September 10, 2015 at 17:44How to find the intelligence algorithm
- Transfer
In our blog, we talk about infrastructure virtualization and related technologies . Something interesting can be learned not only from experience in working with infrastructure projects, but also from theoretical work aimed far into the future. Today we decided to take a look at the book by Michael Nielsen, who discusses the topic of intelligence algorithms.
/ photo Sean Davis CC
As the author notes, the main reason for the study of neural networks lies in the hope that one day with their help we can go far beyond such basic tasks of pattern recognition. Perhaps this or that approach will lead us to the development of artificial intelligence that compares or exceeds human capabilities.
But is there a simple set of principles that can explain what intelligence is? Better yet, find an intelligence algorithm. It sounds too optimistic to be true, but there is hope that in the end everything can be explained by a set of simple but fundamental ideas.
Consider, for example, astronomy in the early stages or the countless chemicals that make up our world, so beautifully ordered by Mendeleev in his periodic table. Even if we move from this to the side, it is perfectly acceptable the existence of forms of intelligence that do not have a human way of thinking, but nonetheless surpass it in one direction or another.
A quantitative approach to assessing the functioning of the human brain can be considered from the standpoint of connectomy, which deals with the construction of connections: how many neurons exist in the brain, how many glial cells and how many connections are observed between neurons. Given the presence of billions of neurons, cells and connections, we clearly will not come to a simple algorithm for the work of intelligence.
In a second, more optimistic approach, the brain is considered from the point of view of molecular biology. The point of the approach is to find out how much genetic information is needed to describe the architecture of the brain. Here you can see that the genomes of humans and chimpanzees differ by about 125 million base pairs of DNA, and the similarity of humans to chimpanzees is about 96 percent.
Each base pair can be described with two bits of information - this is enough to indicate one of four types of bases. Thus, 125 million base pairs are equal to 250 million bits of information. Even if half of the difference is between the human brain and the chimpanzee - a total of 125 million bits or (relatively speaking) 125 million letters (or approximately 25 million words in English). Well, if you recall that we are talking only about part of the brain, it becomes clear that our genome should rather determine the overall architecture and basic principles that underlie the work of the brain.
The idea of the brain from the point of view of molecular biology simplifies the complexity of our description by about nine orders of magnitude. Although this is inspiring, we still do not get an answer to the question about the possibility of the existence of a truly simple algorithm of the intelligence. Can we make this description even simpler?
The experiment, published in April 2000 in the journal Nature, describing the “reconfiguration” of the brain of newborn ferrets, showed that the general principles of perception of sensory information can be based on different zones of the brain. Such a community, in turn, is an argument in favor of the idea of the existence of a set of simple principles at the heart of the work of intelligence.
On the other hand, we know that evolutionary psychologists have discovered a wide range of human universals, common norms of behavior for all people, regardless of their cultural affiliation or upbringing. It was found that many human norms of behavior are responsible for specific areas of the brain that are similar in all people.
Based on the foregoing, some people conclude that brain function has indivisible complexity, which, in turn, eliminates a simple explanation of the brain (and, possibly, a simple algorithm of the intellect). This point of view is supported by the researcher of artificial intelligence Marvin Minsky.
In the 1970s and 1980s, Minsky developed a theory called The Society of the Mind, which is based on the idea that the human mind is the result of the interaction of individually simple (but very different) computing processes that Minsky called agents. Having stated his theory in the book of the same name , Minsky comes to the following conclusion:
Of course, it’s hard to believe that we will find a really short Python program (either C or Lisp, it doesn’t matter) - say, up to a thousand lines of code — which will implement artificial intelligence, but you need to act as if an extremely simple intelligence algorithm exists. The key to artificial intelligence lies in simple, fundamental ideas, and we can and should be optimistic about their search, Nielsen concludes.
PS We share not only our own experience working on the service for providing virtual infrastructure 1cloud , but also the experience of Western experts in our blog on Habré. Do not forget to subscribe to updates, friends!
But is there a simple set of principles that can explain what intelligence is? Better yet, find an intelligence algorithm. It sounds too optimistic to be true, but there is hope that in the end everything can be explained by a set of simple but fundamental ideas.
Consider, for example, astronomy in the early stages or the countless chemicals that make up our world, so beautifully ordered by Mendeleev in his periodic table. Even if we move from this to the side, it is perfectly acceptable the existence of forms of intelligence that do not have a human way of thinking, but nonetheless surpass it in one direction or another.
A quantitative approach to assessing the functioning of the human brain can be considered from the standpoint of connectomy, which deals with the construction of connections: how many neurons exist in the brain, how many glial cells and how many connections are observed between neurons. Given the presence of billions of neurons, cells and connections, we clearly will not come to a simple algorithm for the work of intelligence.
In a second, more optimistic approach, the brain is considered from the point of view of molecular biology. The point of the approach is to find out how much genetic information is needed to describe the architecture of the brain. Here you can see that the genomes of humans and chimpanzees differ by about 125 million base pairs of DNA, and the similarity of humans to chimpanzees is about 96 percent.
Each base pair can be described with two bits of information - this is enough to indicate one of four types of bases. Thus, 125 million base pairs are equal to 250 million bits of information. Even if half of the difference is between the human brain and the chimpanzee - a total of 125 million bits or (relatively speaking) 125 million letters (or approximately 25 million words in English). Well, if you recall that we are talking only about part of the brain, it becomes clear that our genome should rather determine the overall architecture and basic principles that underlie the work of the brain.
The idea of the brain from the point of view of molecular biology simplifies the complexity of our description by about nine orders of magnitude. Although this is inspiring, we still do not get an answer to the question about the possibility of the existence of a truly simple algorithm of the intelligence. Can we make this description even simpler?
The experiment, published in April 2000 in the journal Nature, describing the “reconfiguration” of the brain of newborn ferrets, showed that the general principles of perception of sensory information can be based on different zones of the brain. Such a community, in turn, is an argument in favor of the idea of the existence of a set of simple principles at the heart of the work of intelligence.
On the other hand, we know that evolutionary psychologists have discovered a wide range of human universals, common norms of behavior for all people, regardless of their cultural affiliation or upbringing. It was found that many human norms of behavior are responsible for specific areas of the brain that are similar in all people.
Based on the foregoing, some people conclude that brain function has indivisible complexity, which, in turn, eliminates a simple explanation of the brain (and, possibly, a simple algorithm of the intellect). This point of view is supported by the researcher of artificial intelligence Marvin Minsky.
In the 1970s and 1980s, Minsky developed a theory called The Society of the Mind, which is based on the idea that the human mind is the result of the interaction of individually simple (but very different) computing processes that Minsky called agents. Having stated his theory in the book of the same name , Minsky comes to the following conclusion:
What is the secret of our intelligence? The secret is that there is no secret. The power of our intellect does not come from one perfect principle, but from their [principles] totality.
Of course, it’s hard to believe that we will find a really short Python program (either C or Lisp, it doesn’t matter) - say, up to a thousand lines of code — which will implement artificial intelligence, but you need to act as if an extremely simple intelligence algorithm exists. The key to artificial intelligence lies in simple, fundamental ideas, and we can and should be optimistic about their search, Nielsen concludes.
PS We share not only our own experience working on the service for providing virtual infrastructure 1cloud , but also the experience of Western experts in our blog on Habré. Do not forget to subscribe to updates, friends!