All about AI
Scientists are working on AI technology that has brain-inspired hardware, architecture or algorithms. Such neuromorphic AI could be nimbler, more efficient and more capable than traditional AI, freelance writer Kathryn Hulick reported in “Making AI think more like your brain.”
Hulick reported that mainstream computers, which currently run most AI, separate memory and processing. Some burgeoning neuromorphic technology, such as spiking neural networks, combine the two.
This concept reminded reader Gary Pokorny of an early experience with computers. “The first computer I used … was an Apple IIe, in which I would insert one floppy disk to load word processing instructions, then take it out and insert a blank floppy to save my work, and back and forth while writing,” Pokorny wrote. The personal analogy helped Pokorny “understand why mainstream AI requires huge resources for both memory and processing. I have a harder time grasping, but am fascinated by, the idea of spiking [neural networks combining both], hence more efficiently, and more like our brains.”
Neuromorphic experts’ work to streamline computing systems struck a nerve with reader Linda Ferrazzara. “All the while, I couldn’t help thinking of how human brains develop, with an initial surfeit of neurons and connections that get gradually pared down as the brain is pruned into a more efficient configuration, from prebirth to adulthood.”
Ferrazzara wondered if quantum computers could be adapted to neuromorphic computing systems.
Quantum computers perform powerful computations by leveraging quantum principles, such as superposition, the idea that subatomic particles can exist simultaneously in multiple states, and entanglement, a type of ethereal link between particles.
Quantum and neuromorphic computing are very different technologies, says Daniela Rus, a computer scientist at MIT. “I don’t think you can directly adapt quantum computers into neuromorphic computers, but we might be able to use neuromorphic processes to control quantum computers,” Rus says. What’s more, “ideas from quantum mechanics may be useful to design new chips for neuromorphic computers.”
Quantum and neuromorphic computers could be used to perform different but complementary computations, says computer scientist Prasanna Date of Oak Ridge National Laboratory in Tennessee. “For example, quantum computers could be used to train spiking neural network models, which get deployed on a neuromorphic computer for energy-efficient, real-time machine learning computations.”
Corrections
In the February feature “Holding back a glacier,” the opening image was incorrectly identified as Thwaites Glacier. The image was actually of Pine Island Glacier.
In the March issue’s “Have 5 years of COVID-19 readied us for what’s next?” the last sentence of the second paragraph had a missing word. The sentence should have read: Nearly 17,000 people in the United States died of COVID-19 the last week of that year.
