Evolutionary Artificial Intelligence
Nowadays, everybody is talking about Artificial Intelligence (AI). It is like a festival everyday. New tools, new approaches, new software… Despite some of our fears, it is the new magic of our usual life. On the other hand, I feel different from happiness and fear. I feel that we are stuck in a dead end street while we are trying to create AI. I feel like that because; I know that we can’t see the front enough when we drive too fast. At such times, we should stop and think about what are we doing.
First of all, many people think that we had good progress in artificial intelligence. We have self-driving cars, face recognition systems and even deep fake porn videos. Besides, there are many new areas which haven’t been worked on yet.
I do not doubt that everybody agrees we have some kind of intelligence. Certainly, they are artificial because we have built them by machines. But I think that we forget our main target while trying to solve tiny problems. Can you remember what our main target was?
A. M. Turing published a paper which is called “Computing Machinery And Intelligence” in 1950 . He asked a very simple question in his paper; “Can machines think?”. After this question, humanity has started its most inspiring journey. But after 70 years, today, I want to criticize the point where we have arrived. I don’t think we have machines which can think. Moreover, with current methods, I don’t think that we will be able to create a machine which can think in the future.
In this article, I will try to explain why we are going to neglect to build machines who can think and I will suggest another approach.
The best way of thinking about something is starting to try to understand the terminology. Without understanding the essence of the words, we can’t think about ideas. That’s why firstly we should define basic terms. We should answer those questions; “What is artificial?” and “What is intelligence?”.
I use Cambridge Dictionary online to find the meanings of words. According to this, artificial is something made by people, often as a copy of something natural . This is a great definition because it says that there should be something natural to make a new copy by people. So what does natural mean? It is something which can be found in nature and not involving anything made or done by people.
At this point, we know that what artificial and natural mean. When we use artificial word with intelligence, intelligence should be the natural one according to the definition. When we look at the definition of intelligence, clearly we see that is true. Cambridge Dictionary says that intelligence is the ability to learn, understand, and make judgments or have opinions that are based on reason. Shortly, we are trying to create artificial intelligence which can learn, understand and make judgments. Best natural intelligence in the world is the human brain. It means that the main target is to copy the human brain.
Until today, we had been trying to create an artificial brain which can learn, understand and make judgments. Because if we could do that, it means, we would have a real artificial intelligence. Our main target was to build something like the human brain. The problem was that it is too complicated to be designed by humans. Up to the present, we only made some tiny problem solvers. Today’s artificial intelligence is not even close to the human mind.
In our studies, we focus on the result of the journey and miss the journey itself. Think about the natural one which we want to make an artificial of, human intelligence. It hasn’t been developed by other intelligent creatures. It is a completely natural result of nature. It is the magical presence of the evolution. But we think that we can design it simply. To accomplish this, we are trying to create learning models when we face a real-world problem. With this method, we can solve tiny problems for a short time. At the end of the day, we will not have an artificial intelligence like the human brain.
First of all, there are billions of problems in the universe. Because of the eternity of the universe, we know only a small amount of these problems. Even we can be able to know, we have to create new learning models for every specific problem. If we create a wrong learning model, we can get the wrong solution. Also, there is not any idea about how we can bind all these learning models in one brain in order to create one intelligence.
We should ask ourselves, do we want to create a real artificial intelligence?
Why Is Human Brain Perfect?
In this section, I want to clarify some points. When we create a learning model in the traditional artificial intelligence way, we can call that learning model as a brain. We can train that brain very well if it is perfect. The bad part of traditional methodologies is not the training process. They are bad because we aim to control how a brain should be made. We want to have a perfect brain. But the human brain is too complex and we barely know how it is working.
Let’s think about the human brain for a while. Why it is so precious? How could be so clever?
Human brain -or any kind of creatures’ brain- has been developed by nature in under the circumstances of evolution. It hasn’t been designed by anybody. It didn’t have any aim. The brain we have has a perfect learning model today but it is not here for any specific reason. We have it now because it helped us to survive. That’s why I claim that evolution idea should be the key element of creating a real artificial intelligence.
Evolutionary Artificial Intelligence
You would discover natural selection when you think why a lion is so magnificent. But once upon a time, they weren’t as perfect as they are today. Circumstances were so hard. If a creature doesn’t adapt to its environment very well, it would became extinct. They have to have special skills to survive in their nature such as camouflage, hunting or speed. With this way, we can design an artificial intelligence which could be perfect if we create an evolutionary environment for it.
To create an environment for artificial intelligence, we should understand basic facts of evolution. All parts of evolution should be simulated on computers because our main target is making a copy of real world. Definition and finding equal corresponses of evolutions are our basic steps to create evolutionary artificial intelligence.
In real world, there are organisms which are single living plant, animal or virus . Those organisms live on a planet like earth. Also there are many stimulants which can affect the organism. When an interaction takes place, the organism responds. This process is called cell signaling . Besides, those organisms reproduce themselves. While they are creating a new offspring, DNA chains of ancestors are used. Every new offspring should be the same as its ancestor. But in real world, there is a mutation factor which disrupts the DNA structure. This new organism with a new DNA chain is tested by the environment about how suitable it is to live in the environment. This is called natural selection. After this life cycle has continued for a long time, different kind of animals have appeared. One of them is us which has got a super brain which can learn, understand, and make judgments. And I claim that if we can create same structure on computers, we can produce an artificial intelligence on machines.
In our virtual environment, the artificial intelligence doesn’t have to be a physical creature such as a bug, a lion or a human. We aim to create an intelligence. To make it, we don’t have to have a real body. Instead of creating a body -like a robot-, we can simple create a computer program.
So we can name a simple program as an organism. When a program has been executed, executor of the program (physical presence of computer) would be the planet. In this case, stimulants would be every signal in the system such as camera signals, microphone signals, every data transmission in ports, the data which is located in RAM or HDD. When a stimulant has been processed, the program will produce a new response for the computer. As a principle of evolution, organism doesn’t have any purpose in its actions. All actions are generated randomly. Also, the program should have an ability to create a copy of itself so that it creates new offsprings. While it is creating a new copy, it should change its structure too. Because organisms have DNA chains which define the behavior of the organism. In our virtual world, the program should have its DNA structure. Maybe, the code of the program can be use as DNA. With this method, every program should have its code. Also every program should be able to create a new copy with its code. Of course, as a key point, if we are talking about evolution, while the program is creating a new copy, it should mutate the code randomly, a little bit. As a result, a new copy of the program should be almost the same as the ancestor.
You might think that a system like this can turn to a garbage. You are right because, probably most of programs will be turn to garbage which is something can’t be executed. But in this randomness, some programs which have some ability to survive may occur. This little chance is the hearth of the evolution which has created human brain.
Of course, this method can have some problems in it. That’s why we should start to criticize the theory.
The Problem Of DNA
As I mentioned before, the code of the program can be used as DNA. The problem is that most of the modern programming languages are so fragile. Even a missing comma can break the compiling process. In the copying process, the code should be mutated randomly. But most of the new versions will not be able to compile. Even stillbirth scenario is a something which we see in real life, at least one of the new version should survive. That’s why low-level programming languages, like Assembly, can be more useful to use as DNA.
On the other hand, creating a completely new DNA structure which is less fragile can be efficient too. The new structure should be able to mutate easily and should be less fragile than high-level programming code.
Whichever way (low-level programming language or new DNA structure) we choose, every program should have a DNA which carries information about its behavior.
The Problem Of Time
Evolution started almost 5 billion years ago. The human brain is a very new concept for evolution. The question is, do we need same amount to create a new copy of the human brain? We can create a new version of a program in milliseconds. But we can’t wait another 5 billion years. It is too much time for us. That’s why the problem of time is a very important drawback for us.
Also, I don’t think that this is something which is computable. Because of randomness, we can’t know it will take almost the same time. On the other hand, it doesn’t have to follow the same evolution path. Probably, it will follow a completely different style.
The Problem Of Random Infinity
Hearth of the human brain’s evolution was randomness. But it can be bad as it can be great. Because, if we talk about creating new DNA structure with randomity, I am sure that we will have different type of programs. In their nature (physical presence of computer), we can design a system which can kill them. In those circumstances, some of new programs types can adapt to that environment. We can observe the evolution of new creatures. But they don’t have to follow the same evolutionary path as the human mind. There is no obligation to follow the same path.
First, we can’t create the same environment as the evolution of the human brain has. The environment could be similar but not the same. We can even create almost the same environment (which is not possible), randomness will still follow a different path because of infinity. We can’t know what kind of creatures we will have. We can’t know what kind of intelligence we will have. It is a completely blind shot.
The Problem Of Security Of Humanity
Because of the problem for random infinity, we can’t be sure what we will have. There can be some great creatures which can live on a computer. But also, there can be some destructive creatures too which want to destroy everything. We can create an environment which is closed outside to block undesirable results. But it will reduce the speed of evolution. The speed of evolution would increase as long as creatures can have the data.
We can face the fact that computers destroy humanity.
There are many artificial intelligence methods which we can use today. Most of them are so powerful to solve problems. Probably we will make our life easier with those artificial intelligence applications. After all, I am not happy with this result because they can’t think like the human mind.
I suggest a solution to solve this thinking problem. I claim that we can create artificial intelligence with a very old method, evolution, which is responsible for creating a human mind. I claim that artificial intelligence can occur randomly on a computer if we apply basic evolution rules. On the other hand, I accept that there are some possible problems which should be solved in practice.
The human mind is the most exciting result of evolution. Because it can compose operas. It can write sonnets. Moreover, nobody organized the human brain in order to do those activities. A human mind does this because it can think. What a wonderful thing it is to think, isn’t it? If nature has built it once, it can do it one more time for us.
Maybe the evolutionary artificial intelligence will be the best teacher for us to love each other more despite the fears we have. If you think the same, please contribute to the idea as much as you can. If this is a bad idea, please explain why, and we will forget this theory.
(This post was originally published on ozguradem.net. I also appreciate to my friends (Furkan Başaran, Alperen Yılmaz, Melih Çaylak, Fatih Boyacı and Murat Palankalılar) who reviewed this article before I publish.)