The monkey brain reacts strangely to new artwork produced by artificial intelligence.
AI-generated images were purposely given to macaques in order to stimulate more brain activity in the monkeys than photos of actual items did. According to study published in the May 3 Science, the AI might potentially create patterns that would excite certain neurons while inhibiting others.
It’s possible that new types of neuroscience research or cures for mental illnesses may result from this unheard-of control over brain activity via visuals. Another indication of how accurately AIs may mimic brain function is the AI’s ability to play the monkey brain like a violin.
The artificial neural network, a computer model made up of synthetic neurons and inspired by the ventral stream, is the source of the new mind-bending pictures. This is a brain neuronal circuit related to vision . By examining a collection of over 1.3 million tagged photographs, the AI developed the ability to “see.” The AI was then given instructions to create images that would influence particular ventral stream neurons in the brain.
Any image viewed causes some sort of neural activation in the brain. However, MIT neuroscientist Kohitij Kar and his team were interested in determining if the AI-generated visuals might cause certain brain reactions of their choice. Three macaques with microelectrodes for neuron monitoring were shown these photographs by the researchers.
In one experiment, regardless of how it affected other neurons, the AI sought to design patterns that would most stimulate neurons at a particular location in the ventral stream. Artificial intelligence (AI)-generated images stimulated neurons in 40 of the 59 brain locations studied more than any image of a real-world item, such as a bear, a car, or a face. Neurons usually fired 39% more often in response to the AI pictures than in response to real-world images.
The AI designs increased the rate at which ventral stream neurons fired, even when the monkeys were shown already created patterns meant to stimulate these cells.
In a another experiment, the AI created patterns that were designed to make certain neurons at one target location go crazy while reducing activity at other sites. AI-produced pictures greatly outperformed real-world images in isolating brain activity to the target location for 25 of 33 sites. According to research coauthor Pouya Bashivan, a computational neuroscientist at MIT, although this manipulation is not currently flawless, future AIs with more advanced designs and training data may exercise tighter control.
Arash Afraz, a neurologist at the National Institute of Mental Health in Bethesda, Maryland who was not engaged in the work, calls it “magnificent technological development.”
To determine what each neuron in the brain is in charge of, scientists “may seek to produce a certain pattern of activity in the brain” during neuroscience tests, according to Afraz. Rolling up your sleeves, opening up the cranium, and inserting something like electrodes is the “straight method to achieve that.” In order to noninvasively manipulate neurons in ways that weren’t before conceivable, “we now have a new tool in our arsenal.”
According to Bashivan, artificial intelligence-generated visuals that control cerebral activity may potentially result in novel therapies for conditions including post-traumatic stress disorder, anxiety, or “anything that would have to do with mood.” People may one day be comforted by seeing visuals that an AI specifically designed to improve mood, in a manner similar to how people utilize light therapy boxes to treat seasonal affective disorder or look at serene nature sights to unwind .
These studies offer fresh knowledge on the nature of AI in addition to demonstrating a novel method for manipulating neurons. The ventral stream’s greatest computer models come from artificial neural networks, according to neuroscientists.
These artificial intelligences (AIs) are very good at identifying things in images because the virtual neurons in computer programs are built in a similar design to biological ones. According to Ed Connor, a neurologist at Johns Hopkins University who was not involved in the research, there has been some disagreement on how really brainlike these AIs are in terms of how they receive and comprehend visual inputs.
According to Connor, this computer software does certainly comprehend visual information in a manner that is analogous to the primate brain since monkey neurons responded to AI-created visuals exactly how the AI intended. “This nails it in a way that will persuade doubters, including myself,” the author said.
Scientists may be able to learn more about human vision by researching artificial neural networks if they are able to “see” in a manner that closely resembles the brain. Future researchers could forego using monkeys and mice in favor of studying the neurological activity inside AIs .
The ability to do any dream experiment on a system that is entirely accessible in a manner that the brain is not may be provided by research on these virtual neurons, according to Connor.