SciFM 2026 at UChicago: Inside the Premier Gathering of AI, Foundation Models, and the Future of Scientific Discovery

Walking onto the University of Chicago campus this May, visitors to SciFM 2026 could sense the electric anticipation. As the third installment of the Scientific Foundation Models conference series unfolded, world-renowned researchers, national lab experts, and industry leaders collided in a three-day conversation that asked: How is AI reshaping the practice of science—and what still stands in its way?

Why SciFM Matters

SciFM, short for Scientific Foundation Models, has rapidly become the foremost venue for discussing foundation models and AI agents in scientific discovery and engineering innovation. What began in Ann Arbor in 2024 as a probing look at scientific foundation models has evolved, traversing topics from mathematical reasoning to national-scale facility automation. This year the conference drew 270 registrants, including faculty from 26 universities and students from 13 universities, reflecting an exceptional breadth of academic participation. UChicago’s role as host brought together the region’s deep expertise in computation, physics, and policy, inviting new voices from both the Midwest and national sphere into the debate.

The Crucial Questions

The 2026 conference was structured around three questions that permeated every session:

• Can scientific foundation models truly be grounded in physical laws—and what does “grounded” mean?
• What unsolved problems in domains like aerospace, energy, and manufacturing remain untouched by today’s AI? Why?
• What does scientific discovery look like when AI works alongside humans in real time, with uncertainty and oversight?

These questions aren’t just academic exercises; they probe the realities facing industries from biotech to climate science. For example, a pharmaceutical startup may deploy AI to predict protein structures, but still fail in wet lab tests, highlighting the difference between computational promise and practical results. In manufacturing, robots that recognize patterns may flounder when confronted by unpredictable physics or safety constraints.

A Program Designed for Debate

The conference was packed with keynotes and panels that fostered “productive tension” between optimistic claims and honest assessments, as well as the recurring challenge: What isn’t AI solving, and what must scientific foundation models address?

SciFM 2026 featured speakers from five overseas countries: England, Germany, Switzerland, Japan, and Korea. Attendees included representatives from 13 national labs, as well as the Army Research Lab, Air Force Research Laboratory, and the NSF-Simons National Institute for Theory and Mathematics in Biology, bringing government and institutional perspectives into dialogue with academia.

Industry’s presence was equally notable–41 private companies sent participants, affirming the conference’s relevance to sectors ranging from technology and pharmaceuticals to energy and advanced manufacturing.

Day One opened with discussions about how AI can evolve from passive pattern recognition to active, closed-loop engagement with the physical world. Panels explored the hurdles in industrial transformation, complexities of biotech workflows, and the daunting task of creating embodied intelligent systems: robots and autonomous instruments that must navigate not just the world, but the ever-shifting landscape of scientific experimentation.

Day Two dove into “physical intelligence,” with keynotes and panels focused on the merits of learning versus encoding physical laws. Infrastructure and hardware took center stage as the conversation shifted to the silicon requirements for scientific AI, and how hardware validation cycles can integrate chip design into rigorous experimental science.

Day Three tackled autonomy, scale, and future roadmaps, examining how AI can close the loop on real-time systems, from climate modeling and beamline science to managing national energy grids. The final plenary, “The Road Ahead,” brought together panelists from multiple sectors to reflect on consensus, unresolved tensions, and the direction for future efforts.

Academic Discussion and Societal Impact

At a societal level, SciFM 2026’s discussions touched on fundamental questions: Who will benefit from AI-driven science? How can the community avoid a consolidation of power and ensure democratization of opportunities? What does progress look like in a world where advanced models and infrastructure can cost billions?

Panelists discussed new funding models and challenged traditional methods of scientific publishing, hinting at future platforms for sharing models, data, and methodologies. The need for modern benchmarks and evaluation tools was identified as a catalyst for real progress.

There was candid recognition of challenges: budget pressures, the risk of workforce downsizing as automation accelerates, and the need for frameworks that foster both innovation and equity.

Consensus, Tensions, and the Road Ahead

Throughout the event, there was a notable shift from monolithic models toward agentic systems where specialized agents collaborate, guided by human oversight. The closing panel forecasted a future where scientific discovery cycles accelerate, catalyzed not just by AI, but by community-driven benchmarks, more democratic compute allocation, and cross-disciplinary collaboration.

Yet, several cautions surfaced: rapid acceleration could bring greater inequality, more concentrated scientific progress, and societal consequences that require policy intervention and community engagement. Industry, governments, and institutions must develop new frameworks not only to fund and sustain infrastructure, but to ensure equitable progress and safeguard against unchecked concentration of power.

Looking Forward

As SciFM 2027 was announced, the sentiment was clear: The next decade of scientific AI will demand both imagination and responsibility. It will profoundly change how knowledge is constructed, shared, and applied in fields ranging from medicine to energy. Whether that change accelerates inclusive discovery or deepens divides remains a challenge to be met with collaboration, policy, and technical innovation.

For researchers, engineers, and policymakers looking to shape the future, the SciFM series remains a vital meeting ground—a place not just for advances, but for the open-ended debates and consensus-building that real scientific progress requires.