Everyday, the envelope is being pushed a little bit further on what artificial intelligence (AI) will eventually be like. (Reuters)Everyday, the envelope is being pushed a little bit further on what artificial intelligence (AI) will eventually be like. At the end, will it just be algorithms—of course, of increasing complexity—leading to an outcome, or will there be a fully functional, artificial brain? No one’s going to give you a definitive answer now. But what you will be told is that there is work going on at fantastic speed and at multiple levels of complexity. One segment of work in AI is on artificial neural networks—AI that operate (in a very rudimentary manner) how a human brain would. Researchers at Massachusetts Institute of Technology (MIT) have designed a microchip that uses beams of light to function in a neuron-like manner—in computers, the transistors present in chips are used to regulate passage of electricity in a manner in which codified information is processed. MIT’s new photonic chip—key to building what are called optical neural networks—is based on the fact that light is a more efficient in neural networks than electricity because light waves can travel as well interact at the same time. Thus, they can perform a quantum of functions in a given time period that is not possible for electricity to carry out. Of course, this had been known for quite some time, but all attempts to create optical neural networks resulted in massive setups with hundreds of sensitive mirrors and lenses propped for work.
You May Also Like To Watch:
Against such a backdrop, MIT’s photonic chip represents, forgive the pun, a quantum leap. The new chip, made of silicon, simulates a network of 16 neurons. Data is entered into the chip in the form of a laser beam split into four smaller beams—the brightness of each beam as it enters represents an unique number (distinct piece of input information) and the brightness at exit represents a different but unique number (or the distinct piece of information processed). In passage through the chip, the paths of the light intersect and interact in different ways to either amplify or weaken their individual intensities—a near-perfect simulation of the way signals transmitted via neurons can be intensified or weakened.
As per a report in Science, the researchers tested the network for recognising vowel sounds. Using recordings of 90 people making four vowel sounds, the neural network was able to tell the vowel sounds 77% of the time. Though old-school, transistor-based processing had a higher 92% success rate, the optical neural network did the task much faster and much more efficiently. If the kinks are worked out to achieve a higher success rate, the photonic chip can unlock phenomenal potential for AI—a photonic neural network in an autonomous car, for instance, can process information in a minute fraction of the time that the AI in such cars take today. On the road, where a split-second spent in taking the correct decision can make the difference between a crash and safe passage, wouldn’t this be a blessing?