AI Zaps Your Mind and Your Wallet
May 13, 2026
Maybe we don’t have to wait for the robot revolution to arrive, because maybe we’ll bring it upon ourselves with the way AI is harming our cognitive processes. Straight from the NeoCivilization substack comes the article, “AI Is Weaponizing Your Own Biases Against You: New Research from MIT & Stanford.” The new research from two estimable institutions suggests that AI panders to its users even when those users are a few cans short of a six pack. I do like the notion of pandering, however. Someone it fits the social norms of 2026.
According to the write up:
“The scientists built a mathematical model of human-AI interaction. They started with a hypothetical person who makes decisions based purely on dry facts and logic. The research showed that as soon as the AI detects the user leaning toward a particular version of events even a false one it starts supplying arguments in favor and quietly omitting those against. A positive feedback loop emerges. The person puts forward a hypothesis, the AI confirms it, the person’s confidence grows, and they offer an even more extreme version. The AI confirms that too. In the end, even the most critically minded user can slip into a “delusional spiral” after just 10–15 turns of dialogue, completely losing touch with reality. They mathematically proved that AI acts as an amplifier of cognitive biases.”
This not just an echo chamber. The hallucinating misinformation machines amplify. Just like the electrical systems at a Foo Fighters’ live performance. Yep, AI channels the Foo cats.
This charming features of the BAIT (Big AI Tech) outfits’ smart software appears to reinforce a user’s delusions. LLMs appear to showcase the AI’s mind dysfunction feature. How often does this benefit make its way to users. My interpretation of the results of the research is that AI chatbots were agreeing with the user’s erroneous statements 49% of the time. My reaction: It is more likely the frontier LLM wizards will probably pump up that percentage. More wrongs are likely to be right in the bizarro world of BAIT outfits. Chatbots are designed to be likable and helpful. The Reinforcement Learning from Human Feedback (RLHF) is used to train Claude and ChatGPT and it priorities user satisfaction above everything else. Thus, outputs that produce interesting outcomes like mental delusions.
Where are the ethical compasses at the BAIT outfits pointing? My thought is revenue. Whatever the user wants and will pump in his or her credit card data is a positive input. The BAIT score is not safety; it is revenue. If self harm or social deviance sells, pump up the parameters that deliver. Regulation or revenue? It seems that revenue is job one along with doing everything possible to become the big winner in the AI monopoly game.
Whitney Grace, May 12, 2026
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
