The potential of AI is limitless, and scientists continue developing this field of knowledge, trying to create a full-fledged analog of the human brain capable of computing and thinking, creating, and even creating.
Two researchers at the University of Pennsylvania have succeeded in surpassing all past versions of AI by developing a neural network based on a low-level neural machine code. Such an AI can execute a program just like any computer, so the AI can solve mathematical problems, play video games, and run other artificial Intelligence.
Neural networks are known to mimic the brain's neural connections, so they consist of several layers of interconnected artificial neurons (nodes). Each node has its own weight and threshold value: if data exceeds this value, it is transferred to the next layer, and so on. Therefore, each AI-driven neural network has the task for which it was designed. The neural network finds patterns in thousands of example tasks and learns to compute them to improve. It is called machine learning. However, Jason Kim and Dani Bassett at the University of Pennsylvania have taken an entirely new approach to creating a neural network. A neural network gets trained to execute code in the same way as a standard computer.
Artificial Intelligence trained to mimic the logic circuits of a conventional computer in its neural network could theoretically speed up many calculations and perform them independently. However, the lack of a low-level programming language suitable for neural networks currently prevents the full potential of such an AI from being unlocked. For this reason, Jason Kim and Dani Bassett are presently developing a new programming language that could bridge the gap between neural computers and conventional ones.
Share this with your friends!