Fall 2023 (Volume 33, Number 3)
AI Overload
By Philip A. Baer, MDCM, FRCPC, FACR
Download PDF
Artificial intelligence (AI) and the associated AI-related
hype seem to be everywhere. Are the robots
taking over? Will they outsmart us? Will knowledge
workers’ jobs, including those of rheumatologists and
other physicians, be threatened?
In 2016, Canadian machine-learning pioneer Geoffrey
Hinton spoke at a machine-learning conference in
Toronto and said: “I think if you work as a radiologist,
you are like the coyote that’s already over the edge of the
cliff but hasn’t yet looked down. People should stop training
radiologists now. It’s just completely obvious within
five years deep learning is going to do better than radiologists….
It might be 10 years, but we’ve got plenty of radiologists
already.”1
Rheumatologists were not mentioned, and as of 2023
Hinton appears to be incorrect about radiologists, but
what are the implications for us?
Evidence-based medicine is one of our cornerstones,
so let’s examine the evidence. ChatGPT and other large
language models (LLMs) seem to have impressive capabilities.
They can write code, create reports and letters,
and apparently can pass some medical licensing exams.
However, an AI candidate failed a mock radiology fellowship
examination recently, struggling the most with
musculoskeletal imaging.2 As well, AI chatbots cannot
always be trusted, as they “hallucinate” by inventing
“facts” and references which are false. They may violate
copyrights and privacy and plagiarize the material they
review. Ethical AI is not the current reality. The current
iteration of ChatGPT was apparently trained on what it
could find online up to 2021, so newer research will not
be incorporated into its work product.
Medical journal editors are concerned enough to limit
the use of AI in producing papers and require the disclosure
of AI assistance in generating submissions. Can an
AI program be listed as the author of a medical paper?
Currently, the answer is no.3 Yet there are already reports
of a corporate board including an AI program named VITAL
as a board member.4 Advantages: perfect attendance,
reliable completion of pre-reading, and no costs to feed it.
I reviewed studies related to rheumatology uses of
AI presented at EULAR 2023 for a recent presentation.
The field is burgeoning, with a literature search indicating
an annual increase in papers citing AI of between
20 and 48 percent (Abstract AB1667). Multiple studies
support the usefulness of AI in reporting of imaging studies,
using convolutional neural networks (CNNs) developed
with training sets followed by testing on validation
sets. The CNNs did as well as expert readers in multiple
studies: hand/wrist X-rays looking at Sharpe/van der
Heijde scores in rheumatoid arthritis (RA) (POS0160),
RAMRIS scoring of hand MRIs in RA (OP0002) and scoring
of X-rays (POS0896) and MRIs (AB1013) for sacroiliitis
in axial spondyloarthropathy. LLMs using natural
language processing (NLP) could review patient data in
EMRs and EHRs and find previously unsuspected cases of
ANCA-associated vasculitis (POS1179). Models could
also be constructed in RA to evaluate baseline factors and
predict outcomes one year into the future, potentially selecting
some patients for more intensive follow-up and
treatment if the prediction was for high disease activity in
the future (POS0320).
Meanwhile, what happened in radiology since Hinton
pronounced it was doomed in 2016? Well, there are now
200 FDA-approved radiology AI algorithms ready for use,
according to the American College of Radiology’s AI Central
site. At the same time, there is a global radiologist
shortage, driven in part by overwork. “The amount of
imaging is going up 5 percent per year, and we’re not training
5 percent more radiologists per year,” according to
Jordan Perchik, MD, a fellow in the Department of Radiology
at the UAB Heersink School of Medicine. “The most
commonly used AI tools,” Perchik said, “are ones that
speed up scans, paradoxically increasing the workload for
radiologists.”5
Hinton has also been rebutted by Stanford radiologist
and AI pioneer Curtis Langlotz, MD, PhD. He said, “AI
won’t replace radiologists, but radiologists who use AI
will replace those who don’t.”6 Maybe that is what rheumatologists
can glean from all the AI hype. Our brand of
longitudinal holistic care will never be obsolete, but we
may have to integrate AI tools to cope with the increasingly
severe mismatch between supply and demand we
are faced with.
Now, please excuse me while I follow the advice of the
legacy AI troika of Alexa, Siri and Cortana and invest all
my spare cash in Amazon, Apple and Microsoft.
Philip A. Baer, MDCM, FRCPC, FACR
Editor-in-chief, CRAJ
Scarborough, Ontario
References:
1.Geoff Hinton: On Radiology [video]. Available at https://www.youtube.com/watch?v=2HMPRXstSvQ.
Accessed August 23, 2023.
2. Murphy Hannah. AI 'candidate' fails to pass mock radiology boards. Health Imaging. December
2022. Available at https://healthimaging.com/topics/artificial-intelligence/ai-candidate-fails-pass-mock-radiology-boards. Accessed August 23, 2023.
3. Solomon DH, et al. ChatGPT, et al. Artificial Intelligence, Authorship, and Medical Publishing. Arthritis
Rheumatol. 2023 Jun;75(6):867-868. doi: 10.1002/art.42497.
4. Deep Knowledge Venture's Appoints Intelligent Investment Analysis Software VITAL as Board
Member. Globe Newswire. May 2014. Available at https://www.globenewswire.com/news-release/2014/05/13/635881/10081467/en/Deep-Knowledge-Venture-s-Appoints-Intelligent-Investment-Analysis-Software-VITAL-as-Board-Member.html. Accessed August 23, 2023.
5. Windsor, Matt. This radiologist is helping doctors see through the hype to an AI future. UAB Reporter.
December 2022. Available at https://www.uab.edu/reporter/people/achievements/item/9925-this-radiologist-is-helping-doctors-see-through-the-hype-to-an-ai-future. Accessed August 23,
2023.
6. Center for Artificial Intelligence in Medicine & Imaging (Stanford University). RSNA 2017: Rads who
use AI will replace rads who don’t. December 2017. Available at https://aimi.stanford.edu/news/rsna-2017-rads-who-use-ai-will-replace-rads-who-dont-0. Accessed August 23, 2023.
|
