san francisco Medicine is facing a fundamental upheaval. Artificial intelligence (AI) will improve quality, facilitate diagnoses and relieve medical staff of paperwork. At least Peter Lee is convinced of that. The computer scientist heads Microsoft’s research department and focuses on large language models, the technology behind programs like ChatGPT. “It’s the most important thing in my career,” Lee said.
The models could not only analyze texts or create business models, but also help in the diagnosis of rare diseases. “They weren’t even specially trained for it,” Lee said in an interview with the Handelsblatt. Lee has advanced computer science for the past few decades and has been at Carnegie Mellon University for over 20 years.
Now he and his team at Microsoft are testing the use of large language models. Because Microsoft is a close partner and sponsor of OpenAI, the company behind ChatGPT, Lee and his team have access that other researchers don’t have.
OpenAI’s most advanced language model, GPT-4, passed the US medical licensing exam with flying colors, better than 90 percent of trained physicians. “It was shocking to see how well the models did,” Lee said. The systems are already able to help with documentation, for example. Currently, the collection of data, diagnostics and other information would tie up many hours of work. That could change in the future – thanks to AI.
In a pilot project, Microsoft is working with the medical technology provider Epic in the USA. Many clinics in the United States use Epic’s software to manage patient records. The company, in cooperation with Microsoft, has developed a model that creates suggestions for standard messages. This is being tried at clinics in San Diego, Palo Alto and Wisconsin.
Develop medicines with AI
The chip specialist Nvidia has also developed a language model optimized for medical purposes. It’s not designed to understand human language, but to analyze systems in biology or chemistry to develop drugs faster, says Kimberly Powell, who heads Nvidia’s medical division. The model, called BioNeMo, is optimized, among other things, to recognize and process protein structures.
>> Read here: What is artificial intelligence?
The search engine operator Google has also entered the race for the best solutions for medicine. The company has presented a language model called Med-PaLM that has been optimized for specialist medical questions. For example, it can evaluate medical studies, but also read patient files.
Traditional medical technology companies like Siemens Healthineers are following suit. Siemens Healthineers, for example, states that it is already experimenting with large language models. AI-based decision support for doctors is already being used in the company’s products in a wide variety of forms: not only in imaging diagnostics for diagnosis, but also with a view to the patient’s entire treatment path.
Humans, not AI, have the last word in medicine
Application errors in the laboratory can now also be detected thanks to self-learning algorithms. “Generative AI will have an impact on healthcare,” says Peter Schardt, chief technology officer at Siemens Healthineers. The technology has the ability to evaluate large amounts of text data and increasingly also images, videos or other data – such as new scientific studies.
>> Read here: How one of the largest clinics in Germany uses ChatGPT
“Such services are extremely relevant for a significant increase in efficiency in healthcare,” says Schardt. In the area of error detection and quality assurance, it is quite possible that in the near future AI programs will be able to routinely point out errors and risks or deviations from quality specifications to doctors.
“Generative AI has great potential to fundamentally transform the future of healthcare. It is crucial that it does not hallucinate, does not suppress any relevant facts, remains controllable and is trustworthy,” says Schardt. In his view, the future lies in the combination of AI-based services and human control. “But the last word will always be with people – not with AI,” says the Healthineers chief technology officer.
Professor Johannes Eichstädt warns against too much optimism. The AI expert from the Institute for Human-Centered Artificial Intelligence at Stanford University emphasizes that it will be years before these technologies can really be used widely in medicine. “A lot of studies are still needed,” says Eichstädt.
More: How AI is revolutionizing the biotech industry