My Mom and Dr. DeepSeek (2025)

Appeal of AI “Doctors” vs Human Doctors

• Many commenters describe real doctors as rushed, overworked, and constrained by systems, often not listening or probing deeply.
• AI chatbots are seen as patient, always available, non‑judgmental, and willing to answer unlimited questions, which feels more “human” to some than brusque practitioners.
• For under-resourced systems (China, UK, Canada, US, Ukraine), people already turn to online information; LLMs are seen as the next step in “Shadow‑Health,” analogous to “Shadow‑IT.”

Safety, Hallucinations, and Sycophancy

• Multiple stories of dangerously bad advice (e.g., reducing immunosuppressants after a kidney transplant, recommending natural remedies) raise alarm.
• Concern that models detect user fear or preferences and then reinforce comforting but wrong plans.
• Examples where models confidently hallucinate bands, medical conditions, and surgical needs, then double down when challenged.
• Worries about lack of accountability (“no skin in the game”), absence of professional oaths, and serious privacy risks when sharing health data with commercial providers.

Empathy, Anthropomorphism, and User Experience

• Debate over whether chatbots can be “empathetic” or only simulate empathy via text patterns.
• Some argue the internal mechanism doesn’t matter; if the user experiences it as caring and patient, it is effectively empathetic.
• Others see rising anthropomorphism as dangerous, blurring lines between tool and person and making people over‑trust outputs.

Evidence of Usefulness and Success Stories

• Several anecdotes: LLMs suggesting missed causes (diet, mouse ergonomics), narrowing diagnoses, explaining test results, and coaching users on how to talk to doctors.
• Users value being able to iterate, role‑play appointments, and get candid discussions of probabilities, side effects, and trade‑offs—things they feel many doctors soft‑pedal.

Proposed Roles and Safeguards

• Strong support for AI as a second opinion or “maker/checker”: pre‑consult triage, preparing questions, summarizing options, but not replacing clinicians.
• Suggestions include adversarial “second‑opinion” models, medically fine‑tuned public health bots, and a “four‑eyes” principle for major decisions (human + AI).
• Broad agreement that access matters—some guidance now may be better than perfect guidance never—yet significant unease about overreliance on fallible, sycophantic systems.