Just like a shot of a strong heart, when the world’s first optical quantum computer that surpassed the early classic computers turned out, people’s expectations of the artificial intelligence era seemed to have more enthusiasm: beyond the classic quantum computer Is the quantum computer that beats the supercomputer far behind?
Once the latter is realized, human beings will once again be proud of their computing power and pry into the mysteries of the human brain, thus removing a major obstacle to the study of artificial intelligence. At present, in the face of the human brain, this guy, who weighs only about 1.5 kilograms but has 1011 neurons, makes humans helpless. To simulate the computing power of the whole brain, any computer in the world is currently incapable.
In the recent frontier forum on science and technology of brain science and artificial intelligence sponsored by the Chinese Academy of Sciences and the Institute of Automation of the Chinese Academy of Sciences, many people in the industry have raised the rumor: building a new computer cluster that supports deep learning. Has become an inevitable choice for some artificial intelligence research, then the artificial intelligence research does not need the computing power like quantum computer?
"Our scientists today, especially computer scientists, have used 'calculation' too much, and their dependence on 'computation' is even somewhat 'insatiable'!" Li Deyi, academician of the Chinese Academy of Engineering and chairman of the Chinese Artificial Intelligence Society, was on the forum. Poured cold water for everyone. In his view, artificial intelligence scholars can not only focus on "computational cognition", but also ask how fast the "human brain" research is going, but to put more energy into "memory cognition" and "interaction recognition." Know".
There are not many brain sciences that can inspire artificial intelligence?
Li Deyi is not interested in "computational cognition", but also from a report from Google (microblogging) company -
On May 15, 2015, Google said that the company's driverless cars have millions of miles of testing experience, roughly equivalent to 75 years of human driving experience.
“How did the 75-year driving experience 'calculate'?†This led to Li Deyi’s thinking: When the unmanned vehicle is on the road and the driver’s license is put on the agenda, the driving “measurement†has become a top priority for the traffic control departments of various countries. How do you know how to measure? Information is measured by "bits", energy is measured by "joules", then brain cognition?
It seems that the brain science scholars did not give such an answer, and artificial intelligence scholars could not get any inspiration.
This became a metaphor: brain science, artificial intelligence, two frontier disciplines of the same 21st century, which have been relatively independent and rarely crossed in the past few decades.
Pu Muming, a foreign academician of the Chinese Academy of Sciences and director of the Institute of Neuroscience of the Chinese Academy of Sciences, also mentioned in the forum that day, whether it is domestic or foreign, but with the continuous enrichment of research methods, the research field continues to break through, the intersection of the two Fusion has become a hot spot, and even a new research term has emerged, brain-like intelligence. The United States and the European Union have successively launched relevant research programs, and China has also launched a brain plan. He said that China's plan is to combine brain science and artificial intelligence most closely.
For example, the popular deep learning is an application based on artificial neural networks, which can be inspired by some laws of neuroscience. Pu Muming said, for example, can learn from the synaptic plasticity, memory storage, extraction and regression, and so on.
However, he also admits that the current brain science research can not inspire artificial intelligence.
Pu Muming gives an analogy. The current brain science research is only equivalent to the research level of physics and chemistry in the late 19th century. "To fully understand the brain, it may be centuries, not in this century." ," he said.
Then why do brain-like research, Pu Muming said, we must do some appropriate application at this time, if we do not apply the knowledge already known to the diagnosis, intervention and treatment of brain diseases, then by 2050 our The medical system is likely to face a collapse - then you will find that there is still no brain disease that can be cured.
Accordingly, the same applies to the application of artificial intelligence . He said that it is not necessary to fully understand that some of the neurological sciences have staged results and can also provide inspiration for the development of artificial intelligence.
What is the most important intelligent behavior of human beings?
On the basis of the existing research, Tan Tieniu, a member of the Chinese Academy of Sciences and a researcher at the Institute of Automation of the Chinese Academy of Sciences, came to the conclusion that "pattern recognition" is the most important intelligent behavior of human beings and an important research content of artificial intelligence - the "model" of the machine. The ability to identify "to a certain extent or to a large extent reflects the degree of machine intelligence "humanoids".
In the forum of the day, Tan Tieniu gave several examples of pattern recognition. For example, speech recognition, in recent years, the rapid development of Keda Xunfei, can translate Uighur into Chinese, Chinese into Uighur; and like gait recognition, when you can not see faces, irises and fingerprints, you can pass the steps The state perceives its identity within a few tens of meters.
In addition, there are image recognition, which is representative of face recognition. As early as a few years ago, Ma Yun (microblogging) brush face payment has already detonated public opinion hotspots. Tan Tieniu himself is conducting research on iris recognition and has established the largest shared iris image library in the world, which is shared by many countries. He said that this can be used not only on mobile phones, but also in finding lost children.
Tan Tieniu said that the technical bottleneck of pattern recognition can be improved by referring to the mechanism of biological, and the future bio-inspired pattern recognition can be expected in the field of artificial intelligence. The ultimate pursuit is to simulate the approaching pattern recognition, which is a very difficult process.
He also mentioned that the main bottlenecks of pattern recognition are robustness, adaptability and generalization.
Robustness, to put it bluntly, is that artificial intelligence is "not enough skinny". "Is it a little disturbing, it will go wrong." Tan Tie Niu gave an example, such as chatting at a reception, and having a lot of background noise. If you want to hear the voice of one of them, you should ignore or suppress the interference of other conversations in the background - humans can do this, that is, The auditory system is called the cocktail effect, but can artificial intelligence be?
The so-called adaptive, it is easier to understand, Tan Tieniu said that human eyes will adjust with changes in lighting and environmental changes, which shows that the adaptability is very strong. This can be applied to artificial intelligence, such as face recognition. Have a friend who has not seen it for more than ten years or even decades, can you recognize it again? He said that existing pattern recognition is not ideal in this regard.
It can be generalized. To put it bluntly, it is to say "one in the third." Tan Tie Niu said that when a child knows an apple, even if he only remembers it once, he can recognize other types of apples. This shows that when humans see a thing, they not only know it, but also know why. And knowing why, is the "deep learning" in the field of artificial intelligence. However, the current deep learning of artificial intelligence must be based on a large amount of data, which needs further study.
Tan Tieniu said that to solve these three problems, the key is to look at human beings. At the micro level, the pattern recognition of artificial intelligence can learn from human neurons, which are excitatory, inhibitory, functional plasticity and disseminative. Scientists have been inspired by this to enhance the stability of the pattern recognition dynamic system.
Driverless is the breakthrough of artificial intelligence?
Li Deyi has found a breakthrough in practice: automatic driving. He said that whether it is dialogue, poetry or driving, the Turing test allows the tester to intervene on-site, and the results are approximate and subjective. However, compared to dialogue and poetry tests, the Turing test of driving can be more accurate and objective.
He said that when the car was invented, people were most interested in the structure, machinery, transmission, tires, chassis and body of the car. By the 20th century, people were interested in engines, carbon emissions and passive safety. By the end of the 20th century and the beginning of the 21st century, people generally care about three things, lightweight, clean, and intelligent.
The so-called intelligent, in his view has four stages, the first is rational assisted driving, mainly driven by people; the second is automatic driving, local hands can open hands and feet; the third is automatic driving, that is, automatic Driving to take over the driving right; the fourth is man-machine cooperative driving.
In Li Deyi's view, it is difficult to be anthropomorphic without driver.
He lamented that the car evolved from a horse-drawn carriage. As a power tool, the horse's horsepower can reach 100 horsepower, but the car is far less suitable than the horse's ability to adapt to different loads, weather, roads, and different vehicle conditions. To put it bluntly, the car's perception and cognitive ability are far less than the cognitive subject of the horse. "The old horse knows the way, the car is not as good as the horse!"
Li Deyi said that the fundamental problem is not the car but the people. To solve the human problem, the driver's cognition can be replaced by the robot , so that the robot has the ability of memory, decision-making and behavior, so the new concept is produced - "Driving the brain."
"Driving the brain" is not equal to the driver's brain. "Driving the brain" is the intelligent agent to be the driver. To complete the driving cognition including memory cognition, computational cognition and interactive cognition, he said, this should It is one of the most significant topics in the era of artificial intelligence.
ceramic piston
Yixing Guangming Special Ceramics Co.,Ltd , https://www.yxgmtc.com