Tracking expert predictions on the effects of artificial intelligence on sciencescience

"Medicine shows the most rapid uptake, fueled by the pervasive use of AI in medical imaging, diagnostic algorithms, and clinical decision support systems. The scale of healthcare data, regulatory momentum, and demand for translational applications all reinforce this trajectory, positioning medicine as the leading non-CS discipline for explicit AI engagement by 2030. In physics, the diffusion of AI is proceeding steadily but at a more measured pace, with growth concentrated in data-intensive areas such as astrophysics, particle physics, and large-scale simulations. These subfields generate massive experimental datasets that benefit from AI methods, but much of the broader discipline remains oriented around theory-driven inquiry, moderating the overall rate of adoption."

"Historically, human labor patterns have experienced quite radical transformations over time, even in established sectors. Emerging technologies, rather than sucking people out of the labor market of white-collar work, are more likely to make them work differently and lead to new white-collar roles that can capitalize on this transition."

"A paradox exists for the creating-music and writing-novels thresholds: by definition the exceptional must be rare, but if an AI could write one novel as good as the best in 2025, it could likely write thousands, rendering all of them unexceptional and thus devoid of personality."

"Some of the [Millennium Prize] problems, like the Riemann Hypothesis or the Birch and Swinnerton-Dyer Conjecture, are especially well-suited to AI-supported exploration. They bear a kind of family resemblance to the Four-Color Theorem in their relationship to computer-assisted mathematics. The Four-Color Theorem was famously solved through a hybrid of human conceptual framing and extensive computer verification. As Donald MacKenzie details in his socio-history of that episode, much of the intellectual labor wasn't in the computation itself but in formalizing the problem in a way that machines could meaningfully engage with it and in managing the institutional consequences of proof-by-machine."

"Challenges in reasoning and inter-system collaboration are matters of scaffolding and context and I expect both—especially collaboration—to become much more robust in the coming years. Embodiment will advance as robotics grows and simulated world models improve, driven by greater availability of training data. In contrast, challenges such as memory, hallucination, and generalization are more deeply tied to underlying architectural limitations, making them persistently difficult to overcome."

"While adoption rates among knowledge workers may continue to grow rapidly, a substantial portion of the U.S. workforce performs tasks that are inherently unsuitable for generative AI assistance in its current or foreseeable form. Manual labor sectors including construction, manufacturing assembly, and maintenance require physical presence and manipulation of objects. Direct service work such as food service, retail, healthcare aides, and childcare depends on human interaction and physical tasks. Transportation and logistics, skilled trades like plumbing and electrical work, agriculture, and direct patient care in healthcare all involve work that generative AI cannot meaningfully assist."

"I would expect an increasing proportion of adults, even if not deliberately seeking AI interaction for companionship, drifting into the habit of conversations with AI systems and allowing themselves to suspend disbelief and conceptualise these conversations as interactions with human companions. This proportion will increase rapidly as children grow to adulthood accepting these conversations as natural, having known interactive systems like Siri and Alexa for all of their conscious lives."

"Access to clean, unbiased, legally usable, and domain-specific datasets remains the single greatest obstacle to scalable adoption. Despite larger models and better architectures, progress will continue to be throttled by privacy constraints, copyright uncertainty, and the scarcity of trustworthy labeled data for specialized sectors such as healthcare, defense, and finance."