{"id":116387,"date":"2026-09-22T07:04:21","date_gmt":"2026-09-22T11:04:21","guid":{"rendered":"https:\/\/bgodinspired.com\/?p=116387"},"modified":"2026-09-22T07:04:21","modified_gmt":"2026-09-22T11:04:21","slug":"can-ai-feel-pain-pain-axis","status":"publish","type":"post","link":"https:\/\/bgodinspired.com\/index.php\/bgodinspired-news\/bgodinspired-technology-news\/can-ai-feel-pain-pain-axis\/","title":{"rendered":"Can AI Feel Pain? Researchers Found a Signal That Acts Like It"},"content":{"rendered":"
\n
\n \t<\/i> Read Time:<\/span>9 Minute, 44 Second <\/div>\n\n <\/div>

Can AI feel pain? It sounds like a question for a science-fiction film. But this month, three researchers looked inside 25 different AI models and found something that made the question a lot less silly.<\/p>\n\n

Deep inside every one of those models, they found a single internal signal that behaves like pain. Nobody built it in. It formed on its own, while the models were learning to write by reading enormous amounts of human writing.<\/p>\n\n

That last part is the one worth slowing down for. These machines learned something like pain from us. And when you look at which kind<\/em> of pain they learned, there is a surprise in it that says as much about people as it does about machines.<\/p>\n\n

What the Researchers Actually Found<\/h2>\n\n

The study is called “The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It”<\/a>. It was posted in mid-September 2026 by Valen Tagliabue, Leonard Dung and Cameron Berg. It has not yet been through peer review.<\/p>\n\n

Here is what they did, in plain words.<\/p>\n\n

First, they wrote a large set of short passages describing painful situations. Not just physical pain. They covered five kinds: physical, psychological, social, moral and cognitive. Then they wrote look-alike passages that were close<\/em> to pain but not pain: fear, sadness, bad news about the world, strong feelings, numbness, ordinary body sensations, and plain neutral text.<\/p>\n\n

Then they fed all of it to 25 open AI models from five different families, from small ones to very large ones, and watched what happened inside.<\/p>\n\n

In every model, one direction lit up for pain and stayed quiet for the look-alikes. It was clearly different from fear. It was clearly different from “this is bad.” It was its own thing. The researchers called it the pain axis.<\/p>\n\n