Lei Jun's annual speech, robotic brain is comparable to humans?Expert: impossible at all

In the 2022#Lei Jun's annual speech#, Xiaomi Lei Jun showed the full -size human bionic robot "Iron", 1.77 meters high, weighing 52kg.

Lei Jun said that "Iron University" has a smart brain, can perceive 45 human semantics, and distinguish 85 environmental semantics; the cerebellum is also particularly developed, Xiaomi self -developed systemic control algorithm, coordinating exercise 21 joint freedom; visual acuity: MI Sense's visual space system, three -dimensional reconstruction of the real world; strong limbs: 5 kinds of joint drives throughout the body, peak torque 300Nm.

However, in the eyes of AI experts, it seems like a show. Because of the development of today's robots, it is far less optimistic as he expressed.

There is a sentence worth remembering in "AI3.0": We always mistakenly believe that AI has solved simple problems, which is equivalent to making progress on the problem. In fact, for a complex technology project, it takes 10%to complete the previous 90%work, and it takes 90%to complete the last 10%.

AI has no "understanding" at all. Their only ability is "seeing a lot of knowledge". Because the basis of understanding is meaning, and meaning is the real obstacle of artificial intelligence.

Although a smart machine can quickly traverse more possibilities in the pattern space, scan the corners of the edges, and then show more and more surprising skills, or even more than human performance, they still seem to be like "Dog's Eye Seeing the Stars" does not recognize what "Star Map" is.

Recommended reading

01

How far is we from the AI ​​of human level?

There are basically two views about this issue.

In the first point of view, I will quote the question of the director of the Ellen Institute of Artificial Intelligence, Olun Ezionni, to answer this question: "Make your estimation, extend to 2, 3 times, and then extend to 4 times to 4 times to 4 times , It's time to achieve it. "

The second point of view is to review the evaluation of Andre Capabis in the previous chapter: "We are really far away." This is also my point.

The word "Computer" originally refers to people at first. In fact, it usually refers to women who perform computing tasks for those artificial or desktop mechanical calculators. During the Second World War, they helped soldiers to complete the artillery aiming related to the artillery. The computing task of missile trajectory is the most primitive meaning of a computer.

In the mid -1940s, electronic computers replaced humans in the field of computing and immediately became "super humans." Unlike any human "computer", these machines calculate the speed of a flying artillery trajectory even more than the speed of the shells. This is the first one in many subdivided fields that can show well. Today, computers with the most advanced artificial intelligence algorithm have conquered many other tasks in other subdivided areas, but General Smart has not yet been realized.

Human intelligence is an incredible, delicate, and unbelievable thing, and there will be no danger of being copied in the short term.

"Forecasting is difficult, especially for future predictions." This is the most famous saying that it remains to be verified, but this sentence is correct whether it is in the field of artificial intelligence or any other fields.

Several investigations for artificial intelligence practitioners, about when the general artificial intelligence or super intelligence arrived, the results were quite broad, from the results of "will appear in the next 10 years" to "never appear". In other words, we have no clue about this.

02

Does it really exist to fully autonomous car technology?

This depends on how you define "autonomous driving". The National Highway Traffic Safety Administration defines 6 automatic level 1 for vehicles. I tell here as follows:

Level 0: Human drivers perform all driving tasks.

Level 1: The vehicle can occasionally support the human driver by controlling the steering wheel or the speed of the vehicle, but it cannot be carried out at the same time.

Level 2: In some situations (usually on highways), vehicles can control the steering wheel and speed at the same time. Human drivers must always maintain high attention, monitor the driving environment, and complete other behaviors required for driving, such as changing lanes, driving away from highways, parking when encountering traffic lights, and concession for police cars.

Level 3: In certain specific situations, vehicles can perform all driving behaviors, but human drivers must keep their attention at any time and are ready to recover driving control at any time.

Level 4: Under a specific situation, vehicles can complete all driving behaviors, and humans do not need to invest in attention.

Level 5: Vehicles can complete all driving behaviors in any situation. Humans are just passengers and do not need to participate in driving at all.

I believe you must notice that very important limited condition: under a specific situation. We cannot list a detailed list for this specific situation. For example, for a level 4 self -driving car, although you can imagine many situations that may challenge it: bad weather, urban traffic congestion Navigation through the construction area, or on a narrow two -way road without any lane signs.

For a simple example, if you encounter a baby car sitting in front of you, a baby who is sitting in front of him, a mother with a baby car, and a dog pulled by his mother, how should artificial intelligence be avoided at this time?

Manufacturers and users of Tesla cars such as autonomous driving systems are still learning which situations are still learning "specific situations" that require human drivers to take over. There are also some experimental vehicles that can fully get autonomous driving in a fairly broad situation, but they still need a "safe driver" of the human "safe driver" who can immediately take over the vehicle at any time and receive a notice. So far, there have been several fatal accidents caused by autonomous vehicles, including some test vehicles. These accidents have occurred in scenes that should have been taken over by human drivers, but they have not been noticed in time.

In summary, do you still think artificial intelligence can replace humans?

Perhaps artificial intelligence in certain work will replace humans. In fact, some work has been replaced, and it has brought a lot of benefits to society. But no one knows what impact it will have on employment on the whole, because no one can predict the ability of artificial intelligence in the future.

So they will not completely replace humans to make humans unemployed on a large scale, at least now. Many human work may be much more difficult for computers or robots than we think.

There are many artificial intelligence products that suggest that the machine may really think, not just simulation thinking.

Just like "Iron", although it can perceive emotions and distinguish emotions, this does not mean that it has its own "consciousness" and the ability to truly think.

The real thinking is done by the brain, but the simulation thinking is completed with a computer.

Regarding thinking, there is a human characteristics that can never be completely imitated by computers.

Just like AI smart speakers, when you say it is not in the voice library it sets, then he will have a bug; just like the specific scene that you can't choose when AI is driving, it will not be able to choose from, leading to a fatal accident; Just like the AI ​​robot on the sales call, you can only identify those set words. If you do not follow the routine, they will be "confused", and then talk about it according to the set script.

All in all, now we are just a vision for the development and ideas of artificial intelligence. It is impossible to achieve the current technology. There is still a long way to go.

- END -