Ever wondered what athlete’s foot medication has to do with depression? Ask ChatGPT! The answer is completely correct and even understandable. So do we finally have the omniscient machine to solve our problems? Not really.
dr André Sander received his doctorate from the Charité in the field of medical sciences and has been working with medical terminologies and ontologies for more than 25 years. At ID Information and Documentation, he is CTO, authorized signatory and member of the management board. His main focus is in the areas of medical terminologies and ontologies, semantic analyzes of medical free texts, natural language processing.
First, the reassuring news for the doctor readers: Don’t worry, ChatGPT doesn’t know any more than you know. Strictly speaking, ChatGPT does not provide any answers either, but rather summaries of what we (people) know and have written down – but quite well. So good that millions of people around the world ask questions and read the answers with fascination. Even so many people that the public system on the OpenAI website is currently no longer accessible.
medicine with the latest hype
Blockchain and NFTs were somehow not really applicable to medicine, so it is understandable that the euphoria about the results of ChatGPT is so great that there is already speculation as to whether the “digital native” doctor ChatGPT will be in the future surveyed, so more patients can be treated better in less time. But it probably won’t be that easy (again). The accuracy of the results stands or falls with the evidence of the knowledge, not to say with the number of experts who have built up the knowledge. And if the knowledge is not yet available, ChatGTP cannot summarize anything either. After decades of research, it is now known that amyloid beta plaques are most likely responsible for Alzheimer’s. A treatment, however, is still completely unknown. Ask ChatGPT about it…
For medicine, one of the challenges will also be to integrate the “digital native” patient into such new forms of knowledge representation. In contrast to the specialist staff, they are rarely or hardly able to assess the results correctly. No matter what ChatGPT presents as a result: It always sounds damn smart and right. Even GPT-3, i.e. the underlying language model, was indistinguishable from humans and scientists have increasingly advocated petitions against its use. With ChatGPT, the interactive, the playful has found its way into the application. You can just “chat” with a really smart machine.
The makers have announced that the next generation will know even more. Here, too, skepticism is appropriate: the next generation will know even more. More sources and more data do not necessarily guarantee better results. Will the next generation know the difference between epigenetics and genetics? Between pharmacy and alchemy? Will ChatGPT get better if medieval medicine books are included?
Doctors do not abdicate responsibility
Interestingly, politicians are currently dealing with drafts on the right not to be treated by AI algorithms. Of course, this generally means machine learning (all three forms of artificial intelligence are only correctly addressed at EU level). How absurd (in terms of content) this demand is can best be illustrated by imagining an internet ban in medical practices. My family doctor “googles” from time to time for side effects of medication and I’m actually quite happy about it, because of course he doesn’t know all the side effects of all medication. He does not give up responsibility for the interpretation of the results of his research.
And the provenance of the facts is also an essential factor that is taken into account (in the end it is the specialist information from the manufacturers that is read and not just any advice blog). Neither will change with systems like ChatGPT or other supporting systems. Ironically, there are already established systems in the field of medicines that do exactly what my family doctor researches manually every time: the calculation of drug and therapy safety (AMTS). Established, however, only in hospitals and not in medical practices. This shows once again where we have a much more urgent need for revolution: in the structures of medical care itself!
Such AMTS systems are certified medical products and mostly belong to the area of symbolic AI, whose great strength is the referentiality or the traceability of the results. But access to knowledge will once again be revolutionized by ChatGPT – and that will also change medicine. Since the internet and intelligent search engines, hardly anyone knows how to do research in libraries and books. Maybe we will soon say: Since ChatGPT nobody knows what a research is. So it is the personal acquisition of knowledge that will benefit enormously.
Facts alone don’t help
But be careful: the entire knowledge structure – the generation of knowledge – i.e. the central hope in all the AI algorithms, is not addressed here at all. It would be naïve to think that we just need to put enough facts into an algorithm to make it intelligent and, more importantly, creative. ChatGPT’s creativity is our own creativity, our behavior that we have taught an algorithm. The danger with ChatGPT could also be that knowledge building is even inhibited if ChatGPT learns from published texts that were generated with the system itself. In addition, much information that can be used to build up knowledge is of a personal nature. If machine learning algorithms are used, this has significant consequences for data protection and copyright.
No new unanswered replies
In medicine, revolutions always take a little longer – ChatGPT is a promising tool that once again impressively demonstrates the possibilities of digitization. But we need revolutions elsewhere: Why do we still not have a nationwide electronic health card in Germany? Why do we use a so-called telematics infrastructure whose technological basis dates back to the last millennium? ChatGPT has no answer to that either.
To answer the question asked at the beginning: Will ChatGPT revolutionize medicine? Perhaps. If we understand ChatGPT as a tool and use it where it can actually solve problems.
(mack)
