Humanity faces more questions than scientific capacity
For most of modern science, the limiting factor has not been imagination. Scientists have never lacked for questions. They have lacked some combination of funding, bandwidth, quality of data, and the freedom to pursue ideas that do not fit neatly into a grant cycle or product roadmap.
That leaves an enormous amount of important work waiting across the spectrum of scientific disciplines. Some questions are neglected because they fall outside fashionable areas of funding: too rare, too early, too interdisciplinary, too uncertain, or too far from an obvious market. Rare disease communities know this well. Others sit inside active fields, but remain bottlenecked by complexity. Psychiatric genetics has produced extraordinary maps of risk, but far fewer mechanistic answers. Aging, materials science, agriculture, climate, neuroscience, and many other domains now generate more public evidence than any individual lab can reasonably absorb. The bottleneck is not always a lack of data or ideas. Often, it is the cost of turning scattered evidence into a tractable research direction.
AI changes that equation by enabling a more efficient division of labor. In just the last few years, AI systems have moved from useful assistants for search and drafting toward agents that can carry more of the research loop itself. They can help synthesize literature, integrate public evidence, and run in silico experiments. That is not the whole of science, and it is not necessarily where human judgment matters most. But it is often the work that consumes the most time and distracts scientists from the harder questions of interpretation and judging what it would really take to convince us or change our minds. Used correctly, AI like Marvin aren’t meant to remove people from the loop but rather to make scientists more powerful by moving their attention back to the parts of discovery that require creativity, skepticism, and judgment.
From access to participation
This opportunity only matters if the capability spreads. If AI for science becomes another advantage reserved for the best-funded labs, largest companies, and narrowest set of commercial priorities, there is no doubt it will still help to produce extraordinary work. But it will also accelerate the already-accelerated and further widen the inequity between hot research topics and the underexplored. The better timeline in our view is one where AI is used to help more people turn neglected ideas into tractable research projects and produce actionable discoveries.
The trajectory of the internet offers a great analogy (not because every AI-adjacent blog post announcement needs an “AI is the new internet” comparison), but because the lesson here really is meaningful. The internet did not matter only because it made information easier to access. It mattered because it also made participation easier. People could publish, organize, build in public, and contribute to shared bodies of work. Open-source software, Wikipedia, and social media all grew out of that shift.
We think AI should do the same for science. That is why we are launching the Iluvatar Open Research Initiative (IORI), an open research program powered by Marvin, our autonomous AI research agent.
Open research should evolve with each loop
We do not think the right messaging for autonomous AI in science is to ask people to trust the machine more. Trust has to be earned and that’s especially important when the foundation of research is reproducibility and rigor. All IORI projects are fully transparent and we will publish the work in cycles as it develops: the questions, the evidence, the methods, and open questions to stimulate discussion and follow-up research. IORI projects are also meant to be living research programs, evolving in response to new evidence and analyses. Anyone can propose a project, review or challenge interpretations, and contribute additional literature or resources. We especially encourage scientists who work or run real-world laboratories to build off of IORI projects to generate new data and validate computational findings. All IORI publications, code, and analyses are completely open source under Apache-2.0 / CC-BY-4.0 and all contributors retain all of their rights and IP.
Kicking off IORI by targeting sarcopenia and schizophrenia
We are starting with two projects that affect millions of people and represent huge unmet medical needs. The first focuses on skeletal muscle aging: why aging muscle loses its capacity to regenerate and shifts toward inflammatory, fibrotic, and degenerative decline. Sarcopenia still has no approved drug treatment, despite important work across muscle biology, inflammation, fibrosis, and single-cell atlases.1 As a result of this foundational work, it is exactly the kind of problem where AI-driven computational integration can help turn scattered evidence into sharper, testable hypotheses.
The second focuses on schizophrenia risk. After two decades of genome-wide association studies, the field has identified hundreds of risk loci, but translating those associations into mechanisms remains one of the central bottlenecks in psychiatric genetics.2 The relevant evidence spans genetics, cell-type regulation, neurodevelopment, and drug-target biology. It’s a hard problem that, like sarcopenia, is well suited for iterative autonomous research to align and integrate fragmented, high-dimensional omics and clinical data.
We want to note that while both of these initial projects are biomedical in nature, we view them as starting points, not boundaries. Marvin’s autonomous abilities are multi-disciplinary, and we envision IORI as an initiative that can span the spectrum of high-value, challenging problems facing humanity.
How to take part
IORI’s success as an open source science effort relies on the contributions of everyone. Whether you’re a KOL in a relevant field or citizen scientist or just someone who is passionate about a problem, please feel welcome to participate. As we said earlier, we launched IORI in part because we believe AI should expand the surface area of science, not concentrate research in fewer places, so we encourage you to get involved and share IORI projects no matter who you are. If you or someone you know has a problem that you think would make a great IORI project, send it our way.