I think the danger here could be stated as, "what is the safe advice that can be given?" In an environment where an AI can easily go off the rails, and there are thousands of normal biological variations, and yet the internet tends to feed various forms of psychosomatic conditions?
It's always "safe" to advise someone to see a doctor. Talk to a clinician. Get a test of some sort. It may not be efficient, or clinically indicated, or the lowest cost option. But it's safe.
Especially when your concerns as an AI company are, "can we be sued for this?"
Thus we might find a flowchart of possible patient interactions where every single option ends with, "you need to talk to your physician, and the sooner the better!"
Cosmos - 2 months ago
Organizations will need to evaluate and test solutions EXTREMELY carefully here. As Pim points out, the devil is in the details and the wrong product can make things much worse. Ideally, an AI assistant will help patients craft a complete inquiry without extraneous details, and will research the context of the message to add relevant test results, history, and reference notes to assist the clinician on the other side. Organizations should push back hard on any vendor that can't demonstrably improve this process.
Skeptic - 2 months ago
Look forward to AI-generated questions being answered by AI drafts and/or full responses. An unnecessary layer between humans.
bob - 2 months ago
I agree with Pim ... the choices and the ramifications are all intrinsically linked. i.e. one may beget another
Pim Volkert - 2 months ago
Choosing an answer here is challenging, as it depends on how AI is implemented.
If AI assists patients by helping them frame their questions accurately and providing enough context, it can be beneficial.
However, if its role is limited to simply relaying messages, it may increase workload.
Additionally, excessive AI intervention in rewriting a patient's question could raise liability and safety concerns.
I think the danger here could be stated as, "what is the safe advice that can be given?" In an environment where an AI can easily go off the rails, and there are thousands of normal biological variations, and yet the internet tends to feed various forms of psychosomatic conditions?
It's always "safe" to advise someone to see a doctor. Talk to a clinician. Get a test of some sort. It may not be efficient, or clinically indicated, or the lowest cost option. But it's safe.
Especially when your concerns as an AI company are, "can we be sued for this?"
Thus we might find a flowchart of possible patient interactions where every single option ends with, "you need to talk to your physician, and the sooner the better!"
Organizations will need to evaluate and test solutions EXTREMELY carefully here. As Pim points out, the devil is in the details and the wrong product can make things much worse. Ideally, an AI assistant will help patients craft a complete inquiry without extraneous details, and will research the context of the message to add relevant test results, history, and reference notes to assist the clinician on the other side. Organizations should push back hard on any vendor that can't demonstrably improve this process.
Look forward to AI-generated questions being answered by AI drafts and/or full responses. An unnecessary layer between humans.
I agree with Pim ... the choices and the ramifications are all intrinsically linked. i.e. one may beget another
Choosing an answer here is challenging, as it depends on how AI is implemented.
If AI assists patients by helping them frame their questions accurately and providing enough context, it can be beneficial.
However, if its role is limited to simply relaying messages, it may increase workload.
Additionally, excessive AI intervention in rewriting a patient's question could raise liability and safety concerns.