Is it really not scary when a robot has self-awareness and can adapt itself to the environment?

When robots, like humans, can learn, shape, change and evolve with each other, the tiger sniffs the dark side of artificial intelligence with horror.

Last week, Baidu held its first overseas Talk in Silicon Valley, where Silicon Valley's leading experts in depth computing and artificial intelligence shared the world's most advanced research and philosophy, including Woody Cornell, the head of the Creative Machine Laboratory at Cornell University (University). Lipson (Hod Lipson) shares the latest results of his lab-the next generation of robots that are self conscious and adaptive to the environment.

Cultivating robots with large data and learning to achieve self consciousness

Lipson is talking about a robotic technique that has been started, called self-awareness (self awareness). "It's not a technology you'll see tomorrow, I bet it's the technology you'll see in your lifetime." ”

Lipson defines self-awareness as the ability to be a self imagine. He made an analogy, can you imagine yourself going to the beach tomorrow? Can you imagine the smell of the ocean? Can you feel the sand now? Can you imagine what you look like in my eyes?

"Why can you imagine a situation where you have not really experienced it, it is because of self-awareness, you are conscious." In a way, you are not making decisions based on the experience you have actually experienced. This is what I describe as the definition of self-awareness, which is what we humans have-and we want the machine to eventually have the ability. ”

Darwin said that it was not the strongest species or the brightest species that could survive, but that the most responsive and varied species in the environment eventually survived. So adaptation is very important. More than 10 years ago, we began to decide not just to design, to make robots, but to nurture (breed) robots. ”

Lipson said the robot actually contains two levels, one is using a supercomputer simulation robot (simulator), the other is through the 3D printer manufacturing robot.

So what is called a nurturing robot, Lipson explains: We will ' reward ' the robots ' performance, copy them and make small changes in some places so that they can crawl on the floor as much as possible, so that they will be copied more and more. In this way, we allow evolution to continue, eventually to get the conscious robot. ”

"We use the printer to make these robots and let them go from the virtual environment to the real world." These robots are constantly nurtured by deep learning techniques, not designed directly by people. It's exciting. It's just plastic, but we'll have it printed into titanium in the future.

At this stage, I began to design more complex robots. One of the incredible robots, it has the wind engine, it can jump up, it can flip on the ground, and I'm proving to my colleagues that this technology is going to work. So we made this robot. We put it in a big cage, and they keep tumbling backwards and forwards in this cage, and you can see that these movements are random and self fulfilling, meaning that it is learning how to walk.

"We collect data from robots, collect sensor data through sensors, capture action data, and data that robots feel." We collect all this data and use it to nurture simulators. So, rather than just giving the simulator data, we Now "feed" data to both the robot and the simulator. As a result, the robot's performance is getting better now, and our robot has the ability to imitate and adapt to itself.

"The robot has begun to learn to imitate itself in its environment and to make intuitive judgments based on it," he said. When things are not intuitively judged, they add to their own imagination. Now let me show you this robot. “

"These robots will learn to walk. They will first create a lot of assumptions, through its engine to randomly move and perceive, feel the sway around. They collect all of this data and generate their assumptions/judgments.

To test their adaptability, we did something very cruel, and we broke one of its legs. You can see its self adaptation through its own imagination, which loses a leg in its own imagination. You can see this pink leg shorter than the others. With this new ego, it finds new ways to move forward.

"You know, if I'm going to design a robot that lacks a leg, I can make a robot that works better," he said. But what we're doing here is not saying that the sensor tells the robot that it's missing a leg, but it lets it feel and change dynamically. The robot's own imagination has changed, so the way the robot behaves is changed. In other words, it's adaptive. ”

Robots can think for themselves

"What I've shown you so far is that the robot adapts to its own mechanical structure, but can you imagine that the robot can think for itself?"

Robots are not shaping themselves, they shape each other. We already have the idea that human beings are shaped by each other, you shape me, and I shape you. You can think, I can think what you think, you can think how I think of you, cycle. And we can do that for robots. ”

This stunning ability to reinvent each other means that the robot's "Awakening", viewed and exploited from the positive, can lead to a more convenient and efficient life for people.

Lipson that bridges can discover their own weaknesses, faster and more accurately than engineers, because they are self-conscious.

Imagine that self-driving cars will mold themselves, and cars themselves know what they can and can't do. In the ecology of self-driving cars, each self-driving car will also be able to shape other vehicles, and better in the driving environment to pre-judge what other vehicles to do and how to do it.

Why these sci-fi-like futuristic technologies can happen today, Lipson: "These ideas have been in fact long ago, but because of two technologies, we have been able to achieve today's robotics." One is the new manufacturing process, 3D printing, and the other is a breakthrough in depth learning. ”

' Today we can print batteries, print muscles, and even wire and electronic devices can print, ' said Lipson. Because of the future possibilities of 3D printing technology, tomorrow's robots will be completely different from the robots in sci-fi movies, whose body structure will no longer be a cold metal, mechanical, but a more organic organization that can wrap equipment and components inside.

And in depth learning, in the past 25 years the pace of the technology has not been as exciting as it seems today, and the recent strides have been made in leaps and turns, on the one hand, the constant support of large data, and the massive "service" of low-cost supercomputers capable of processing large data, Unlike the previous generation, which relies on cloud computing for a deep learning model, this generation of technology, stability and high performance is the key to driving deep learning to a breakthrough.

Isn't the Lipson robot enough to be worrying?

Lipson is excited about the progress of his lab, "If you believe that robots will kill humans, you will miss out on great technological advances." "But in response to the possibility of the robot killing, Lipson seems to be a little evasive. "I don't worry about robots hurting us, and I'm worried that some people will use robots to hurt people," he said. ”

It's like we all know the great possibilities of biological cloning for human progress, but we can all imagine the disastrous consequences it can bring.

Of course, optimistic about this is not only Lipson, Baidu chief scientist Wunda is also more willing to see the positive side of AI progress, "in the past six months, some people talk about the evil robot will kill, or fear that the future of the computer will become more intelligent than us, they will occupy the world." We are on the frontline, shaping this technology. I believe we know what to do today and it seems to be all very well at the moment, without worrying about the evil killer robots. I personally do not believe that these speculations have realistic possibilities. ”

(Responsible editor: Mengyishan)