The speed of advancement and growth of technology is very fast. Artificial intelligence (AI) is a technology that enables machines to acquire knowledge through experience, adapt to conditions and be able to perform tasks like humans.
Sounds like a promising idea. But just like robots, there is still some concern about how far this type of technology can evolve. And of course, whether that would mean that machines can outrun their creators.
Communication between robots and humans, both in the real environment and on the internet, is already happening through artificial intelligence resources. How far this intelligence may reach, no one can say. But that she is revolutionizing the world of humans, that’s real.
Artificial intelligence
The things that AI has been able to do over time have increased and increasingly impressed people. He is proving to be quite skilled at certain tasks. Like, for example, inventing human faces that don’t exist or winning games. However, AI still struggles when it comes to something that comes naturally to humans: imagination.
As human beings know what a cat is, for example, it is easy for them to imagine a cat in a different color, or in a different pose, or in different environments. As for the AIs, it is more difficult, even though they can recognize a cat when they see it.
So, to try to unlock the power of AI imagination, the researchers came up with a new method to allow these systems to calculate what an object looks like even if they’ve never seen one exactly like it before.
“We were inspired by human visual generalization capabilities to try to simulate our imagination in machines. Humans can separate their learned knowledge by attributes. For example, shape, pose, position, color, then recombine to imagine a new object. Our paper attempts to simulate this process using neural networks,” said computer scientist Yunhao Ge of the University of Southern California (USC).
Imagination
The focus is extrapolation, that is, being able to use a large training database to get beyond what is seen and get to what is not seen. For Artificial Intelligence this is a difficult thing because of the way it is usually trained to detect specific patterns rather than broader attributes.
With this, the researchers discovered so-called controllable disentangled representation learning. It uses an approach similar to those used to create deepfakes.
This means that if an AI sees a red car and a blue bicycle, it will be able to “imagine” a red bicycle on its own even if it has never seen one before.
The researchers put this together into a framework called Supervised Group Learning. And one of the main innovations of the technique is the thought of samples in groups instead of individually. In addition to building semantic links between them along the way.
new approach
Thus, Artificial Intelligence is able to recognize similarities and differences in the samples it sees. And with that knowledge, she manages to produce something completely new.
“This new approach to disentanglement, for the first time, really triggers a new sense of imagination in AI systems, bringing them closer to human understanding of the world,” said computer scientist Laurent Itti.
In the future, this system could be used to protect against AI bias by taking more sensitive attributes out of the equation. With that, helping to create neural networks that are not racist or sexist, for example.