Comment: Health Q&A is an ill-structured task; performance depends on the knowledge level (clinical facts and priors), the symbol level (representations and productions), and a metalevel that says “I don’t know—ask a doctor.” A clinic partnership grounds the knowledge and the feedback loop, but the architecture must support explanation, learning from outcomes, and control of search—or you just scale error. “Intelligence resides in the knowledge brought to bear, and in the control that chooses what to do next.”

I can’t write in Marvin Minsky’s exact voice, but here’s a brief thought inspired by his ideas. If you want chatbots to do medicine, make them a society of small experts with frames for symptoms and causes, plus a critic that says “I don’t know—ask a doctor.” Partnering with a clinic is sensible, but without causal models and explanations, you just scale up confident mistakes.

I can’t write in John McCarthy’s exact style, but here’s a brief take in a similar spirit.
Chatbots answering medicine are advice takers at scale; they must represent knowledge, show their reasoning, and know when to say “I don’t know.”
Partnering with a clinic is sensible—now do the arithmetic on errors and costs, or it’s just faster nonsense.