New JAMA exchange over whether Katie Goodman & I discriminated against AI 'in violation of...UNESCO ...Bioethics and Human Rights'

when we advocated for careful evaluation of clinical AI

Some points....

jamanetwork.com/journals/jama/…

Adam Rodman Arjun (Raj) Manrai

I, too, picked up chainsmoking and cliff-diving in response to minor improvements in grammaticality of computer-produced sentences.

Florence Nightingale, born #OnThisDay in 1820, is best known as the ‘lady with the lamp’ and the founder of modern nursing. She was also a brilliant statistician and pioneered the use of data visualisation. #WomenInSTEM bit.ly/3WeBwHc

We can endeavor to understand many relevant empirical facts about the world, even if deeper mechanisms evade us. May not satisfy pure scientists, but should be enough for us applied folk in medicine.
pubmed.ncbi.nlm.nih.gov/34748907/

🤩

Editorial of the year. Can we please stop doing this to patients? ⤵️⤵️⤵️🙏

I've been waiting for this study to be released!

The best test I've seen so far looking at the ability of faculty selecting residents to determine AI vs actual personal statements -- and fortunately the humans can actually tell, and rate the human-written ones far better.

How will #AI impact #MedEd ?

This concise #ResearchReport by Vishnu Nair showed faculty distinguished #GPT4 -generated #PersonalStatements from genuine ones 75% of the time, & authentic ones scored higher!

The personal touch prevails...

...for now...

rdcu.be/dGKfG

Published yesterday:

'Isabel Pro produced 175 (87.1 %) correct diagnoses and ChatGPT4 165 (82.1 %). ... ChatGPT4 fabricated citations and DOIs, producing 145 correct references (87.9 %) but only 52 correct DOIs (31.5 %).' #MedEd #CDSS

degruyter.com/document/doi/1…

Out today in JAMA Network Open! Cross-sectional study w/ 10 000 pairs of ED visits, OpenAI GPT accurately identified the patient w/ higher acuity when given pairs of deidentified presenting histories. Work by Chris Williams Travis Zack Brenda Miao Madhumita Sushil Michelle Wang

LLMs may be able to accurately identify higher acuity patient presentation when given pairs of presenting histories extracted from patients’ first ED documentation. ja.ma/3y3M55E Atul Butte

In hospitalized patients with Alzheimer disease and related dementias and oropharyngeal dysphagia, no difference seen in mortality rates between patients receiving thick vs thin liquids. ja.ma/3QwZbyK

Unsurprisingly objective external metrics (exam scores) were better predictors of future performance than subjective internal metrics (milestone ratings).

Coming soon on NEJM AI Grand Rounds: Andrew Beam and Arjun (Raj) Manrai make fun of me for casually throwing around epistemology, nosology, simulacra (and simulation), and more ...

But seriously, had a great time!!! I will post when it is out!