Musk's latest interview: About autopilot, AI, and Tesla's humanoid robots

Recently, Elon Musk, the world's richest man, participated in an exclusive interview hosted by TED head Chris Anderson.

In the interview, Musk talked about some of his views on AI, autopilot, the acquisition of Twitter, and Tesla's previous development history, and the following is a compilation of some of the interviews, mainly related to AI autopilot and Tesla humanoid robots.

1

About AI and autonomous driving

Anderson: "The interior of Tesla's gigafactory is really shocking, and I think what I see is not a factory that produces electric cars in the traditional sense, but a complete system for the production of electric vehicles, and the capacity increase that this system has achieved in the process of continuous optimization is bringing considerable revenue growth to Tesla."

It's remarkable, but one of the issues that I'm very interested in here is about the development of autonomous driving. You told me five years ago that 'Tesla will be fully autonomous this year.'

But in fact, no one has achieved true automatic driving until now. What do you think about this? ”

Musk: "I think the reason is that the development of autonomous driving has really appeared many times in the past 'false dawn', including me, many people have thought many times 'Oh, this real sense of automatic driving is really coming this time', but in fact, it did not come later, and taking 'dawn' instead of replacing it is more problems that need to be solved, the problems that need to be overcome, the ceiling that needs to be pushed in technology."

If we can draw a curve for the development of autonomous driving technology, it should be like a curve made up of many log functions, which can first climb a while, then tend to flatten, then fall for a while, and then may then tend to be flat or turn upward.

The core problem here should be to figure out a 'reliable AI system that can really be used', which is compatible with our road system.

Our road system is designed to be compatible with our biological neural network, that is, the brain, and also with our biological vision, so if we want to develop an autonomous driving AI that is compatible with the road system, the core problem is how do we consider various factors and develop an AI system that is real and reliable and compatible with the external conditions just mentioned.

For example, the system will need cameras and silicon-based neural networks on hardware to be promisingly compatible with existing road systems."

Anderson: "So what do you think of the existing framework, which is enough to get us up the curve of autonomous driving technology?"

Musk: "Maybe my answer to that question will be less popular, but I think we'll be able to fix it this year."

For example, when you're measuring what is true autonomous driving, one piece of information that can be used as an indicator is whether the probability of the self-driving program having an accident is lower than the average for human drivers, and I think we can achieve this this year.

While it's also possible that we'll still be discussing this next year or the year after that, I'm confident we'll be able to achieve it this year. ”

2

About Tesla's humanoid robot

Anderson: "Are you deliberately making these radical remarks, maybe if no one had done that, we wouldn't have been able to do it sooner?"

Last year, for example, Tesla's AI made a lot of progress in understanding its surroundings, and then you revealed Tesla's humanoid robot program.

Do you have the confidence to think ,'we can do this' because Tesla has made a major breakthrough in its R&D of self-driving systems? ”

Musk: "Yes, as mentioned earlier, it took me some time to realize that the key to solving autonomous driving is to achieve a highly compatible and reliable AI, and to put this concept on an autonomous car, you can think of the car as a robot on four wheels, but we can actually generalize this concept further to 'general-purpose humanoid robot on two legs', and now the difference between achieving this is a strong enough to explore the surrounding environment autonomously. Intelligent systems that accomplish meaningful tasks based on the results of exploration.

In addition, I think that in addition to the software problems, we will also need a strong enough manufacturing production system in the hardware, and whether it is software or hardware, the two points just mentioned are actually tesla's good at.

So, if Tesla wants to build a humanoid robot, all it needs to do is come up with a design solution that integrates the power unit and the induction device. Of course, the power and sensing devices also need to be designed by ourselves.

At present, many people do not realize that building humanoid robots is actually much more meaningful than what kind of car to achieve. ”

Anderson: "You just talked about the application of AI systems, and you think that the initial application is on the manufacturing system, but it will eventually go into everyone's home.

So, if there is such a robot in the home, it can really and effectively understand the 3D structure of your home, know what is where, or where what should be, and can identify what is what, what can it do? Like cleaning up our room? ”

Musk: "Yes, clean up the room, cook, take care of the elderly and so on" and so on.

Anderson: "It recognizes all the members of the family, brings tea to the elderly, takes pictures, plays a game of catching balls with your children." ”

Musk: "Yes, another thing we need to be aware of is that we can't let robots be the infrastructure to build a 'dystopian society.'"

The robot must have a local chip that the network cannot rewrite, for example, if a person issues a 'stop' command to it, it should stop, and this function cannot be rewritten by online operations.

I think the robots of the future need to have this safety mechanism in design. And I also think we need an AI regulator, and while individuals don't like to be regulated, I think it's important for public safety. ”

Anderson: "So, in your opinion, you think, for example, in 2050, everyone can have such a housekeeping robot at home." ”

Musk: "Yeah, I think it's very likely, maybe then we'll all have a good robot brother or sister." ”

Anderson: "So what can our good brothers and sisters do besides housekeeping, do you think a robot with some special function is possible?" ”

Musk: "Yeah, I think it's going to be inevitable, for example, I promised on the Internet that I'd build a cute Catwoman robot, but then again, I think the robot of the future will be able to do everything people want it to do." ”

Anderson: "When do you think the first robot of this kind will be built and hit the market?" ”

Musk: "The first humanoid robot we're currently designing will be aimed at jobs that are more dangerous for people and more repetitive for people.

We may come up with a prototype within this year and then launch a model that can enter the market next year, but in general, I think in about two years, we will see the first robot of this kind, and then we will witness the product iteration process of humanoid robots, increased functionality, increased production capacity, and lower prices. ”

Anderson: "How much do you think this robot is going to sell?" ”

Musk: "I think the price of this robot should not be too high, for example, it may be cheaper than a car." ”

Anderson: "But if we were to do the math in this place, if we could save a human worker $30,000 to $40,000 a year in salary expenses by buying a robot, assuming that the robot would sell for about $25,000, and assuming that everything was going well, the robot could work longer a day than a human worker and not need a vacation, it would be a good deal."

So, we have reason to believe that there will be a lot of people in specific jobs in the future who will be replaced by this robot, should we be worried about this? ”

Musk: "I personally don't have any concerns about this.

I think we're going to see, in fact, see a massive shortage of manpower in the market, so even in the future with this kind of robot, this manpower shortage should still exist, and the kind of large-scale unemployment caused by robots that everyone imagines may not actually appear.

The future is likely to evolve into a world with a better balance of supply and demand, where people will be able to easily and cheaply access any goods and services they need.