Bob Wachter
speaker
382 appearances
1 recordings
1 series
first heard Apr 2024
last heard Apr 2024
Bob Wachter’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsNo recordings in the last 12 months.Older appearances are listed below; set an alert to hear about the next one.
Appearances
I'll say one more thing about trustworthiness.
I just had another paper that I wrote with uh Juliet Namilstein and a colleague in Toronto that'll be coming up Panjamma soon.
But the thing that in some ways worries me the most in medicine, if the AI was right 50% of the time, we wouldn't use it.
It would be worthless.
If it was right a hundred percent of the time, it would be great, except we wouldn't have jobs.
For the foreseeable future, it's gonna be right ninety percent, ninety-five percent, eighty-five percent.
And so the safety system, of course, will be the AI s says something, makes a recommendation or gives you a draft diagnosis or drafts your note.
And then the fail safe is the doctor reads it over.
And our point in this article is that it's an inherently unsafe system because humans suck at that.
And they suck at it for a lot of reasons.
One is we're not good at vigilance, particularly when the s the thing that the thing that's giving us the first pass answer is generally right.
So we we will fall asleep at the switch.
The second is we will de-skill, we'll we'll we'll we'll sort of lose over time, we'll not be as good as we used to be, and we won't know uh we won't know that's wrong.
The third is this thing called automation bias, which is the bias that we have to trust the technology.
Even when it's wrong.
So the argument we have made is we better approach this intentionally and strategically and say, you know, this system, which seems really safe, the doctor will always do the final reading, is an inherently flawed system.
And there are ways, we can talk about it if you want, but there are ways of probably making it more robust.
Uh but if you just say that's the system, the AI will do its thing and the doctor will do the final sign off, uh, that's gonna get
R get it wrong a fair amount of the time.
chatbot like like the Bank of America or United Airlines.
Showing 121–140 of 382 · page 7 of 20
← Previous
Next →