Recent research shows that a computer working with light instead of digital keys can significantly reduce energy needs in artificial intelligence applications. Scientists working on this prototype consider this approach as a new computing paradigm. The study revolves around a computer prototype developed by Microsoft and proposes that this system, called an analog optical computer (AOC), can perform certain AI tasks more efficiently and faster with less energy.
Compared to the processes in traditional computing where billions of digital switches are toggled, AOC uses different intensities of light and voltage, and with the help of small LEDs, sums and multiplies numbers within a feedback loop. The system computes a problem repeatedly, progressing on top of the previous results, and continues working until it reaches a stable state. It was initially demonstrated that AOC could perform a simple machine learning task, such as good classification, as effectively as a digital computer.
In the future, a broader AOC capable of processing more variables could outperform digital computers significantly in terms of energy efficiency. One of the experiments conducted in this context involved reconstructing a 320×320 pixel brain scan image using only 62.5% of the original data. The digital twin of the AOC successfully completed this task, indicating potential to shorten MRI procedures.
In another experiment, the team used AOC to find the most efficient transfer path for a series of financial problems and to minimize risks. The success achieved with stock exchange and financial transactions drew attention due to its high success rate compared to quantum computers.
