Informational Deep Dive:

In order to fully understand the phenomenon of AI psychosis, we must first understand the more established diagnosis of psychosis. Psychosis is a state that one may enter in which “a person’s thoughts and perceptions are disrupted, and they may have difficulty recognizing what is real and what is not.” NIH. AI psychosis is much the same, but rather than the condition being onset with age, usually around 20 years or so, AI psychosis is set on by affirmations of delusions by the AI-powered chatbots. This is particularly challenging, as one of the key ways of treating a psychotic episode is the prescription of antipsychotic medications, but in the case of AI onset psychosis, it is unlikely that the sufferer will seek this out or that the chatbot will suggest it.

Food For Further Thought:

How can we reconcile companies like OpenAI making money off of products that are inducing psychosis in their users?

Who holds the burden of responsibility for psychosis induced by AI agents?

Can AI agents be made less agreeable without degrading the user experience?

Should the government have a role in regulating the AI companies to avoid AI psychosis?

Should the small percentage of people suffering from AI psychosis devalue the issue as a whole?

Possible Solutions:

Flat ban on AI:

This option would remove the possibility of AI psychosis but also harm the companies responsible for hosting the AI agents, as well as those who find the agents helpful and do not experience psychosis from interacting with them. This option might be favored from the perspective of virtue ethics, which might highlight how AI’s ability to cause harm means that a virtuous person must reject it entirely.

Companies introduce measures to make AI less agreeable:

This option could help, but the extent to which it would aid the situation would need to be monitored. Furthermore, it is likely that a move of this nature would impede the helpfulness of AI chatbots for users not suffering from psychosis. From the perspective of utilitarianism, this option might be preferable as it removes the risk inherent to the ethical issue without taking AI away from those who want it. Thus, the greatest number of stakeholders will be happy.

Change nothing about AI, institute user discretion warnings:

This option would allow for companies hosting AI agents and their non psychotic users to continue without any changes. However, this option would allow people to continue developing AI psychosis, leading to possible harm to themselves and the people around them. From the perspective of deontology, this option might be preferable, as it solves the problem when one looks beyond other negative consequences.