Artificial intelligence can’t deal with chaos, but physics training can help

(ORDO NEWS) — Although AI systems continue to take tremendous steps forward, they are still not particularly good at dealing with chaos or unpredictability. Researchers now believe they have found a way to fix this by teaching AI physics.

The Hamilton function gives the AI ​​information about the entire dynamic system: all the energy contained in it, both kinetic and potential.

Neural networks designed to mimic the human brain as a complex, carefully weighted type of AI then have a “more complete picture” of what is happening, and this may open up opportunities for AI to solve more and more complex problems.

“The Hamiltonian is really a special sauce that gives neural networks the opportunity to study order and chaos,” says physicist John Lindner of the University of North Carolina.

“Thanks to the Hamiltonian, a neural network understands the underlying dynamics that a regular network cannot understand. This is the first step towards physically savvy neural networks that can help us solve complex problems.”

Researchers compare the introduction of the Hamilton function with a swinging pendulum – it gives the AI ​​information about how fast the pendulum swings and its path, rather than just showing a snapshot of the pendulum at some point in time.

If neural networks understand the Hamiltonian flow, that is, where the pendulum is, in this analogy, where it can move and what energy it has, then they can better control the introduction of chaos into order, according to a new study.

Not only that, but they can also be built to be more efficient: it is better to predict dynamic, unpredictable results without a huge number of additional neural nodes. This helps AI quickly gain a better understanding of how the world actually works.

To test their recently improved AI neural network, researchers compared it with a generally accepted standard called the Henon-Hales model, originally created to model the movement of a star around the Sun.

The Hamiltonian neural network successfully passed the test, correctly predicting the dynamics of the system in states of order and chaos.

Improved AI can be used in all areas, from disease diagnosis to piloting autonomous drones.

We have already seen how AI simulates space, diagnoses health problems, modernizes films and develops new drugs, and the technology, relatively speaking, is just beginning – there is still a lot to come.


Contact us: [email protected]

Our Standards, Terms of Use: Standard Terms And Conditions.