UAE’s medical AI ‘will not replace doctors’

Reuters/Cover Media Pictures

• Potential but ‘not there yet’
• Patient-doctor interaction still needed
• AI cannot say ‘I don’t know’

The UAE’s unveiling of a homegrown clinical artificial intelligence program is a breakthrough in healthcare technology say industry experts, but AI will not yet replace doctors talking to patients.

M42, an offshoot of G42 Healthcare and Mubadala Health and based in Abu Dhabi, announced last month that it had built a clinical Large Language Model (LLM) designed to provide answers to medical inquiries.

Med42, developed with 70 billion parameters – or different values that control the behaviour of a machine learning model – is designed to streamline clinical decisions and offer consolidated medical data to healthcare professionals, researchers, regulators and patients.

• TruDoc’s new owner seeks to expand virtual health provider
• Gulf accelerates longevity push with AI and healthtech
• AI education takes centre stage in the UAE

The model is geared towards developing tailored treatment plans based on patient history.

“There’s massive potential for use in clinical decision making, but we’re not quite there yet,” Dr Sameer Mankani, general manager at Health at Hand in Dubai, told AGBI.

“LLMs are not yet legally approved medical devices.”

Integrating LLMs into patient care presents challenges because of the technology’s inability to maintain context over multiple interactions, Mankani said.

The technology also has a tendency to “create information” when lacking data, a phenomenon known in the industry as “hallucinations”.

“LLMs don’t know how to say, ‘I don’t know’,” Mankani said.

“If I ask a question on something very current and there’s only one article or one opinion piece out on that, then [for the AI] that becomes the truth. When it comes to medicine, that could be disastrous.”

Dr Pankaj Shrivastav, director of Conceive Fertility Hospital in Dubai, said the legal implications of incorrect diagnosis as a result of the technology were unclear.

“AI will definitely have a big role to play in the future,” he said. “But you cannot diagnose purely on the basis of AI, which essentially is just taking into consideration lab results and scans and so on.

“In the work that we do, a lot of factors come up as a result of patient-doctor interaction.”

Emotional support

Shrivastav said that AI also cannot replace counselling and emotional support. “So many surrounding issues come up, such as marital disharmony, or financial issues,” he said.

Despite these challenges, Mankani believes LLMs offer considerable operational benefits, such as improved scheduling and translation support, ultimately leading to cost savings. 

Mankani also sees value in AI models in supporting clinical decisions by providing direct answers from the different sources that doctors use to look up diagnoses, new treatments and dosages. “It takes doctors a lot of time to comb through new articles, journals and studies,” he said. 

“LLMs are really good at summarising an incredibly large amount of data. The potential and value for that is really, I think, huge.”