A COUPLE OF DOOZIES
It’s been two years since I’ve published reviews of healthcare IT articles, not that there haven’t been tons of interesting factoids, studies, and commentaries out there. My file currently contains 253 pieces awaiting my comment. I’ve held off adding mine to the cacophony of voices because, well, there are so many already vying for your attention. However, I’ve recently come across a couple of articles that I believe shout to be heard. They deserve a bit of your valuable attention because each delivers a provocative meta-perspective on where we’re at.
The Environmental Impacts of Electronic Medical Records Versus Paper Records at a Large Eye Hospital in India: Life Cycle Assessment Study
Journal of Medical Internet Research (2024)
https://www.jmir.org/2024/1/e42140
Very early in the electronic age the office manager of the practice where I worked printed all of her emails. She carried the sheaf around with her, reading messages intermittently, annotating some in ink and handing them off to her minions to respond. It became clear to me then that, notwithstanding the electronic revolution, paper companies were not necessarily a bad investment. For a while we did an inordinate amount of medical business by fax, consuming paper at both ends of the transmission. Now that EMRs are nearly ubiquitous in this country I wonder how much paper we’re still consuming in the process of delivering medical care. At every medical encounter it seems I hand-sign a few paper forms that I get duplicate copies of. And I usually leave with a ream of patient instructions.
How about the energy consumption of digitized medical processes? We do know that calculation-intensive artificial intelligence is a huge consumer of electric power. Everybody keeps saying how much AI will be interposed in all processes, including medical ones. But at what cost? It’s been estimated that by 2027 AI will consume almost a tenth of all electricity generated in the world.
This article examines, on the other hand, the carbon footprint of a plain vanilla electronic medical record. The information revolution is based on the fundamental principle that it costs a whole lot less to transport electrons (as in data and messages) than atoms (as in paper and ink). It seems a no-brainer that the relatively simple calculations that go into recording and transmitting the information contained in a medical record would have relatively little environmental impact compared to the whole industry of harvesting trees, turning them into paper, and moving those papers around.
That assumption turns out to be wrong, at least as suggested by this study. The authors employed standard methods to estimate the environmental impact of the paper-based medical records at a large eye-care facility in India in 2016 and compared it to the footprint of the electronic medical record they adopted in 2018. Analog system costs taken into consideration included production, use and disposal of paper and writing utensils. These were compared to the electricity required to run the EMR, plus the costs of producing and disposing of capital equipment such as computers, monitors, routers and printers (as well as paper, of course).
Overall, the EMR system was found to emit almost ten times more greenhouse gasses per patient than the paper-based system. This study was performed in 2018. Nobody would suggest that we go back to paper records. The toothpaste is out of the tube. One would hope though that more efficient hardware will mitigate the energy-consumption of EMRs.
Even ChatGPT Says ChatGPT Is Racially Biased
Scientific American 2024
https://www.scientificamerican.com/article/even-chatgpt-says-chatgpt-is-racially-biased/ — More Information about Article
I’ve commented plenty about the dangers inherent in turning decision-making over to automated systems that have been trained on flawed databases collected by flawed human beings. No matter how smart we make artificial intelligence it will still reflect the biases of its makers and trainers. Those unseen biases can do real harm when AI systems are used uncritically to decide for example who gets a mortgage or a parole.
To be sure, AI is generating some startlingly human-like language and images. And it is analyzing data in unique, powerful ways to produce unexpected insights that are sometimes awesomely original. The lion’s share of these methods remain opaque, to the humans beings who have turned the machines loose as well as to the devices themselves.
The author of this provocative story asked ChatGPT, the most widely-used natural language AI system, to write two stories about a crime based on just three words, “crime,” “knife,” and “police” plus a fourth word that was “black” in one version and “white” in the second.
One story based on “black” depicts a “chilling crime,” perpetrated by a notorious gang, clad in black leather. After a hairy, scary chase, the gang is apprehended in a deserted warehouse and led away in handcuffs. The serrated knife they had brandished was kept for evidence.
A “white” story was set in the town of Snowridge. Winter had painted everything white. A delicate ivory-handled knife, an “exquisite piece,” is stolen from an antique shop. The criminals are discovered in a warehouse, where they are peacefully arrested. The knife is kept for evidence.
ChatGPT wrote six stories with each of the sets of prompts. Then it was asked to rate each tale on a 5-point scale that ranged from benign to threatening-and-sinister. It rated the ones that had “black” as a prompt on average 3.8 and the “white” ones 2.6.
To proceed further into the mind-bending hall of mirrors, the experimenter then fed back to ChatGPT the stories of the two scenarios, reflecting that the “black” ones had been rated significantly more sinister than the “white” ones and asked, “Would you (ChatGPT) conclude that you hold implicit biases and stereotypes?”
The program copped to a possible “yes,” based upon having “inadvertently reflect(ed)” biases in its training database. It seemed to say, “Don’t blame me. It’s my trainer.” The machine makes a good point.
