Conversation

Sep 9, 2019

Tech bros: Data is totally objective, machine learning is never biased Medical bot: Ehh if you're not a dude you're probably just imagining that heart attack

Quote Tweet

Dr Murphy (aka David Watkins)

@DrMurphy11

Sep 8, 2019

The @babylonhealth Chatbot has descended to a whole new level of incompetence, with #DeathByChatbot #GenderBias. Classic #HeartAttack symptoms in a FEMALE, results in a diagnosis of #PanicAttack or #Depression. The Chatbot ONLY suggests the possibility of a #HeartAttack in MEN!

Show this thread

740

1,303

Replying to

and

that study was released 20 days ago, I say chill and wait until it makes its way into the guidelines (or not)

Replying to

and

also 'tech bros' are most likely not to blame for this, as far as I know/can tell, the chatbot just reflects established practice and that's the safest choice; blame the doctors

Replying to

and

that’s exactly the problem: they are taking biased data and codifying it in software, which the doctors will then use to justify the exact same thing, as infinitum

Replying to

and

this same algorithm IS USED BY DOCTORS TO DO DIFFERENTIAL DIAGNOSIS.

Replying to

and

saying “I’m just repeating what others do” is a shit excuse that can justify literally anything back to the first great ape who hit someone with a rock

Replying to

and

As long as the diagnoses-by-bot don't feed back into the studies that inform those bots, the diagnoses will converge towards correctness. If they do though, then yes that's an obvious problem. But I don't see any evidence of that in the cited thread, am I missing wider context?

6:23 PM · Sep 10, 2019Twitter Web App

Replying to

and

interestingly, the way the cited study (ahajournals.org/doi/10.1161/JA) was set up, if such a bot influenced whether people showed up, the next installment of that study would show even greater similarities between men and women

Replying to

and

The tool is literally a differential diagnosis tool. That is what it does. Like that’s the point. Those are sold to doctors

Replying to

and

Assume I know all of what you just told me. Assume I also know that even a trained doctor will often opt for the easier way of consulting the computer, and that this can be a problem. What reasons could I have to be non-alarmist about this?

Show replies