Machine intelligence requires neurological
COLWELL, BOB. Machine Intelligence Meets Neuroscience, Computer, 38 (1), 2005, 12-15
Shi Jun translation
Key Laboratory of Intelligent Information Processing of Chinese Academy of Sciences
Shij@ics.ict.ac.cn
When I accepted high school education early in the 1970s, we were the closest to regular contact with the computer, and we used the telephonic typewriter. These electric mechanical monsters can receive telephone transfer of ice-cold and slow (110 baud) and convert the bit into a variety of unharitable sounds. These noise noise is generated by a metal part of a metal part of a machine. Paper maintains the same consistency, and smoothes seems to be from the paper towel found in the living room. The data stream portion is generated by the sound migratory used at the time. I may also have a modem of 300 baud, which is converted by two electric guitars and two tennis shoes.
Earlier
My personal education begins in a friend. It is recommended that we sneak into the school's computer room to see what computer is like. After a ghost, I used the key that opened the school anything to open the door. We are close to those mysterious devices and explore the inside.
The keyboard or less is a bit familiar, but there are many parts that are more than any typewriter I have seen. I also noticed those "computer" to make a snoring sound, as if the electronic breathing. Just as I want to be close to see more, the whole lashup snacks, jumping, and goes forward to emit extraordinary sound waves. I don't know if it is normal, I just know that I have never heard the sound.
I know that I didn't touch anything, although I may really have ambiguous plan. I didn't see my friend touched anything. Therefore, I still face a big mystery: What promotes the monster to work? In short, what is it doing?
We cover your ears with your hands, close to see those texts being printed on endless tape rolled out from the head. At that time, I thought that the machine was printing text at a huge rate. But I know, the speed at that time and now it is just a joke.
I am surprised, what causes the machine to print those non-rules, or even understand the importance of the style or what "Basic" will appear on each page? The machine has been printed from the sky for a while, and then it seems to be a while, and more text is printed in the intelligence.
After I thought for a while, I guess certain someone controlled this machine somewhere outside the room, but what happened to what happened. Prerequisites the objective phenomenon, I am horrified that what wonder I can't understand.
What is intelligent?
After that, I want to know what is "smart" in the first place. I know that I will have a lot of good ideas, but I also know that I have enough ability to do a very stupid thing that is generally believed to motivation. So "smart" is something that we or what we did?
More broadly, what is the concept of intelligence, where is it from? To a certain extent, we are consisting of substances that operate in a predetermined manner - the minimum definition of a machine - this is very clear, the machine can have intelligence.
I am deliberately skipping some of the themes, such as the same thing (you are finishing is the fictional thing I think), mysticism (may be right, but can't lead to any basic analysis), There are countless external topics, such as schemeable, creative, and dolphins. If we try to patch up the evidence of the intelligent machine existing, why don't you understand a carbon-based human brain?
These sounds are very logical, even inevitably, and universities I am going to enter the study in this regard. Maybe I can help it in the manufacture of such a smart new product. This is really exciting! A few years later, I heard some universities tried to achieve the efforts of "artificial intelligence". Those who are very interesting in I have learned: we can learn a lot from itself, maybe it will also help us insight into different intelligence issues, and will help us create cheap and more powerful machines. This sounds reasonable, regardless of such a machine display, they may differ from human intelligence. Similarly, the artificial aircraft can't fly as birds, and the propeller of the ship cannot mimic fish in the water.
The artificial intelligence we want to achieve seems to be boundless. For example, when a programmer writes a computer program, he may generate a lot of errors. Why don't you let a smart machine you can write a perfect program in any time? These machines can also write poetry, stories, books, and music. If they give them chemical rules, they can also produce chemical formulas for manufacturing new drugs. They can solve the latest theorem of Fermat, think about the physical community's large-scale theory, tell us who will hide on the grass in the grass. This is how big is human intelligence!
But at 25 years ago, the event is not very clear. Research on artificial intelligence makes people learn a lot of intelligent knowledge. Intelligence is not a constraint computer to perform and clarify the task in a favorable manner, but is a new intelligence from the original goal. I have never completely lost my beliefs, and maybe one day we will eventually reach the goal. But with the birth of each year, this goal seems to be far farther away from us.
New understanding
In October 2004, the pioneer Jeff hawkins of the handheld computer broke into the field of artificial intelligence and enabling the field. The title in his new book cover is "On Intelligence", which proposes an understanding of brain new, which will result in a true intelligent machine.
In a hurry, it will immediately make the neurons since 1977 have not been stimulated, so that Hawkins will make them start again. For more than a decade, Hawkins has always been a fanatic fan of the nervous system. However, until today, it is only possible to invest enough time to seriously think seriously, and pay his point of view.
In addition to providing reviews in the field of artificial intelligence and neurological science, Hawkins also describes how his own cerebral skin is the nature of intelligence and intelligent machine, which will make our development of neurotropic models and operations (perhaps) He himself) and finds the way to achieve hardware that achieves this model.
Artificial intelligence researchers have learned a lot of knowledge about intelligence.
Intelligence and intelligent behavior
Hawkins started from the careful area of intelligence and intelligent behavior, which made me twice with him twice. The first time I saw the phantom of the telephonic typewriter during my high school (printing for a while, then thinking about it later, then print), the second time he analyzed the famous Turing test.
Just like Hawkins himself, I have never been satisfied with the expression of the map spirit. Intelligence does not make smart objects think they are also intelligent. At first, the proposal of Tuling gave me far-reaching insight, but
Over time, I gradually realized that this is mediocre and useless, more is escaping instead of useful definitions.
Hawkins describes John Searle's "China House" test, this test puts a smart body in the room, he can follow the mechanical instructions anywhere in the room, which will cause him to answer one correctly Series issues (these issues are written in Chinese who cannot be understood by the subject).
Selle pointed out that the recipient of Chinese output inferred that the subjects in the room must understand Chinese, possibly, and even show the deep insights in the surprising. But the main body doesn't do anything, - the machine can't do it - he simply follows the procedures written by some people. Sec said that he didn't know what smart, but this test indicated that the computer did not have intelligence. Sel's experiment clearly illustrates that at least in my opinion - Tuling test is easy to be fooled. We must do more efforts to do more than just some subjective judgments.
Others believe that the main, instructions, and the combination of rooms together reflect the intelligence - try to avoid the fact that we have to face the intelligence does not understand much, although we have long a long-term work in many scientific branches.
I don't want to cause an unnecessary argument, because I think so, there is no benefit to Hawkins want to lead us to the light. I will mention here because it considers what is intelligent and its physical representation, or the starting point of external performance.
Brain behavior
Hawkins believes that intelligence is a large number of neuron emerging behaviors, with a continuous constant prediction of future incidents with memory-based world models. He believes that the time is what the brain is doing and how to do the most important part. He believes that there are three critical components of the brain: the brain works in the input time series, involving many feedback (there is evidence that the neural network is constituted in this manner), the hierarchy of the real network has Mode, it seems important to their features.
I especially agree with Hawkins's view. The brain is essentially a smart device independent of our limbs and feelings. He reminded me that the brain is the only way to feel the feeling of feeling and think about the outside world. Each part of the brain continues to pass the related modes of real-time metrics in the external world. We consider visual, auditory and tactile, they are very different, but they will be delivered to our brain. We have evolved from millions of years, making our brain and feelings and movement control tightly. We are easy to get into the traps of Tuling, think "If it looks smart, he is intelligent"
Hawkins referred to the Hellen Keller. Helen Keller, although she was also afraid, but she learned the language and became an excellent writer. Yes, our brain uses the way to help us generate the sensor of visual and auditory sensors to the brain. But intelligence can significantly produce without those feelings.
Hawkins recommends that the neural network of the neuroplast is regarded as a distributed stored mode sequence, in the unchanged form, and stores, and arranges in a hierarchy.
This model gave me a huge feel. I think it can help us explain why so many smart people will think in similar thinking patterns: they can see, the mode under the very different transactions. When they don't know, there are many surfaces that seem very surprisingly related. According to Hawkins's expectations, these people may store specially constant conceptual expressions, and use it to form a large amount of shock-eye-eyeic ideas when necessary. I often feel that only a few ideas is that computer discipline is unique. For example, most of our current learning is just a new name in our minds in our mind, just whether we link them.
Neural network storage
When I read this book, I found that I personally endorsed Hawkins's point of view. For example, in order to help understand the sequence mode of neural network storage, he suggested that readers consider those who have happened in them and assume friends who tell them these stories.
This training reveals that the story must be regarded as a sequence from beginning to end. You can start from the middle, but you can't reverse tell stories. Each state "wake up" next state. This is also the truth that can be able to play long and complex music with memory. But you let them start from the middle part, they will fall into the dilemma.
Most of our current learning is just new names that have been stored in our minds.
I suspect that these the same phenomenon is the reliable way to remember the name of the song: They sing in the heart to the name of the songs through spiritual imitation, saved in the brain. Again, those continuous storage methods are Hawkins to help us understand. Hawkins also pointed out whether we are reading a story stored in memory, or written it or written.
We all take them from memory warehouses in the same way. Those information is together with other functions of the brain, help us to complete the handshake, properly move, recite, and mobile arms and arms to write.
According to Hawkins's theory, intelligent feature - it seems that the basic function of all the neural networks he proposed is to form a continuous prediction and correctly store this mode of the worldview obtained by the sense.
Write a policy
I have been above a class related to form, the program is to write a game program playing TiC-Tac-ToE. I spent a whole hour to consider how to identify, the best strategy encoding is to be applied to that game, but it seems that I am in trouble. After that, I realized that the game is so simple, I can let it go while "learning".
Essentially, the procedure I wrote is done as follows. It is always moved to the "most adjacent" position, and the other moves out. If you want to win, that looks like it is the easiest strategy. If the program is completed, it will lose every game. But this is a table handling course, so I have to write a list of each step of moving it over a game. After realizing the simplest adjacent mobile strategy, the program will check all previously failed game lists and refuse to make the same move again.
In fact, you can defeat the program in the game, but never have the same situation twice. Because TiC-Tac-ToE is such a simple program, the program will cost 10 to 20 games.
In the middle of that project, I clearly realize that not only this program doesn't have intelligence, but also some programs that are not very similar to it also do not have intelligence. The computer is not really playing TiC-Tac-Toe- I am. It simply follows the instructions, it has to be done because it is not terminated.
Intentionally
What we lack, I feel, those purposes rather than let the machine to do something. So I am particularly enthusiastic about Hawkins, intention is the characteristics of neurological activities.
Change this problem in a way. Imagine, a person is hired to guard a door to a school gym. This work is to look out the outside and inside. But there is a context here that that person understands what you should do, not to be told: the guard work is to protect those who are fit in the gym, and the safety of the whole room. And don't have to care about what people do in the house.
If there is any undesirable thing, such as a fire, a guardian will listen to the arrangement of authoritative people, do not need deeper understandings. Now, imagine if it is a computer to watch the same gate - it will reliably fulfill its responsibilities under normal circumstances, and ignore the watchman to sleep due to fatigue. Once the situation changes, for example, a fire has occurred, or a police officer suddenly requires an incorrect thing. It doesn't really understand context, and there is no purpose of unconscious work, intention is very critical.
Lost content
After carefully study, the expected observation and the rapid thinking model of the reader, Hawkins changed the direction of the way in the original book. He sets the topic as a smart machine looks like, the obstacle to manufacturing such a machine, and it will bring benefits in those aspects.
A famous cartoon drawn drawn by Sydney Harris has described two scientists standing before writing two blackboards with different equally equally equally equally equally equally equally equally equally equally equally equally, and the equation is separated by a set of words, "then is a miracle". The title is "I think your second step is more clear". I feel like I have passed the quantum tunnel after reading this book. There is a chapter lost. Compare his estimates, stable, ruthless, group facts, and explanation of his theory, the last part of the book is like a cousc in someone's fantasy. In a fair man, I don't think that Hawkins has solved each of the latest problems before publishing this book, which is the challenge to him to make problems. These issues may be in "Step 2" - Reduce the neural network of the cerebral skin can form a silicon-based machine idea.
Maybe I am responding to the intelligent neurons that Hawkins awakened in my machine. I hope that Hawkins is right. I think in the future, the machine will not be very stupid, but intelligent and we work together. Will the personality will appear? Will continuous evolution or can it be avoided?
Ask such a problem, it looks like science fantasies. But after reading the work of Hawkins, we will gramify the power of these fantasies again. Let your neurons get nutritious!
