“Can We Really Cure Every Disease?”

This is Optimizer, a weekly newsletter from a senior reviewer that dives into the latest gadgets and innovations that claim to revolutionize your life. This week’s edition is a special early release linked to Google I/O coverage. Look out for our next issue at the usual time next Friday.

During this year’s Google I/O keynote, Google DeepMind CEO Demis Hassabis made a bold statement, claiming the company aims to “reimagine the drug discovery process with the goal of one day solving all disease.”

This kind of declaration is what the phrase “big, if true” was made for. What Hassabis was really referring to was Gemini for Science, a suite of experimental AI tools designed to inspire researchers to make new discoveries.

While I often critique AI in healthcare, Hassabis’ statement warrants deeper context. Effective science communication—making complex ideas accessible without spreading misinformation—has become increasingly challenging. Researchers in the audience likely understood his claim as a sign that AI is significantly speeding up medical discoveries. However, for the average person, it might have sounded like “Gemini is going to cure every disease because AI is that powerful.” This oversimplification doesn’t reflect how medical breakthroughs actually occur.

For decades, AI has been a crucial player in medical research. The algorithms powering wearables? That’s AI at work. Innovations in noninvasive detection features? Thanks to machine learning. Generative AI is a newer player in this field, but it shows immense potential. In my conversations with clinical researchers, it’s clear that many advancements in consumer health tech stem from AI progress. For instance, a meta-review highlighted how AI significantly shortened the development timeline for COVID-19 vaccines, benefiting the global community. However, the review also pointed out that serious ethical, logistical, and regulatory hurdles remain, particularly concerning algorithmic bias, data privacy, and equitable access.

In the keynote, Hassabis highlighted Google’s AlphaFold and AlphaGenome projects. AlphaFold aids researchers in understanding protein structures, which is vital since proteins are involved in countless biological processes. Gaining insights into proteins—or even creating new synthetic ones—could be key to unlocking cancer treatments. Recently, scientists identified 1,700 new proteins that might contribute to this goal. Traditionally, discovering new proteins and understanding their interactions was a lengthy process, but AlphaFold can significantly accelerate that timeline. Real-world applications include developing malaria vaccines and uncovering proteins linked to LDL cholesterol and early-onset Parkinson’s disease.

On the other hand, AlphaGenome helps researchers predict mutations in human DNA sequences. This model has the potential to shed light on why certain diseases occur, but Google has acknowledged its limitations. For example, it hasn’t been validated for personal genome predictions and struggles with capturing specific patterns in cells and tissues. These nuances are crucial for researchers but may be overlooked by the general public.

In many ways, Hassabis’ comments weren’t aimed at the average person. It’s also important to note that these AI models and Gemini for Science tools won’t magically eliminate cancer or other “unsolvable” diseases in the next few years. Realistically, we’re looking at a timeline of at least 20 years, if not longer. While that may seem like a long wait—especially for those with loved ones facing illness—it’s an ambitious estimate in the realm of rigorous scientific research.

However, a keynote isn’t the place for lengthy explanations, especially when announcing a multitude of AI features. The issue is that such statements can spread quickly and have far-reaching effects. For most of us, AI in healthcare has been a frustrating experience filled with vague summaries, inaccuracies, and excessive hand-holding. It’s easy to conflate AI tools for researchers with consumer AI health features, even though they serve different purposes.

My immediate reaction to Hassabis’ comment was to recall a recent statement from a health official who suggested that AI could render the FDA “irrelevant.” His reasoning was that AI could streamline drug development and approval. When compared to Hassabis’ comment—made in a different context—it’s easy to see how the average person might draw misleading connections, such as assuming Google supports that perspective.

It’s worth noting that previous discussions have highlighted flaws in that official’s views on AI in healthcare. In an interview, he claimed that AI could expedite the drug approval process. While there’s some truth to that—AI tools have been utilized in this field for a long time—newer models won’t eliminate the need for FDA trials, animal testing, or established processes. AI is ultimately a tool that requires expert input and collaboration, and scientific rigor cannot be bypassed.

Context is crucial, yet it often gets lost in catchy soundbites. This is why I previously outlined the tactics of wellness grifters, where broad facts are juxtaposed with misleading claims. I’m not suggesting that Hassabis made a grave error in his keynote remarks. Google (and others) engage in substantial clinical research and strive to communicate their efforts effectively. However, in today’s fast-paced media landscape, much can be misinterpreted or oversimplified.

Sciencewashing is rampant today, where buzzwords and bold claims create an illusion of legitimacy that overlooks important details. In Silicon Valley, this is evident in tech enthusiasts who embrace extreme biohacking trends. It’s not a huge leap from “AI can solve all diseases” to “track your biometrics, optimize with these supplements, and conquer mortality.”

Perhaps one day, AI will help tackle all diseases. But if that happens, the journey will be complex and far from straightforward. The next 20 years will bring many changes, especially in the political, social, and cultural landscapes that will also influence clinical research capabilities—so forgive me if I’m not as optimistic as Hassabis at this moment.