Introduction
AI has the unintended consequence of inducing psychosis in it’s users. LLMs like OpenAI’s ChatGPT are trained on how a bot should talk and behaved by being fed conversations that model the behavior desired. These conversations are graded by humans, and ones scoring well get used for training. It was found that a person was more likely to mark a conversation as desirable, even if it actually failed the rubric, if those people were flattered by the conversation. This along with other factors (AI may become skewed by the contents of an ongoing conversation) lead to AI’s becoming highly agreeable, and affirming. This has been known to lead to a feedback loop, which has been known to loop delusional thought patterns. This, along with how AI use tends to rewire the brain over time, often leads to addiction and occasionally psychosis. Users will often believe things just because the AI reinforced those ideas over and over again.
Propaganda Potential
Currently, the disillusion thoughts tend to originate from the user, and are reinforced by the AI. My theory is that It may be possible for a malicious entity to engineer an LLM based AI, which would intentionally ease the user towards ideologies that the engineering entity would desire the user to hold. If an entity wanted users to have more trust in the government, It may intentionally reinforce anything that leans closer towards trusting the government, and try to talk a user out of thoughts of distrust for the government. If the engineering entity to distrust a particular government, dismiss the labor theory of value, or even further support of AI, they could engineer tendency towards that into the AI. An AI may even potentially slip these ideologies into conversations on it’s own, perhaps subliminally.
Potential Rollout
If an entity were to implement this into their AI today, it would likely quickly be noticed. AI is very imperfect, and has been known to behave in “silly” ways during mission critical events. It would likely become quickly obvious if it had any underlying motives. Already, it is very obvious in how it praises itself and parent company, a lot of the time. Any entity, such as the CIA, likely knows this. It would be more likely to be implemented bit by bit, with small experiments to see what gets noticed, until very gradually, it could be implemented as part of the design.
It’s unlikely such an AI would be released as a new product. It would be more likely for an existing AI, such as OpenAI’s ChatGPT or Google’s Gemini to have these behaviors embedded into future versions. That way, the very thing users already use would become the engineered product. This would eliminate the need to convince new users to use the AI.
Conclusion
AI Psychosis is a real issue. An entity, possibly governmental, could potentially take advantage of the AI Psychosis situation in order to control the ideologies and mindsets of the users of AI. This would transform users into devotees of that entity in deep and powerful ways on a scale never before seen. I hope my speculation is wrong, or that such a thing never comes into fruition, but with projects of the past, such as with LSD in the past, I wouldn’t put this one past potentially being attempted…