I think this claim is absolutely incorrect and I have already told you this in previous conversations. Yes, we will have to connect the neural network to some body in order for it to learn and develop, and develop things like consciousness, but this body really does not need to be at the level of complexity of a biological body, this is an unfounded claim that has nothing to rely on and in my opinion is absolutely incorrect.
rival
We have already agreed that to simulate a human brain, one should also simulate the body. Not only the muscles and bones - also the nerves, the endocrine system, the senses and so on. We know that a person will not remain sane without these things.
Try a VR game once and see how nauseating you get under certain conditions. Think of people with amputations (and even those born without limbs) - and think of the excruciating pain they live in. Cut off a person from his environment - and he will lose his sanity.
I could go on much longer - and I would still miss things. Will we overcome all of this in 7 years (it's already been 3....)? seriously?
Miracles,
1. One of the goals of the project is indeed the creation of new research tools for the study of the brain, but another stated goal is the creation of a complete simulation of the human brain within a decade. That's why I believe that most or at least a significant part of the researchers and scientists who are partners in the project also share his vision and believe that this is possible, even if it takes 20 years and not 10 years.
2. I really don't think it will be necessary to go down to the level of resolution of individual molecules or to the level of complexity of a biological brain in order to simulate the brain, they also understand this and therefore the system they are building allows for simulation at different levels of resolution from the macro level to the micro level. Once they understand the principles well enough, they will also know what can be removed from the simulation and what is not relevant without harming the results.
When you rush, you make a mistake - a million neurons per chip.
I fixed: 1024 neurons and 256 synapses per chip.
anonymous. Quantum computing will also give such power, I agree, but also a supercomputer based on synapse chips 48 million neurons and synapses per chip times 1000 chips. I was wrong - today I went through IBM's website, it is about 1 million neurons per chip and 256 synapses. And the supercomputer is based on 4096 chips. The chips consume only 70 milliwatts and these are low-power computers - just like our brains. And there is another way available even today for simple company employees like me that gives very nice results. Decentralization of calculation between regular computers. There are quite a lot of scientists in the field of artificial intelligence, and there was a series of articles that were very harsh but said we are not there. It's a beautiful world and I must say that I follow it even though I am aware that one day they will replace us. Like we replaced the dinosaurs. The only way to preserve the human would be to produce a cyborg. Of course, every day I think about the livelihood, but it happens.
rival
Many scientists think that the European project is not serious.
Even the members of the project itself do not think that they are going to build a brain in 10 years - they clearly say that what is important is the tools that will be developed along the way.
In any case - the numbers I mentioned are correct in my opinion, and as long as no one corrects me - I claim that we are very far from simulating a human brain.
A more correct description of the Hollywood movies would be as a kind of human mental simulations and not as a scientific theory
Since artificial intelligence is a human creation and also a learning system will use its basis in information
Collected by people, we may receive behaviors/simulations of human behaviors in robots, especially if we are outside
to the fast learning cycle and certainly bad results can accumulate for us in this system,
Because even people have difficult behaviors right now, a few kilometers north of us we are witnessing human hell
Our people have experienced horrors not short of the worst horror scripts of evil robots,
The difference is that bad people can be faced with good people and there is a chance to win
Against an intelligence that surpasses you in every possible parameter (it is not yet clear when it will have a subjective feeling)
Only an intelligence of that level could handle it,
It does not need to take over or replace us, it will only make us redundant
We are currently seeing the buds of this robotics but already at the basic level the tremendous speeds of processing
Information from the non-biological systems does not bode well for our ability to be meaningful in the future
We will look like a snail in terms of information processing for such a system, for every thought we have it will think a million thoughts
And that's before quantum computers went into full operation that seemed like alien technology
Connecting these 2 quantum and digital systems together will create a power that is almost unfathomable to human perception,
The interesting question is whether there will be any field where a biological brain can do better than artificial intelligence
Will it be at all relevant to development or will we become some side niche of the evolution that is helping us
jumped into non-biological matter and became like the human species that disappeared from the face of the earth,
As a programmer and developer of automation systems, it is very difficult for me to see how such complex systems can be controlled
Even in the complex systems we build today there are bugs and errors and we are always inventing improved methods to deal with the ever increasing complexity. It is difficult to see how you can control a learning system that comes to you
in which only the initial layer was written by, and perhaps also rewritten by repeated feeding by the system itself,
What is sure to be interesting in the future is the question of whether it will be good? And you have to invest a lot to make it good
Joseph
Computational power sufficient for the calculations you are talking about will exist when a quantum computer exists.
Until then - dreams don't cost money...
rival
Did it occur to you that maybe it's you who doesn't understand those scientists and a few other people who are discussing this with you?
Miracles,
I don't currently have too much time to actively participate in the discussion, but I'm just curious, what do you think all the people who are involved in this huge project, including computer and neural network experts and world-class brain experts, don't know or don't understand that you do?
As of today, there is no machine that passes the Turing test. Human robots exist and will continue to exist only in dictatorial societies (eg the Jewish Home).
Joseph
I agree with that.
Thanks for not mowing me down.
Clarifying my words: in my estimation, we are not a step away from creating consciousness. What is missing are theories. As convolutional neural networks came around 2004 and upgraded neural networks. Raw computing power will be. Is it dangerous? Yes. I think the article handles this nicely. Probabilistically, I think the chances are that it will go in a bad direction because there are many bad directions (getting wrong) and there are only a few or only one optimal routes. Second law of thermodynamics: there are many ways to break a glass, there is almost only one path for the shards to come back and be soldered together into a whole glass.
You have to be optimistic and believe in something. Don't owe God.
Joseph
The series you mentioned from amoeba to man is problematic - man is not more developed than amoeba. We are just as developed.
There is another persistent property of the neuron and the DNA molecule. The ability to survive and develop. Although not in contradiction to the second law of thermodynamics, the principle of increasing entropy, the two above-mentioned entities produce on a micro-environmental level that increases the order and the amount of stored information of itself, encoded into a small amount of cells, and survive 3.5 billion years on Earth. Asteroid, plagues, wars. and improve itself as a machine: amoeba, amphibian, dinosaur, mammal, monkey, great ape, human. It has a precise mathematical formulation in my opinion. I keep it not out of arrogance - I am aware of my smallness. I want to publish an article once and it took me a long time to understand. I believe that what took me 3 years to understand, I can transfer to others by copying in an hour.
The article started with the right question but could have been much more daring. Why is it so difficult to imagine a consciousness that is smarter than us in many things? We are the product of 3.5 billion years of slow evolution. Intelligent creatures but focused on themselves and have difficulty with panoramic thinking in time and space. It is not difficult for me to imagine a developmental evolution that will pass us by. In one synapse chip from IBM, there are 48 million neurons and synapses. Explanation by the way that about 1% of the brain is used, but we may be wrong. I agree that it will be necessary to develop theories of smarter systems than the artificial intelligence that exists today, to understand our intelligence, and we are at the stage of understanding sensing and decoding - not yet a consciousness like us. Do not copy the same connections. That's why 1000 chips are required to get the same number of neurons that you mentioned, dear Nissim (without double quotes) - I saw your response to me and it is acceptable to me, regarding homeopathy. In my work I developed artificial intelligence software for some difficult task. She does it with astonishing quality. I used a total of 20x8 neurons there, which is about 200. After we have a database of a billion neurons - which is what IBM is developing for DARPA - the agency of the American Ministry of Defense - a military system for tracking and eliminating terrorists. Now a deep learning network - a convolutional network - is produced in the connections. These are the complex connections we talked about. It has an unimaginable learning power. I compare this trillion with the 200 I use. With much less than that (about 50 million) you let an application like Watson play a trivia game, play a Hungarian cube arrangement not according to a fixed algorithm, chess, decipher an MRI of a lung, and many more dozens of applications. In all of them he learns from scratch and becomes stronger than a master craftsman. Therefore, even though I am a person who believes in the Creator, but also believes in the scientific method, calculation power, as you point out, already exists raw today.
After watching Yariv's clip, I noticed that if we replace the professor's voice with the voice of Rabbi Amnon Yitzchak - then it will still be just as convincing!
A/Rival
Our brain has 100 billion neurons and 100 trillion connections. On each connection there is a signal that is made up of levels at varying time intervals. The order of arrival of the pulses is important and not just the frequency. You have to simulate this activity at the same time, so it cannot be simulated with any software - you need a parallel system. Don't forget that these signals have a variable progress rate that also needs to be simulated - for each and every connection.
In order to portray a person - you need to know all these things at an exact given moment. Today we know how to read some information from individual cells in the brain, and read the average value of the information flow with a resolution of 2 mm in each axis. Keep in mind - there is an averaging here in space as well as in time.
If you want to simulate on a molecular level - you have to multiply the number of cells by another 100 trillion.
Can we be serious? 🙂
I recommend watching until the end.
Here it is, the segment starts at minute 2:57
https://www.youtube.com/watch?v=a1XcY-xAvos
A',
What you said is exactly what is being done in the human brain project (Europe's flagship project) where they create a complete simulation of the electrical and chemical actions that take place in the brain, down to the single molecule level.
They also simulate hormonal actions there (those that are supposed to cause the appearance of emotions) and not long ago they put their computerized brain into a state of "sleep" and as a result had some interesting insights, but I need to find the lecture again to remember what it was about.
Miracles
Why build such a thing?
You can run a simulation. As royal neural networks are not really built today, these are just a simulation of them. If you create a complete simulation of the brain on a computer, you don't have to include only the electrical operations. In the same way she will be able to calculate the chemical actions. (By the way, I don't think it will work. But time will tell)
rival
Come be honest and straight with yourself, at least…
You did not show how emotions emerge from a neural network installed on a chip.
Mainly, for the reason that there are no such things as you describe.
Put the irrelevant things - such as your feelings - aside, and try to use common sense and logic to formulate your answers. With all due respect. And there is respect.
rival
Our emotions are not exactly part of consciousness. Part of the emotion of love, for example, is related to the hormonal system.
The only way to get a "person" is by building a machine that is assembled like a person, down to the micro-organ level of each cell, the size of a person, that is sensitive to chemicals like a person, and sensitive to the environment like a person.
And if such a thing is built - it will be problematic like man, exposed to the same diseases, and will age within a few decades. He will also need to sleep 6 hours every night.
And of course - you have no assurance that some robots won't join Daesh, or vote for Bibi...
Correction - they can't.
Miracles,
You claim that emotions cannot emerge from an artificial neural network, all in all I presented you with a simple thought experiment that shows you a neural network on a chip that emotions emerge from.
What does it matter if this network is a replication or not, it is an artificial neural network from which emotions emerge as soon as it is activated and you claim that it is impossible.
And - duplicating a person on a computer chip is equivalent to duplicating a corpse.
After all, the robot needs "animals"..
It is not possible to copy and expect that from then on the work will manage on its own.. This scenario does not make sense.
What is possible is building from scratch. Just as you clone a sheep and start from a simple cell, so they will clone a brain and grow it on top of the robot. But this is where consciousness comes into play...
More than that - the main point is that these replications have been going on for billions of years.
After all, the offspring is a duplication of its parents, yet they are not identical. And the consciousness that develops in the offspring is individual.
rival
Decide - are we duplicating a person or does consciousness in a robot emerge from the complexity of the synthetic brain? Do you understand that these are two very very very very different things?
...and if that average person, after duplicating his mind on the chip, decided that he doesn't love his wife and wants to divorce her - will the robot also choose the same decision? And if not, why? What will cause the robot to change its behavior and not behave according to what is expected of it?
Miracles,
If we take an average person (who has feelings of course) and copy their brain one by one into a computer chip, then connect the chip to a humanoid robot, and place the wife of the person whose brain was copied into the chip in front of them, do you think the neural network that will run inside the chip will not feel strong feelings towards the same woman? She won't feel love for her?
If not, explain and justify why?
rival
Someone here wrote "an emotion like love is a tool designed to help us reproduce"... without the need or ability to reproduce I don't think such an emotion will develop.
Miracles,
In my opinion, as consciousness will emerge from the neural network, so will emotions emerge for them. They will be happy, and sad, and loving and angry just like us.
And if we annoy them then they will show us what it is.
Nostradamus; rival
'They will invade the Earth for resources'. 'They won't let humans reach Mars'... and if they build the robots on Earth then they won't try to take over the Earth? If they build them here, then the robots will allow humans to reach Mars? - You are both wrong.
No matter where they build the robots. And no matter where they put them, still, they will develop intelligence.
And similar to humans - robots are also expected to develop the trait of egoism. Which will not contribute to the relationship between humans and cyborgs.
The main question is:
How will it be possible to get along with the cyborg?
How to teach the robots to be kind to the environment?
First, when talking about artificial intelligence, you should be very careful of statements that "follow logically from the simple fact that...".
Second, you don't even need an artificial intelligence that is one-dimensionally plotted to achieve one or another feature in order to royally screw up the situation. It is enough to look at Enron, for example, a company that worked to obtain maximum profit for the owners under the guise of activities to achieve capitalist control of the electricity market, under the guise of streamlining the electricity market, and in practice succeeded in achieving terrible power outages in California (at a time when there was an excess production of electricity in California!), corruption on a scale never seen before So bankruptcy is one of the most glorious things the USA has ever known.
rival
An emotion like love is a tool designed to help us reproduce. I would not want to give up this feeling. And I certainly wouldn't want to give up what accompanies this irritation.
It is also clear to me that our brains are not designed for a very long life.
Therefore - robots as you describe cannot be a reproduction of humans. They will always be robots, and there is no reason to think they will be nice to us. You too, I'm sure, agree that their consciousness will be the result of emergence, and therefore we will not be able to know in advance what will come out, nor will we be able to control the outcome, or even understand the outcome.
jubilee,
These robots (which by the way can look completely human-like externally) will be our upgraded descendants, they will be our direct continuation even if they are produced in factories and not in the way that is accepted today.
I actually don't like the idea of Kurzweil.
No need to merge with the computers. The direct continuation of our lives are our children.
If there is eternal life then there will be no more children.
I like the prediction of Ray Kurzweil, in his opinion there will be no such competition because we will merge with the technology and become one entity.
Already today there are tens of thousands of people walking around among us with bionic hands and feet, with bionic eyes and hearing aids, with an artificial heart and even with digital chips in the brain that can be updated from the outside with new software (Parkinson's patients), this phenomenon will only continue to expand.
A lot of news lately about artificial intelligence, we are probably being prepared for something, society get ready.
And maybe they will see us as intruders and won't allow us to get there...
Because even the robots on Mars will eventually be able to develop and develop an independent will and possibly invade the Earth for resources
And why not develop such robots on Mars? And let them build the planet so that humans can also live on it?
Why not develop a smart robot and send it to Mars to learn there? Why on earth? Why not make the planet Venus or Mercury the home of the robots? They can visit Earth if they have been properly trained, for example. So why not?