The Future of AI and Health, Part I: Forecasts Reveals Little Consensus

David Shaywitz
The likely impact of AI on health and medicine is … highly dependent on who you ask.
Consider the spectacularly wide range of opinions offered in just the last several weeks.
A $500 billion AI project, dubbed “Stargate,” was announced with great fanfare on January 21 in the Roosevelt room of the White House by President Trump. He was flanked by Masayoshi Son, CEO of SoftBank; Larry Ellison, chairman of Oracle; and Sam Altman, CEO of OpenAI.
Stargate, according to OpenAI, is “a new company which intends to invest $500 billion over the next four years building new AI infrastructure for OpenAI in the United States.” The “key initial technology partners” are Arm, Microsoft, NVIDIA, Oracle, and OpenAI.
Altman touted the promise of AI in healthcare.
“We will see diseases get cured at an unprecedented rate,” he said. “We will be amazed at how quickly we’re curing this cancer and that one and heart disease. And what this will do for the ability… to cure the diseases at a rapid, rapid rate, I think, will be among the most important things this technology does.”
Ellison highlighted the possibility of an mRNA cancer vaccine.
“You can do early cancer detection with a blood test,” he explained, “and using AI to look at the blood test, you can find the cancers that are actually seriously threatening the person. You can make that vaccine, that mRNA vaccine, you can make that robotically, again using AI, in about 48 hours.”
Oncologist and gadfly Vinay Prasad, for one, wasn’t buying this. “AI will do a lot of good things. But it won’t cure cancer,” he said on X. One problem, he argues, is that AI’s ability to inhale all published literature won’t help it sort through the vast amount of fraudulent or irreproducible work.
Another problem he perceptively highlights is “biology itself” – that is, “understanding what is happening in the cell that has not been seen or detected in current experiments. AI has no way to do this unless it again picks up a pipette.”

Derek Lowe
Grizzled pharma veteran Derek Lowe is similarly skeptical. Writing in Chemistry World, he argues that even if AI can help at the earliest stages of the drug discovery process, it “gives you a speedup in the earliest and least expensive part of the whole process, one that is often as not the shortest as well. Unfortunately, that’s not quite the material of a revolution.”
He continues,
“Compounds found (wholly or partly) by such AI methods are still going to be subject to the same white-knuckle dice-rolling as all the others when they get into human trials, because we have (as yet) no computational tools that really help us predict whether we have picked the right target, the right disease, the right biochemical pathway, or the right compound to affect it without doing anything unexpected along the way. When AI systems start to help with those questions, the revolution may really be at hand. But as it stands, some of the current press releases sound like someone trying to sell a new car model by pointing out that its windows roll up and down much more quickly than the competition.”
Lowe’s hesitations are shared by Andreas Bender, an AI expert with big pharma experience who has since transitioned to academia. As I discussed in TR in October 2022, Bender worries, essentially, that AI is solving stylized, reduced problems, which lead to publishable papers but may not be especially applicable to the challenges of real-world R&D.
Another experienced drug developer, Regeneron’s legendary chief scientific officer, George Yancopolous (a favorite of this column – see this April 2021 discussion) also thinks the promise of AI has been wildly overblown.

George Yancopoulos, president and chief scientist, Regeneron Pharmaceuticals
“There’s no miracles coming from stem cells and AI,” Yancopoulos told Andrew Dunn of Endpoints News.
Yancopoulus offered a visceral response to the term itself, Dunn reports:
“I just get a reaction to even the name ‘AI.’ There is no ‘I’ in AI,” Yancopoulos said last week in an interview at the JP Morgan Healthcare Conference. “It’s all about machine learning, even this generative AI stuff. We use machine learning and these approaches as much as anybody, but we understand what it can give you and what it can’t give you.
“The hype in these stages is just greatly exaggerated compared to the reality,” he added.
Yancopoulus, Dunn says, called the awarding of the 2024 Nobel Prize in Chemistry to the developers of Google DeepMind’s developers of AlphaFold “the stupidest thing I’ve ever heard.” (The prize was shared with University of Washington biochemist David Baker, for his work on computational protein design.)
Yancopoulus says he appreciates the pattern recognition capabilities of AI/ML and hopes to turn these technologies loose on ever-larger integrated datasets comprising genetics, medical records, and proteomics, as they are doing now in partnership with the UK BioBank, and aspire to do the same in the United States.
“Our dream is to then get to 50 million [integrated data sets] and then to the entire US population,” he said. “We’re very willing to work with whoever wants to help on this, because I think this is an incredible resource.”
Historically, the issue hasn’t been so much potential scientific value of such a rich and integrated dataset, but rather the challenge of assembling it, particularly given privacy concerns and the highly siloed nature of medical data, data that patients nominally have the right to but often struggle to access easily.
The tendency to gear EHR data to maximize billing may also get in the way. The rigorous and thorough protocols that patients (at least ideally) go through to enroll in a clinical study can differ significantly from the way patient conditions are documented in routine clinical practice – a discrepancy that healthcare leaders like Amy Abernethy, in particular, are striving to eliminate.

Subha Madhavan
Yet another perspective on AI in pharma can be found (like Bender’s pivotal work) in Drug Discovery Today (here), co-authored by Pfizer’s brilliant data scientist Subha Madhavan and me. We offer a cautiously optimistic take and focus on the application of AI to clinical development.
We argue that “the most vexing challenge facing drug developers – as well as our most significant opportunity – is managing the exploding complexity at every stage of the drug development process,” and explain that “these hurdles are the direct result of biomedicine’s remarkable success and accelerating progress.”
AI, we suggest, can help drug developers manage this burgeoning complexity, and we offer a range of early, promising examples from manufacturing to document management to trial design.
We also emphasize the changes in “ways of working” that are likely to be headed our way, highlighting the need to reinvent workflows (as discussed in TR here in June 2023). We envision approaches to drug development that are “less siloed and more collaborative,” involving “shared comprehensive and timely data (enabled by AI), as well as common analytics and visualization dashboards.” (The value of such dashboard during the development of COVID vaccines was discussed in TR here in May 2022.)
Read Part II: The Future of AI and Health, Part II: Andreessen and Colleagues Weigh In