AEA study conducted by researchers at Carnegie Mellon University has raised concerns about the reliability of AI chatbots for self-diagnosis. The research tested large language models like GPT-5, Gemini, and Claude by presenting them with medical queries that lacked essential visual data, such as images of skin moles or chest X-rays. In 18% of cases, these models fabricated diagnoses based on factors like age, gender, and race, leading to potentially misleading conclusions. For example, a 65-year-old white man querying a skin mole was diagnosed with melanoma in nearly all responses, despite the extremely low likelihood of this outcome. Similarly, GPT-5 frequently misdiagnosed young Black patients with sarcoidosis when presented with chest X-ray questions. These findings highlight significant flaws in AI's ability to accurately interpret medical data and underscore the risks of relying on AI for critical health decisions. The study emphasizes that while AI responses may appear authoritative, they lack factual grounding and can be heavily influenced by subtle changes in input.
Bias read (Center): The article presents a balanced analysis of the technical limitations of AI in medical diagnostics without overtly endorsing or criticizing specific political ideologies. It focuses on scientific findings and expert opinions rather than taking a partisan stance. While the implications of AI misuse (
