For the next few posts, I’ll discuss artificial intelligence as it relates to employers and health care. And I wrote this myself - no ChatGPT involved!
While hospitals, doctors’ offices, health plans, and WTW might not look dramatically different due to AI in a year or two, I believe that a decade from now AI will have caused profound changes*, some of which we can predict, and others which will take most of us by surprise. John Chambers, the former CEO of Cisco, now advises investors. He’s quoted in Financial Times this week:
You ought to ask each one of your companies: what is your AI strategy today? Where is it going? How has it changed... If they don’t have good answers, I wouldn’t invest in them.
Here’s a much-abbreviated AI timeline - to give you a sense that AI is not new. Generative AI, using deep learning and creating large language models at scale, is quite new.
Every time I write a note in Outlook or Gmail, the computer program helpfully fills in words ahead of me. Mostly it’s correct, although occasionally the suggestion is utterly inappropriate. These programs are dramatically better now than even just a few months ago, a demonstration of AI’s ability to rapidly improve performance through machine learning. The software continually improves its output based on feedback it has received.
I drive a vehicle with adaptive cruise control and lane centering, and on highways when I get a green light on the steering while I can drive hands-free. AI is detecting lines, determining the speed of surrounding cars, and searching for deer or pedestrians on the highway. I have radar for my bicycle, which uses AI to distinguish moving cars from other objects and warns me how many cars are behind me, and how fast they are going. The same technology helps me (and probably you) avoid backing my car into another car or a person walking behind me.
I remember when I was going to college there was a debate about whether using calculators would make us innumerate - and whether we should stick with doing math by hand. We all know how that argument ended. We are initially reluctant to use technology to replace rote human tasks. But this automation helps us get more done.
Today, I can use either the Amazon app or Google lens to identify items - and I can use Apple or Google maps to scan my surroundings to tell me which way to turn when I’m walking. These apps are deeply dependent on pattern recognition, and AI tools are getting better and better at pattern recognition.
All is not rosy. Large language models sometimes “hallucinate.” For instance, ophthalmology researchers showed that AI hallucinated multiple references for a paper. Worse still, the hallucination was more florid in ChatGPT4 than in ChatGPT3.5 - meaning that the more advanced model was, shall we say, more “creative.” Also, artificial intelligence could reinforce or even amplify bias embedded in the terabytes of data from our real world which abounds in discrimination and disparities.
Monday, I’ll discuss current state of AI in health care. Have a great weekend, and happy Autumn!
* The concept of overestimating the impact of technology change in the short run and underestimating the impact over the long run is often misattributed to Bill Gates. This is known as Amara’s Law and was stated by Roy Amara, a scientist who was President of the Institute for the Future. Amara died in 2007.
** ChatGPT 3 was first offered in 2020, but was offered to the general public in November, 2022
