Here’s what you’ll learn when you read this story:
For all the talk of things that AI can do better than humans, there remains one especially important skill that even the most advanced machine learning systems can’t match: the raw computing efficiency of human gray matter. Our brains are essentially immensely powerful computers that somehow run on roughly 20 watts of power. AI data centers, on the other hand, churn through millions of watts (and lots and lots of water), much to the detriment of the environment and your electric bill.
Of course, scientists have long known about this massive energy discrepancy between human biology and artificial computing, and they’ve sought ways to merge the two together for decades. Now, in an era where everything seems to be AI, creating ways for these systems to be more efficient is becoming critical, especially if we want to conserve energy and other resources. To that end, a new study published in the journal Nature Nanotechnology investigates a novel way to create artificial neurons that can actually “talk” with real, living brain cells. The hope is to provide a flexible platform (similar to neurons) that has energy advantages over powerful-but-rigid silicon devices.
“The way you make AI smarter is by training it on more and more data. This data-intensive training leads to a massive power-consumption problem,” Mark Hersam, the senior author of the study from Northwestern University, said in a press statement. “Because the brain is five orders of magnitude more energy efficient than a digital computer, it makes sense to look to the brain for inspiration for next-generation computing.”
Creating an artificial neuron is a great place to start. After all, the average human brain contains around 86 billion of them. Specifically, our brains have more neurons in the cerebral cortex—which is responsible for consciousness, thought, memory, emotion, and language—than most animals. While silicon devices (complete with billions of transistors) are powerful, they’re not flexible like human neurons, which are capable of reshaping or even dropping certain connections over time. This is why humans can learn and adapt using very little energy.
To create these artificial neurons, Hersam and his team relied on aerosol jet printing—a technique commonly employed to make electronic components like resistors and capacitors—and used electronic inks as an electrical conductor and semiconductor. In this case, the inks were formulated from flakes of molybdenum disulfide (which acts a semiconductor for the neuron) and graphene. In the past, polymers have been a bit of problem, as they tend to interfere with electrical current. To combat this, Hersam and his team developed an ingenious method for making the substrate more brain-like.
“Instead of fully removing the polymer, we partially decompose it,” Hersam explained. “Then, when we pass current through the device, we drive further decomposition of the polymer. This decomposition occurs in a spatially inhomogeneous manner, leading to formation of a conductive filament, such that all the current is constricted into a narrow region in space.”
This narrow pathway is what allows this electronic inked-up polymer to behave more like a neuron, as it can produce single spikes, continuous firing, or burst patterns rather than a simple, one-off pulse. But the real test came when Hersam’s team interfaced these artificial neurons with a sliver of a mouse’s cerebellum. The team confirmed that artificial voltages matched biological ones, and even triggered activity in the mouse brain tissue.
“Other labs have tried to make artificial neurons with organic materials, and they spiked too slowly. Or they used metal oxides, which are too fast,” Hersam said. “We’ve demonstrated signals that are not only the right timescale but also the right spike shape to interact directly with living neurons.”
More work needs to be done to perfect these artificial neurons—to say nothing of the additional steps needed to create artificial synapses that can connect everything together. But this study shows that creating the building blocks of a future, brain-like AI computer is possible. Now, we need the cerebral mortar and the detailed cognitive blueprints to make this decades-long dream a real, thinking reality.
Darren lives in Portland, has a cat, and writes/edits about sci-fi and how our world works. You can find his previous stuff at Gizmodo and Paste if you look hard enough.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。