Reply to thread

Message

<blockquote data-quote="Umbran" data-source="post: 9857102" data-attributes="member: 177">My wife is a veterinarian, and takes a lot of referrals from other clinics.The cost of AI is hidden in that &quot;accurately&quot; grabbing &quot;important&quot; details.&nbsp; While I cannot speak to your friend's experience, more broadly the observed failure rate is still too high - too often the information is inaccurate, or details important to diagnosis and treatment are not present in AI summaries, and when you hand them off to the next person, they wind up in a bad situation, and patient care can suffer.Moreover, research is finding that, &quot;writing your own summary,&quot; and, &quot;having an AI write a summary, and you review it for correctness,&quot; are not cognitively the same thing for humans.&nbsp; The result is that the human who reviewed an AI summary will themselves remember the details less completely and holistically than the person who wrote their own summary. One thing we must be aware of is that risk is a combination of probability and severity of the bad result.&nbsp; In one way, a 5% error rate doesn't sound too bad, until you wrap into it that the fact that the bad result may be death.And, adding the caution that &quot;ultimately, the user must review AI generated results for accuracy&quot; (and the legal terms associated with tool use that we never read) sets to absolve the tool maker from responsibility for the results of the tool use.</blockquote>

[QUOTE="Umbran, post: 9857102, member: 177"] My wife is a veterinarian, and takes a lot of referrals from other clinics. The cost of AI is hidden in that "accurately" grabbing "important" details. While I cannot speak to your friend's experience, more broadly the observed failure rate is still too high - too often the information is inaccurate, or details important to diagnosis and treatment are not present in AI summaries, and when you hand them off to the next person, they wind up in a bad situation, and patient care can suffer. Moreover, research is finding that, "writing your own summary," and, "having an AI write a summary, and you review it for correctness," are not cognitively the same thing for humans. The result is that the human who reviewed an AI summary will themselves remember the details less completely and holistically than the person who wrote their own summary. One thing we must be aware of is that risk is a combination of probability [I]and severity[/I] of the bad result. In one way, a 5% error rate doesn't sound too bad, until you wrap into it that the fact that the bad result may be [I]death[/I]. And, adding the caution that "ultimately, the user must review AI generated results for accuracy" (and the legal terms associated with tool use that we never read) sets to absolve the tool maker from responsibility for the results of the tool use. [/QUOTE]

Verification