Science & Technology

An AI Designed These Freaky Images to Super-Stimulate Monkey Neurons

Artificial intelligence is all the rage these days, but this new technology industry is not just creating electronic personal assistants like Siri and Alexa. AI algorithms can actually teach us a lot about how the human brain functions. In the case of the XDREAM algorithm, researchers gained a better understanding of how the brain picks out relevant stimuli in its environment.

XDREAM On

You might not think of neurons as having preferences, but the building blocks of your brain are pickier than you'd think. Most visual neurons are pretty selective in their image preferences and fire more in response to certain stimuli — for example, some respond most to specific colors, other to certain curves or directions, and still others to complex shapes like hands or faces. Those discoveries were made via the painstaking approach of trying out different stimuli and seeing which ones led to greater firing rates.

In a study recently published in the journal Cell, researchers designed a genetic algorithm called XDREAM that generates images meant to maximize neural firing rates. The algorithm begins by presenting a series of images while measuring neuronal firing rates, then gradually mutates and combines the images to create a new stimulus that neurons in the visual cortex will "like" best. In essence, the XDREAM algorithm allows neurons to build their preferred images from scratch.

"In this way we have evolved a super-stimulus that drives the cell better than any natural stimulus we could guess at," said Margaret Livingstone, a senior author on the study. "This approach allows you to use artificial intelligence to figure out what triggers neurons best. It's a totally unbiased way of asking the cell what it really wants, what would make it fire the most."

Separating the Signal From the Noise

The XDREAM team tested the algorithm on macaque monkeys at Harvard Medical School. They found that although the first few sets of images looked like unintelligible noise, the evolved images resembled familiar sights in the monkey's environment.

Evolved XDREAM images looked like distorted versions of monkey faces, human faces wearing surgical masks, and the food hopper in the monkey's room. Based on this finding, researchers believe the brain is able to distinguish relevant stimuli in its environment.

"We are seeing that the brain is analyzing the visual scene, and driven by experience, extracting information that is important to the individual over time," said Carlos Ponce, a co-first author on the study. "The brain is adapting to its environment and encoding ecologically significant information in unpredictable ways."

The research team believes that the XDREAM algorithm can be applied to other types of neurons that process sensory information, such as auditory, hippocampal, and prefrontal cortex neurons. This could have implications for understanding how the brain processes other sensory stimuli — and how individual neurons pick their "favorite" or most relevant sights, sounds, and memories.

And learning about the brain could even help scientists to create better artificial intelligence technology in the future.

"As artificial intelligence researchers develop models that work as well as the brain does — or even better — we will still need to understand which networks are more likely to behave safely and further human goals," Ponce said. "More efficient AI can be grounded by knowledge of how the brain works."

Get stories like this one in your inbox or your headphones: Sign up for our daily email and subscribe to the Curiosity Daily podcast.

Learn more about AI in "The Sentient Machine: The Coming Age of Artificial Intelligence" by Amir Husain. We handpick reading recommendations we think you may like. If you choose to make a purchase, Curiosity will get a share of the sale.

Written by Andrea Michelson May 17, 2019

Curiosity uses cookies to improve site performance, for analytics and for advertising. By continuing to use our site, you accept our use of cookies, our Privacy Policy and Terms of Use.