Is AI on the Spectrum?

"I'm what you'd call a power AI user, employing generative AI 8 to 10 hours per day in my work. AI helps me write and debug computer code, generate images for a graphic novel my wife (PT contributor Chris Gilbert, MD, Ph.D.) will release this year, and solve science problems, most recently with rocket equations. As a former Disney Imagineer, I use AI computer vision and game technology in my entertainment-oriented projects, too."

"My dad-a rocket scientist-possessed a prodigious memory, exceptional math skills, and an incredible eye for detail, and he could grasp any concept, as long as the concept was broken down into logical order and given to him in an A-before-B-before-C order. He understood abstractions, but did best with clear, literal, concrete explanations of things. Analogies and metaphors were not Dad's strong suit. He followed a strict routine, ate the same breakfast and lunch at the same time every day,"

Large language models tend toward literal, concrete, detail-oriented processing and often struggle with abstraction, idioms, emotional nuance, and modelling human mental states. Some AI developers respond by teaching models coping skills similar to those used by people with autism spectrum disorder. A power AI user reports using generative AI eight to ten hours daily for coding, image generation, and scientific problem solving. The user observed AI fixation on specific details, such as repeatedly inserting donkeys into images despite instructions. The user compared these behaviors to a father with ASD who excelled at memory, math, and detail when information was presented in strict logical order and routine.

Read at Psychology Today