I had two unusually interesting experiences with AI at Work this week. In one, Chat GPT diagnosed a quite arcane computer software problem, involving a commerical site, my lap top, and my router.
In the other, I asked Chat GPT to take a kind of weird unexpected position on the value of Point of Care Testing, and it came back with some interesting ideas.
### SOFTWARE DEBUGGING
Header: Chat GPT fixed a complicated multi-part software problem that sidelined my work for half a day.
Nowadays we use a lot of cloud-based software (Google Drive, etc). I use a database and information management system called NOTION. For a day or two it starting to perform really badly - web pages refused to open, or took 60 seconds. After even simple pages loaded, a progress symbol would keep spinning, as if trying to download more.
My first guess was a (rare) server problem at NOTION, but their system was up. Chat GPT advised me to try running my desk top off my iPhone wifi - and NOTION ran far better. Chat GPT announced the problem was not NOTION, not my office computer, but my ... router. After some back and forth, we disabled the 6GHz band on my (4/5/6 GHz) router, which, Chat says, also unloads router software called "WIFI 7." NOTION now worked.
While this solution may sound simple when explained, getting there from a mysterious glitch was an impressive pathway. I never could guessed that my ethernet (hard wire) computer could be messed up by a wifi setting (6 GHz) I had never even heard of.
(Boring, but long dialog here.)
### NEW IDEAS IN POCT
Header: Chat GPT Came up with "interesting new ideas" by applying a novel theory to a studied problem.
I was reading some recent review articles about Point of Care Testing (POCT). For some reason, I had a stray thought about "Shannon Information Theory." I barely know what that is, but I gave Chat GPT several recent review articles, and asked it if it could figure out new, interesting value implications by looking through the lens of Shannon Information Theory. It came up with a number of at least very interesting ideas. To my eye, it was an example of AI doing pretty interesting "thinking" and "ideas."
Note: I am "NOT" saying it came up with publishable, ground-breaking ideas or that anybody would do a PhD thesis on this. Rather, that it was doing interesting or suprising work.
For more detail of what it actually said, here.
Summary of its ideas here:
Conventional health economic and outcomes research (HEOR) analyses of point-of-care testing (POCT) emphasize turnaround time, workflow efficiency, and operational convenience. In the present project, a paired conceptual analysis applies Shannon information theory and modern decision-theoretic frameworks to re-examine POCT as an information-processing system rather than a logistics innovation.
Using qualitative reasoning and simplified mathematical formulations, the essays model the diagnostic encounter as a communication channel linking latent disease states to clinical action. The analysis demonstrates that POCT fundamentally restructures this channel by reducing information loss (“erasures”), preserving clinical context, enabling feedback-driven sequential decision-making, and aligning diagnostic information with disease dynamics. Even when analytic sensitivity and specificity are identical to central laboratory testing, POCT increases effective information yield, improves physician–patient communication fidelity, and raises the expected utility of diagnostic information. At the system level, same-visit clarification reduces care-pathway entropy, improves triage accuracy, and stabilizes downstream resource allocation.
This new information-theoretic framing provides a unifying explanation for observed POCT advantages reported in the [conventional] HEOR and implementation literature and offers a complementary theoretical foundation for evaluating POCT value beyond speed, cost, or assay performance alone.
