Researchers have introduced a new artificial intelligence model that utilizes a network of physical oscillators, promising to be significantly more energy efficient than conventional computing methods. The model, named “Un-0,” was developed by Unconventional AI, a startup established by a team of notable AI researchers.
The founding members of Unconventional AI include Michael Carbin, an associate professor at MIT; Sara Achour, an assistant professor at Stanford University; MeeLan Lee, a former Google engineer; and Naveen Rao, former head of AI at Databricks. Details about Un-0 were published in a technical blog post on the company’s website on June 25. The model is also accessible through GitHub.
Un-0 serves as the first proof of concept for the company’s technology, which integrates Achour’s research on nonlinear physical substrates with Carbin’s work in machine learning and physical dynamics. The model operates as a “physical dynamical system,” leveraging physical motion over time to execute computations.
Traditional computers utilize transistors that act as electrical switches, toggling between on and off states to process information. These transistors can represent binary data, allowing computers to perform complex calculations by layering millions or even billions of them together.
Current neural networks, such as those used in AI image generation tools like Midjourney or Dall-E, rely on extensive mathematical computations. These systems iteratively refine images by subtracting noise from a static base to approach a target image, often requiring multiple passes to achieve clarity.
In contrast, Unconventional AI’s approach is rooted in physical principles rather than abstract mathematics. The technology centers around oscillators, which are devices that generate continuous waveforms. The concept posits that interconnected oscillators, despite moving at different rates, can synchronize their movements through mutual influence.
This principle, when applied to thousands of linked oscillators, could potentially transform how computations are performed, making the process more energy efficient compared to current methods.
