This form of artificial intelligence uses light refraction in lieu of processing numbers with algorithms. Back several decades ago, the cutting edge was a light computer. Mind that I read about this idea at a time before computers were ubiquitous. The prospects for a computer which is powered by light rather than electricity would theoretically run faster and cooler than one that used electrons.
The light computer would have a light pump instead of a power source, it would use optical fibers instead of wires and the processing components would also work with light. As far as I have read, they are still working on devising components that will operate on photons rather than electrons.
As a thought, I am researching the use of available components but the wiring would be optic fibers. This would lower the use of electricity but it would still be required. That isn’t an issue as a power source is still needed to run the light pump. Replacing the wires or circuit board traces with optics would lower the time for the signal to be routed from chip to chip as they would move a close to the speed of light rather than the speed of electrons running through a conductor. The overall heat will be reduced as the optic fiber has less resistance than the electrical conductors. To convert the light data to electrons, the fibers need to have an optocoupler.
Researchers at UCLA have demonstrated this AI computing by printing 3D layers of transparent material. The layers have a diffraction pattern imprinted. These patterns work to diffract the light being inputted and manipulates the light to perform mathematical operations on the light.
There are different ways to encoded information on a light beam. One way is to toggle the light on and off, by this method the ‘on’ state would be seen as a ‘one’ and the ‘off’ state of ‘zero’. This is similar to how computers pass their data through electronic conductors, only it uses voltage. Another way is by having light with data encoded interfere with another light with encoded data. When the light patterns both are in the ‘on’ state, the output will be a brighter light. When one is ‘off’ and the other is ‘on’ the output will be a dimmer light and when they are both ‘off’ the output will be no light.
This is the basic method of what they have done with these transparent layers. By having the diffusion pattern on one layer in line with another diffusion pattern on the next layer, the output from one layer will be transferred to the next and the interaction of the data will manipulated on by each layer.
This could be used like a traditional computer uses a math co-processor and even a video rendering board. A co-processor runs along side of the CPU, when the CPU has a math calculation it will output the data to the co-processor and the CPU can process other data at the same time the co-processor is doing the math computation. When the co-processor is done, it sends the answer back to the CPU.