From reading books aloud to mastering chess to predicting structures of complex proteins, machine learning grew mighty sophisticated over the past decade, said Read Montague, a Virginia Tech researcher whose career focuses on artificial intelligence.
Processes driving machine learning are similar to concepts behind human and animal thought, Montague said during a lecture Thursday evening.
“Over the last nine years, an explosion in neural network learning … has had a convergence with the kinds of algorithms that evolution put in your head a long time ago,” Montague said. “Probably 300- or 400-million years ago.”
There’s more to learn yet about those ancient and durable algorithmic processes, he said. They impact how molecules like dopamine and serotonin are produced and controlled in the brain, all the way to what leads you decide between chocolate ice cream or carrot cake for dessert.
People are also reading…
“It is pretty clear that the algorithms that are living in a bee’s brain and living in your brainstem, at the very basic level, are really the same,” Montague said. “These are the systems that make you chase food, sex, salt and water. You are driven to do these things, it propagates the species.”
Algorithms common between bee and human brains are also in the minds of crocodiles, he said. The difference for humans comes from our comparably larger brains — and the impressive inventions, like computers, human brains create.
“I couldn’t dump 500 million pictures into your head in 17 lifetimes if I showed you a picture every second,” Montague said. “You don’t live long enough … and your brain isn’t built to do that.”
But new machine learning models and the computers powering them are designed to process that much data and more. Although it comes at the cost of efficiency, because those computers produce a ton of waste heat, he said.
“We run on very low energy. That’s because we evolved,” Montague said. “Being efficient was crucial to staying alive. You have to eat less. And so your algorithms in your head are also efficient for the same reasons.”
The mysterious teachers inside your brain are an unthinkable number of neurons firing off information to learn from. Inside just a cubic millimeter of human brain are more than a billion connection points, and miles of “biological wire,” he said.
“And it all can heal itself, knows how to fix itself,” Montague said of the brain. “We don’t know how to build a computer like that. It has to be doing something fiendishly efficient, because it runs on very little energy at all, unlike our computers.”
There’s more for researchers to learn about biological learning, but growing convergence between animal and machine learning has great potential to further improve all sorts of industries and personal lives alike, he said.
“The brain is a really complex device,” Montague said. “We’re really in early days of understanding how it emits all these fantastic perceptions and behaviors.”
That comingling of machine learning and human thought is a modern frontier that Montague said he is exciting to be a part of as a professor and researcher at Virginia Tech and its Fralin Biomedical Research Institute. It’s a line of study with growing cross-disciplinary applications.
“One of the reasons my work has ventured into economics and psychiatry, the law and various other disciplines is that you’re the decision maker at the center of all that, people are,” Montague said. “These are natural excursions of neuroscience into domains that matter.”
Montague’s speech Thursday was the first in the institute’s 2023-24 Maury Strauss Distinguished Public Lecture Series, which runs monthly through May. Lectures are livestreamed and archived.