Researchers at Stanford just announced that they've created the first low-power, high performance artificial neural synapse. The new artificial synapse uses a tenth of the energy of traditional computer memory and works on the same low voltage as the human brain. Plus, rather than working in binary states like today's neural networks, it can recall 500 states of charge. According to Stanford News, "every part of the device is made of inexpensive organic materials" like carbon, salt, and water. Physically, it's two thin plastic films and three electrical terminals in a wash of conductive salt water. While this research is still young, it could be a promising link in the process of making safer artificial intelligence.
Artificial synapses and artificial intelligence
As you know, the jury is out on whether or not we can develop truly safe AI.
Sam Harris is certain superhuman AI may not be survivable for the human race. Solution driven experts like Elon Musk have also often expressed concerns, even as he works on his own solution--called neural lace. He shared recently, "If you assume any rate of advancement in AI, we will be left behind by a lot . . . Even in the benign situation, we would be so far below them in intelligence that it would be like a pet, or a house cat."
Making progress toward a better brain/computer interface
Right now, you could say we do "interface" with AI and computers, but we do it through screens and keyboards, or with embedded sensors. Not great. It's okay for action/reaction, but too slow for real time communicating with artificial intelligence. According to Musk and others, one of the best solutions has been a theoretical one--a third layer of digital interface that allows you to completely integrate with digital systems--you know, cyborg style.
This more direct connection is challenging, though. For one thing, chip-based computers use a lot of power relative to our brains--their memory systems are like lightning storms compared to the gentle rain of around 70 millivolts we use inside. "The kinds of tasks we expect our computing devices to do requires computing that mimics the brain, because using traditional computers to perform these tasks is becoming really power hungry," said Alec Talin, an author of the paper just published this week in Nature Materials.
An organic neural network for lower power AI
Our brains recall information patterns through neural charges. So does this new artificial synapse--and that's why it's so exciting. Yoeri van de Burgt, Ph.D, a lead author on this week's paper, puts it this way. "Instead of simulating a neural network, our work is trying to make a neural network."
This promising new synaptic technology is a lot closer to how we understand how our own brains pattern experiences, create memory traces, and train our behavior. A version of it could help bridge high power, high bandwidth computers and our low power, high performance brains. Stay tuned for more interesting news--Elon Musk is expected to have an announcement of his own on "neural lace" technology soon.