I often use Microsoft Copilot or Google’s artificial intelligence (AI) to quickly summarize topics without having to read numerous source documents. I have noticed that the web browser versions of AI are good at summarizing but not critical thinking. In that regard, it is not so much artificial intelligence as it is an algorithm that looks for common themes and parrots them. I stumbled across a YouTube video about the sluggish economic growth in Great Britain since the 2008 Financial Crisis. The broad economic slowdown was attributed to (among other things) government austerity, not investing enough in public services and military support for Ukraine. I found those answers unsatisfying. I asked Copilot, which gave me an identical answer. Then I used Google and found the same thing. But here is the catch: I had not found numerous independent sources all confirming my question. I had likely found articles based on related articles sourced from various Keynesian journal articles and Leftwing opinion pieces dating back years. Taxing consumers (reducing their spending) to increase government spending is not an investment that will boost economic growth. Rather, it sounds like British Labour Party dogma.
Training AI requires consuming copious amounts of data. Consume bad data and the output will be wrong. In computer science this is called garbage in / garbage out. In some AI models once a bad piece of information is learned it becomes exceedingly difficult to purge it.
The integrity of AI training data becomes extremely important for health care AI. Medical science is constantly changing. The Women’s Health Initiative released a report in 2002 that claimed hormone replacement therapy (HRT) increased cancer rates in women. HRT fell by 80% almost overnight. Women willing to take their chances could not find doctors willing to prescribe HRT, even if they were not at high risk for hormone-sensitive cancers. Nearly two-dozen years later the original data was re-examined, and the study was reinterpreted to say HRT is safe for most women. Yet, it may take years for doctors to change their thinking and AI chatbots may be no different.
Despite all the barriers, medical AI has potential. The following is from The Conversation:
Imagine walking into your doctor’s office feeling sick – and rather than flipping through pages of your medical history or running tests that take days, your doctor instantly pulls together data from your health records, genetic profile and wearable devices to help decipher what’s wrong.
Sounds great! Now allow me to take this scenario a step or two farther. Maybe you do not even see a doctor. Maybe you walk into CVS pharmacy, pay a nominal fee to use one of their MedTech cubicles and plug your health card into the kiosk. Maybe a nurse practitioner (NP) is there to assist. Maybe nobody is. Decision support tools are already available for doctors and NPs. This scenario sounds great, but it is not yet ready for prime time, according to Professor Turgay Ayer, with the Georgia Institute of Technology:
But for all its power, AI can make mistakes. Although these systems are trained on data from real patients, they can struggle when encountering something unusual, or when data doesn’t perfectly match the patient in front of them.
As a result, AI doesn’t always give an accurate diagnosis. This problem is called algorithmic drift – when AI systems perform well in controlled settings but lose accuracy in real-world situations.
Then there are privacy concerns. AI needs to churn through not just medical literature, but billions of pieces of real-world patient data to be effective. HIPAA can make that difficult. There is also the expense of training clinical staff to use AI and the cost of integration.
Finally, developing an AI system that works well involves a lot of trial and error. AI systems must go through rigorous testing to make certain they’re safe and effective. This takes years, and even after a system is approved, adjustments may be needed as it encounters new types of data and real-world situations.
AI has the potential to make health care more efficient, cheaper and increase access to care. Even when information integrity barriers are overcome, there is economics. I am talking about how the owners of the algorithms will want their piece of the health care pie. Instead of waltzing into CVS, paying $20 bucks and walking out with a prescription, you may have to make an appointment at a hospital, wait in line while they verify your insurance coverage. Then you may have to pay a $100 copay in order for an actual doctor to plug in your health card to the AI interface. AI may change how care is diagnosed. It will not change how care is funded and the politics around who is the gatekeeper. That will require other reforms.
The Conversation: AI in health care could save lives and money − but change won’t happen overnight