In The Matrix Morpheus Neo states that her digital appearance is based on her "residual self-image". That is, the characters look like they envision their own mental models of themselves.
In the real world, scientists have been trying to teach robots this trick. This is because a real robot with a precise self-image, unlike the martial machines, can be of use to the Matrix of Humanity. This allows for faster programming and more accurate self-planning, and helps a device self-diagnose if something goes wrong. It could even help a robot adapt to the damage it suffers.
And on Wednesday, two scientists from Columbia University said they had given a robot arm that self-confidence and thus a new learning potential. Her research is published in the journal Science Robotics .
The Self-Image of a Robotic Arm
The newspaper is surprisingly readable and its summary reads in its entirety: "A robot has modeled itself without prior knowledge of physics or its shape and used the self-model to perform tasks and detect self-harm. "(Sounds like a Netflix movie description … I'd take a look!)
The researchers bought a standard model robotic arm – intimidatingly called WidowX – and taught them to visualize themselves. They went through 1,000 random trajectories, basically watching what had happened: how certain movements felt, what was possible, what was inefficient, everything. The authors even compared it to a person who had first learned the skills of his own limbs and wrote, "This step is not unlike a babbling baby watching his hands."
Armed with all this data, the robot took advantage deep learning to create his own self-image, ie an exact model of himself. It took a while for the first models to be far away, but after about 34 hours of training, the self-model was 4 centimeters accurate. That was good enough to become an expert in picking up and moving small balls – a typical substitute for robot skill. The self-understanding of the robot was good enough that he could accomplish a completely different task without further training: writing a word with a marker by hand. (The robotic arm, by the way, says "hello.")
Bigger Robot Things
To simulate a sudden injury or minor damage, the researchers replaced the arm the robot used with a slightly longer and deformed one. The machine quickly improved its self-image to accommodate the new situation, and was soon busy again doing the same tasks with approximately the same accuracy.
Overall, the authors make a convincing case that robots make robots create a precise self -images are possibly the best way to create accurate, self-diagnostic and efficient machines. "Self-imaging will be the key to enabling robots to move from the limitations of so-called narrow AI to more generalized skills," they write. Then they go a little further: "We suspect that this separation of self and task may also have been the evolutionary origin of self-consciousness in humans."
It is definitely cool and all as long as we stop making our own machines too much like the one in The Matrix .