MIT's new robot learned by watching.

MIT's new robot learned by watching. 

The new breakthrough in AI and robotics is a tool that can learn like humans. Those robots can look at the monitor and then learn and repeat those things. That means the robot is easier to teach, and no programming is needed. The ability to learn like humans makes robots more versatile. But the thing is that those robots are only one part of the AI-based tools. 

The robot can act as a medium that transmits data that it collects to the AI-central computers. And basically, the AI can learn things by using any camera system in the world. That gives the ultimate possibilities for the AI. If we think that the AI learns to drive, it's possible to use the robot as the medium. 

The robot sits in the classroom and drives a car following orders that the instructor gives. Then the system can transfer those models to the robot vehicles. Or, maybe the driver robot includes the service if somebody buys a robot car. The man-shaped robot can carry things like luggage to the car. And the same robot can wash the car and clean the house if its master wants. 

The problem with a human-shaped robot is that it can do the same things as a human. The human-shaped robots can learn things simply by looking at things on TV. That makes it possible to use the regular movies to train robots in civil and military actions. The thing is that the robot's shape must not look like a man, so that it learns things. The robot aircraft, or drone, or robot car can use the same methodology as a human-shaped robot to learn new tricks. 

Digital twins can make it easier to teach AIs. The digital twin can be the character that a person drives in computer game-type simulations. The digital twin can be the tool that helps to see how the system controls the real robot. The digital twin can also make it possible to advance reactions. That physical robots can face in their missions. The digital twin makes it possible for that system to advance AI. And test the AI and program's ability to respond to challenges. 

The AI can use things like aircraft or satellites to record maneuvers that potentially hostile actors make. Then the AI can drive those maneuvers in the digital simulation. Or simulate a battlefield and create counter-maneuvers to that thing. In the same way, the AI can collect data from the flight profiles of the aircraft, and then the system starts to create counter modes for that maneuver. This is the thing that makes the AI an effective tool in civil and military operations. 

https://www.rudebaguette.com/en/2025/08/that-robot-learned-just-by-being-watched-mits-terrifying-ai-builds-3d-control-map-from-video-alone-no-sensors-or-programming-needed/

https://scitechdaily.com/ai-twins-could-help-save-the-planet-but-only-if-we-fix-them-first/

How to teach AI?

 How to teach AI? 

If we want to make AI that operates smoothly in everyday life, we must make one thing clear to ourselves. The thing is that the AI must handle many variables so that it can operate independently. And it's hard to program those things into the AI’s source code. Even the best programmers forget something. So the answer can be the robot that sits in the classroom and learns like a human. This helps to create all the necessary variables for the AI. The robot acts as a medium between humans and the AI. 

If we want to make self-developing AI, we must use two AIs. The first AI makes a requirement of what it needs for the other AI. Then another AI creates source code. Then, that system must connect the AI to the source code of the first AI. And after that, the system reboots the servers. Before that, the system must simulate how it controls the virtual robots. 

The AI’s self-development happens as in natural organisms. The AI must detect what it needs, and then the biggest problem is how the AI can create code that involves the data it needs. Source code is the DNA of the AI, and there must be something that tells the AI what a certain code sequence makes. The answer to that problem would be the digital twin, the other AI that it can use to simulate what certain source code makes. The best way to get the code is simply to ask for it from some other AI. However, that means the other AI must have access to that source code. 

That requires that the AI tell its problem to other computers. And those other computers give it the code that it needs. But there are legal and other kinds of problems for that kind of self-developing model. So, how do we improve our skills? If we don’t know something, we search for data on the internet or visit the library and read about it in some books. Or, maybe we can take some courses on the topics. 

The thing in AI self-development is how the AI describes the problem to itself? When we talk about teaching AI, we can imagine a situation where the robot is sitting in the classroom and listening to how the teacher explains something. This is one version of the AI training. The AI can also search how some coders make solutions and then copy that code into its digital twin. If we want to teach AI to drive cars, we can use the same method as we use when we learn to drive cars. The driver must drive a car that records the actions for the AI. Then the AI asks necessary questions, and the human accepts that process if it's safe. 

We must remember that there are many variables that the AI should notice in the traffic. It would be very hard to describe all of those variables and traffic signs for the AI by writing them into its source code. The answer would be that the programmers make the base model for the AI. And then AI fills that model. And one of the most interesting ways to learn how to drive is to put the human-shaped robot into a driver’s school. The robot must know that it must avoid humans, know the traffic light, and other regulations. Then the robot that the AI uses for its developing platform or transmitter that transmits data to the AI will finish that training.