Would you trust your health to an algorithm? The technology is not there yet, but the tech giants are working on it. Google, in particular, hopes to gain a foothold in hospitals by applying the new technology: generative artificial intelligence. The goal: to help healthcare professionals, but also to stay ahead of their competitors… and avoid the many ethical and legal risks along the way.
Among the company’s strengths: Med-PaLM-2, a generative AI chatbot specifically trained on health issues. The algorithm, introduced by Google in March, can generate answers to questions, but also summarize documents and organize patient data. The objective: to obtain more reliable results than with generalist chatbots such as ChatGPT. Med-PaLM-2 would have obtained 85% of the correct answers to an American medical exam, according to Google.
And the software has already entered the full-scale testing phase. The Mayo Clinic, a network of the best US hospitals in the world, has been testing Med-PaLM-2 on its wards since April, according to sources cited by the Wall Street Journal. One of the objectives: to facilitate the exploration of patient data, through the creation of a dedicated search engine reserved for the hospital. Google has yet to announce a broader release in the future.
Evacuate administrative tasks
Should we expect a revolution? After all, medicine has been using AI and machine learning for years. Analyzing X-rays, finding the starting point of cancers of unknown origin… Applications did not wait for generative AI to become more and more numerous.
But generative AI like Google’s brings new possibilities. In particular on apparently innocuous uses: to streamline administrative tasks. Automatically transcribe the conversation between a doctor and his patient, summarize the important points and enter them in the patient’s file… A process that could take doctors several hours can be solved in a few minutes thanks to the new software improved with “big language” . “, thus leaving more time for human consultation and analysis.
Patients can also find their account directly there. Responses from chatbots like Med-PaLM-2 and ChatGPT are overwhelmingly rated as more empathic than responses from physicians, according to two studies published by Google and University of San Diego researchers, respectively. By summarizing consultations in a synthetic and clear manner, language models could also limit oversights and further integrate the patient into their care pathway.
Democratize access to medical information
A promising market that attracts greed. The sector is already invested in by many startups, particularly in terms of support for administrative tasks. And, of course, other tech giants are counting on AI to make their mark. In particular, Microsoft, through its partnership with OpenAI, the company behind ChatGPT.
Its latest version, GPT-4, would also have gotten more than 80% correct answers on a US medical exam, according to a study. Microsoft has also acquired Nuance, a company specialized in the field, for almost 20,000 million dollars, and will infuse it with the technologies developed by OpenAI, always to help doctors in the most mechanical tasks.
For Google, this is just the beginning. “AI could democratize access to expert medical information around the world,” Zoubin Ghahramani, vice president of Google DeepMind, one of the group’s main AI research units, told Tech&Co. Especially in countries that lack doctors, according to an internal email cited by the Wall Street Journal. And why not, eventually, provide a diagnosis instead of the human one.
An extremely sensitive sector
But most AI specialists and companies agree that algorithms are still a long way from replacing human professionals. Healthcare is a highly regulated and high-risk sector that cannot accommodate chatbots that generate incorrect answers, at the risk of endangering patients.
While Med-PaLM-2 is rated more empathetic, it still includes more incorrect or unnecessary information than human doctors, according to the same Google study. And many past examples remind us that algorithms can give discriminatory results: In 2019, a study already revealed that an algorithm tasked with assessing the urgency of care in American hospitals consistently underestimated the needs of African-American patients compared to those of African-American patients. whites.
For this reason, both Google and Microsoft insist on a measured deployment of these technologies. “We want to be very careful and responsible in the way we use such a powerful tool as generative AI, especially in healthcare,” insists Aashima Gupta, global director of healthcare strategy at Google Cloud, to CNBC.
The security of personal data is also a crucial issue, particularly for health data. Google ensures that the data sent to Med-PaLM-2 as part of the tests is encrypted and that it is not saved or used to train the model, according to the Wall Street Journal.
Source: BFM TV
