Advancing AI with Brain-Inspired Technology
Scientists are working on AI technology that incorporates brain-inspired hardware, architecture, or algorithms. This neuromorphic AI has the potential to be more agile, efficient, and capable than traditional AI systems. According to freelance writer Kathryn Hulick, mainstream computers currently used for most AI applications separate memory and processing. However, emerging neuromorphic technologies, such as spiking neural networks, are combining these functions.
This concept resonated with reader Gary Pokorny, who recalled his early experience with computers. “The first computer I used was an Apple IIe, where I would switch between floppy disks to load instructions and save work,” Pokorny wrote. This personal anecdote helped him understand why mainstream AI requires significant resources for both memory and processing. Pokorny found the idea of spiking neural networks fascinating, as they potentially combine both functions more efficiently, similar to the human brain.
The efforts of neuromorphic experts to streamline computing systems struck a chord with reader Linda Ferrazzara. She noted the similarity between human brain development and neuromorphic computing. “I couldn’t help thinking about how human brains develop, with an initial surplus of neurons and connections that get gradually pruned into a more efficient configuration from prebirth to adulthood,” Ferrazzara wrote. She wondered if quantum computers could be adapted for neuromorphic computing systems.
Quantum computers perform powerful computations by leveraging quantum principles such as superposition and entanglement. While quantum and neuromorphic computing are distinct technologies, computer scientist Daniela Rus from MIT believes that neuromorphic processes might be used to control quantum computers. Additionally, “ideas from quantum mechanics may be useful in designing new chips for neuromorphic computers,” Rus says.
Computer scientist Prasanna Date from Oak Ridge National Laboratory suggests that quantum and neuromorphic computers could be used for different but complementary computations. “For example, quantum computers could train spiking neural network models, which are then deployed on neuromorphic computers for energy-efficient, real-time machine learning computations,” Date explains.
Corrections
In our previous issues, we corrected two significant errors. In the February feature “Holding back a glacier,” the opening image was misidentified as Thwaites Glacier when it was actually Pine Island Glacier. In the March issue’s “Have 5 years of COVID-19 readied us for what’s next?” a sentence was corrected to read: “Nearly 17,000 people in the United States died of COVID-19 in the last week of that year.”
