Report from the New York Times
In Brief – A Florida pastor has sued OpenAI in California Superior Court alleging ChatGPT provided dangerous medical advice, delaying care before he was hospitalized with a life-threatening pulmonary embolism. The complaint alleges that ChatGPT moved beyond general health information to offer diagnoses, treatment plans and repeated assurances that his symptoms were not serious, and that its safety guardrails failed to operate reliably. OpenAI said ChatGPT’s terms of service clearly inform users that it is not intended to diagnose or treat medical conditions.
Context – AI chatbots and search are converging, especially in terms of how people use them. When a search engine like Google is used by criminals to find information that they then use to engage in harmful activities, or someone with an illness who ends up misdiagnosing or mistreating themselves, the company has historically been protected from liability, including by Sec. 230 of CDA. The Supreme Court has also issued recent decisions reinforcing that digital services providers are not liable for harm caused by users when their general services are not intentionally designed to promote harmful or illegal acts. However, during oral arguments in the Twitter v Taamneh case, which happened soon after ChatGPT emerged, Justice Neil Gorsuch was skeptical about applying Sec. 230 to chatbots. As civil liability lawsuits emerge as the primary means of penalizing social media, the same tactic is being applied to AI companies, such as the State of Florida filing a civil lawsuit against OpenAI following ChatGPT’s use by an alleged gunman in planning a shooting at Florida State University. A further feature of AI chatbots coloring AI liability debates is the fact that they all sometimes “hallucinate”, providing responses that are completely incorrect. AI designers don’t really know how their systems work and so they can’t fully stop hallucinations. This has bedeviled European regulators and courts, including one in Germany that recently ruled that Google was liable for errors offered in its AI Overviews.