False emptiness
About a week ago, I read The Empty Brain essay by Robert Epstein, Ph.D. from Harvard University and a leading research psychologist at the American Institute for Behavioral Research and Technology in California. In spite of the fact that this is not typical of me, I decided to write an answer and affirm that the original essay does not at all reflect the real state of affairs.
In his essay, Robert Epstein argues that the mental model of the human brain and thinking processes cannot be compared with the PI - information processing - occurring inside a computer.
It seems that the author describes the IE just as a set of predefined algorithms laid down by programmers, or as a process of recording photos on a hard disk. Therefore, he argues, we are mistaken when we use this abstraction to describe our brain.
')
Robert Epstein also cites examples that at first look very convincing, but, in fact, mislead the reader, because it is difficult to imagine a person thinking like this (and I don’t know any scientist whose opinion would resemble it):
Well no. This is not how it works.
The human brain does not have a degree in physics. It develops on the basis of what a person perceives, uses input data received from the senses, and is improved through countless attempts to do something, receiving positive or negative feedback.
Obviously, the physical and mathematical algorithms are not intended to be used inside the brain, since they were invented to model the world in a very precise manner, whereas we are very good at approximate calculations.
However, the fact that in the head of a baseball player is not a mathematical formula does not mean that the algorithms do not participate in decision making.
But why standard computer algorithms? After all, we have our own, growing themselves within our brain, those that rely on a wide range of perceived information, and not on the physical and mathematical model of the world.
There are many different technologies behind this term, and, importantly, the brain's work is not limited to the work of neural networks , however, I will briefly describe them. Based on the usual low-level computer logic, neural networks create an abstraction that is completely different from saving photos or creating PowerPoint presentations.
In general, neural networks are trying to find a solution to a complex function, not knowing the function itself. For example, the function is to answer Yes if the object is present in the photo, and No if it is not there.
Since the neural network does not know what the function is, the only way to make it work is to train with a set of signed photos. Each time a neural network receives a photo, it answers Yes or No with some degree of confidence, and we confirm or reject the answer. If the network answered correctly, everything is OK, otherwise it changes itself to find a way to give correct answers.
If we take a closer look at the neural network, we will see a set of simple mathematical equations, each of which answers in a certain way to the received part of the input data. Each individual neuron has no idea what is happening in the whole network.
In a very simplified way, this is how this metaphor works . Do you see? This is far from the point of view of the author. No pre-built algorithms, only those that were generated in a natural way, and which, by the way, are for the most part unique ...
Further in his article, Robert Epstein says that " there is no reason to believe that the two of us will change the same, having received the same experience ." I fully agree with this statement, really. But here lies the fact that struck me so much that I started writing this post:
If we run two identical neural networks on two identical computers, and then give them different input data (as their “lives”), we will end up with two different neural networks. After that, if we give them more input, identical this time, they will react differently, and the changes that have occurred to them will also differ.
It is a mistake to think that computers are not capable of working with fuzzy logic and noisy data. It’s just a matter of size and organization, and it’s widely known how many neurons are in the human brain, and how many connections there are between.
And, although the whole post was non-aggressive, now I’ll be a little mad. The last paragraph brought me almost holy anger. Just look at this:
Few insights, if any? Is it really? A good part of machine learning is based on this metaphor, a huge amount of AI research is based on the study of the human brain, and is that a little insight?
This last statement looks to me like an offer to surrender in brain research. Why bother? We are not computers, that there!
It is true that the human brain is much more complex and mysterious than the most advanced modern computers. Its structure and processes are not yet known, but - like many other phenomena - not unknowable.
And even if our brain is a computer, it is beautiful. Get over it, please. There is nothing ugly or bad about being both a computer and an organism .
Ps. I knew about neural networks, but not how they work. Therefore, thanks to Alan for explaining and correcting errors.
Brief introduction
In his essay, Robert Epstein argues that the mental model of the human brain and thinking processes cannot be compared with the PI - information processing - occurring inside a computer.
It seems that the author describes the IE just as a set of predefined algorithms laid down by programmers, or as a process of recording photos on a hard disk. Therefore, he argues, we are mistaken when we use this abstraction to describe our brain.
')
Robert Epstein also cites examples that at first look very convincing, but, in fact, mislead the reader, because it is difficult to imagine a person thinking like this (and I don’t know any scientist whose opinion would resemble it):
A glance from the PI requires the baseball player to estimate the set of initial ball flight conditions: the impact force, the angle of the trajectory, this kind of thing - then to create and analyze the internal model of the path the ball will move, and then use this model, constantly adjusting its movement to catch the ball.
Well no. This is not how it works.
Primitive algorithms
The human brain does not have a degree in physics. It develops on the basis of what a person perceives, uses input data received from the senses, and is improved through countless attempts to do something, receiving positive or negative feedback.
Obviously, the physical and mathematical algorithms are not intended to be used inside the brain, since they were invented to model the world in a very precise manner, whereas we are very good at approximate calculations.
However, the fact that in the head of a baseball player is not a mathematical formula does not mean that the algorithms do not participate in decision making.
But why standard computer algorithms? After all, we have our own, growing themselves within our brain, those that rely on a wide range of perceived information, and not on the physical and mathematical model of the world.
Neural networks
There are many different technologies behind this term, and, importantly, the brain's work is not limited to the work of neural networks , however, I will briefly describe them. Based on the usual low-level computer logic, neural networks create an abstraction that is completely different from saving photos or creating PowerPoint presentations.
In general, neural networks are trying to find a solution to a complex function, not knowing the function itself. For example, the function is to answer Yes if the object is present in the photo, and No if it is not there.
Since the neural network does not know what the function is, the only way to make it work is to train with a set of signed photos. Each time a neural network receives a photo, it answers Yes or No with some degree of confidence, and we confirm or reject the answer. If the network answered correctly, everything is OK, otherwise it changes itself to find a way to give correct answers.
If we take a closer look at the neural network, we will see a set of simple mathematical equations, each of which answers in a certain way to the received part of the input data. Each individual neuron has no idea what is happening in the whole network.
In a very simplified way, this is how this metaphor works . Do you see? This is far from the point of view of the author. No pre-built algorithms, only those that were generated in a natural way, and which, by the way, are for the most part unique ...
Uniqueness of experience
Further in his article, Robert Epstein says that " there is no reason to believe that the two of us will change the same, having received the same experience ." I fully agree with this statement, really. But here lies the fact that struck me so much that I started writing this post:
If you and I attend one concert, the changes that take place in my brain when I listen to Beethoven’s 5th symphony will almost certainly be completely different from the changes in yours. These changes, whatever they may be, are based on a unique neural structure that already exists, and each of the structures has evolved over a lifetime of unique experience.
If we run two identical neural networks on two identical computers, and then give them different input data (as their “lives”), we will end up with two different neural networks. After that, if we give them more input, identical this time, they will react differently, and the changes that have occurred to them will also differ.
It is a mistake to think that computers are not capable of working with fuzzy logic and noisy data. It’s just a matter of size and organization, and it’s widely known how many neurons are in the human brain, and how many connections there are between.
Insights
And, although the whole post was non-aggressive, now I’ll be a little mad. The last paragraph brought me almost holy anger. Just look at this:
We are organisms, not computers. Get over it. Let's continue to try to understand ourselves, but not being burdened with unnecessary intellectual burden. The metaphor of the PI has now been half a century, and it has provoked little insight, if any. It's time to press the DELETE key.
Few insights, if any? Is it really? A good part of machine learning is based on this metaphor, a huge amount of AI research is based on the study of the human brain, and is that a little insight?
This last statement looks to me like an offer to surrender in brain research. Why bother? We are not computers, that there!
It is true that the human brain is much more complex and mysterious than the most advanced modern computers. Its structure and processes are not yet known, but - like many other phenomena - not unknowable.
And even if our brain is a computer, it is beautiful. Get over it, please. There is nothing ugly or bad about being both a computer and an organism .
Ps. I knew about neural networks, but not how they work. Therefore, thanks to Alan for explaining and correcting errors.
Source: https://habr.com/ru/post/394957/
All Articles