Deep learning alternative can help artificial intelligence agents operate the real world

New machine The learning approach that is inspired by the way the human brain appears to be design and learning about the world is able to master a number of simple video games with great efficiency.

The new system, called Axiom, offers a substitute for dominant artificial neuroma in modern artificial intelligence. Axiom, developed by a software company called VERSE AI, is equipped with prior knowledge about the way the organisms interact physically with each other in the game world. Then it uses an algorithm to model how to expect the game to act in response to the input, which is updated based on what it notes – a process called active inference.

The approach requires an inspiration from the principle of free energy, a theory that seeks to explain intelligence using principles derived from mathematics, physics and information theory as well as biology. The principle of free energy has been developed by Karl Friston, a famous neuroscientist who holds the position of chief scientist in the verses of the “Cognitive Computing” company.

Friston told me, via a video of his home in London, that the approach may be particularly important for building artificial intelligence agents. “They should support the kind of perception that we see in real brains,” he said. “This is required, not only of the ability to learn things but in reality to learn how to behave in the world.”

The traditional approach to learning games includes training nervous networks through what is known as deep reinforcement learning, which includes an experience and replacement of its parameters in response to positive or negative comments. This approach can produce algorithms playing the super game of humanity but requires a lot of experimentation to work. Axiom mastered many simplified versions of the famous video games called Drive, Let and Hunt and jumping using much lower examples and lower mathematical strength.

“The general goals of the approach and some of its main features follow what I see are the most important problems that must be focused on to reach AGI,” says François Cholete, a AI researcher who has developed ARC 3, a designed an indicator for the testing capabilities of modern AI algorithms. Chollet also explores new methods of automated learning, and uses its standard to test models capabilities to know how to solve unfamiliar problems instead of just imitating previous examples.

“The work amazes me as very original, which is great,” he says. “We need more people who try new ideas away from the exciting path of large linguistic models and thinking language models.”

Modern artificial intelligence depends on the artificial nerve networks, which is almost inspired by the brain wires, but it works in a basis in a basis mainly. Over the past decade, deep learning, an approach that uses nerve networks, enabled computers to do all kinds of impressive things, including copying speech, identifying faces, and creating pictures. Recently, of course, deep learning has led to large language models that increase the amount and ability to increase Chatbots.

In theory, Pedesium is a more efficient approach to building artificial intelligence than scratch. Gabi Rene, CEO of Verses, says it may be particularly effective to create agents who need to learn efficiently from the experience. Rene says that one of the financial companies has started experimenting with the company’s technology as a way to mix the market. “It is a new structure for artificial intelligence agents who can learn in an actual time, more accurate, more efficient and much smaller,” says Rene. “It is literally designed like the digital brain.”

One of the paradoxes, given that Axiom provides an alternative to modern artificial intelligence and deep learning, the principle of free energy was originally affected by the work of the British Canadian computer world Jeffrey HentonWhich got both Torring Award And the Nobel Prize for his pioneering work on deep learning. Hinton has been a cohistor of Friston at College University for years.

To learn more about Friston and the principle of free energy, I highly recommend This WIRED 2018 feature article. The effect of Friston’s work on An exciting new awareness theoryDescribed in a wireless book reviewed in 2021.