A country’s scientific capacity may increasingly depend on how much computing power its researchers can access. As AI becomes more involved in generating hypotheses and deciding which experiments to test, institutions with greater access to advanced AI systems may be able to pursue more research questions.
On October 4, 2026, 17 countries endorsed the Kyoto Vision for a Golden Age of Science at the Science and Technology in Society Forum in Kyoto. The White House announcement confirms a coalition that includes the United States, Japan, and governments across several other regions.
The declaration treats advanced AI as an increasingly important part of scientific research infrastructure. Its implications extend beyond whether researchers use AI tools, as access to compute-intensive systems could affect which institutions and countries can pursue certain forms of research at scale.
What “Super Intelligence” means here
The Kyoto Vision uses the term “Super Intelligence,” abbreviated SI, but provides no formal definition or capability threshold. Instead, the document points to scientific capabilities such as reasoning across large bodies of knowledge, building more accurate models of physical or biological systems, and supporting “autonomous, closed-loop discovery.”
The term does not mean the signatories claim AI has already reached human-level general intelligence. Instead, the document focuses on scientific functions while leaving the meaning of “Super Intelligence” unresolved.
Closed-loop discovery connects research decisions in a feedback cycle. An AI-enabled system may generate a hypothesis, select an experiment, analyze the result, and use that evidence to revise the hypothesis or choose the next test. The degree of automation can vary, but the defining feature is that experimental results influence subsequent research decisions.
Early versions already exist in laboratories
In a Nature paper published on May 19, researchers described Robin, a multi-agent system for experimental biology. Robin combines literature research with data analysis to propose hypotheses and experiments, then revises those hypotheses after new results become available. The authors describe the system as semi-autonomous.
The team applied Robin to research on dry age-related macular degeneration, where it identified therapeutic candidates that showed activity in laboratory experiments and proposed follow-up work on a possible biological mechanism. These findings represent early experimental evidence, while clinical effectiveness would require separate investigation.
A second Nature study on perovskite solar cells connected machine-learning-based materials discovery to automated manufacturing. Feedback from fabrication and measurement informed subsequent experimental choices, while the system identified a molecule that enabled high-efficiency devices.
Unlike systems that stop at hypothesis generation, this workflow connected computational decisions to physical production and fed experimental measurements back into subsequent choices. It offers a clearer example of what closed-loop discovery can look like when AI interacts with laboratory processes.
Although both systems show that parts of scientific discovery can already be automated, their capabilities remain limited to specific research domains. They do not yet represent a general-purpose AI scientist that can independently investigate problems across different fields.
The Neuron’s coverage of Google’s AI co-scientist examined a related approach in which multiple AI agents generate and refine research hypotheses. Systems such as these are focused on specific stages of scientific work, but they illustrate how AI can take a more active role in determining what researchers investigate next.
Why computing access becomes science policy
Kyoto groups access to AI tools and scientific data with computing infrastructure and experimental facilities. One implication is that the effective value of a research grant could increasingly depend on how much model inference, simulation, or automated experimentation it can support.
Two teams with comparable expertise may produce very different amounts of experimental evidence if one can test more candidate explanations computationally and send promising options into an automated laboratory. Greater computational capacity does not guarantee better science, however, because results still depend on experimental quality and the validity of the assumptions behind each system.
Questions about access therefore become part of science policy. Governments and research institutions may need to decide who receives scarce computing time, whether smaller institutions can afford recurring access to advanced research systems, and how those costs should be funded.
The U.S. science strategy released in July had already proposed scientific foundation models and autonomous laboratories, alongside investment in datasets and AI-enabled verification. Kyoto moves several of those priorities from a U.S. policy document into a declaration endorsed by 17 governments, although each country still needs to determine how those goals would be implemented.
Seventeen endorsements still need implementation
The declaration establishes no binding funding commitments, shared infrastructure arrangements, or implementation deadlines. It also calls for experimentation with research funding and greater use of metascience, while scientific integrity and support for early-career researchers remain part of the broader agenda.
The AI-for-science agenda creates a separate challenge: governments will need evidence that faster or more automated research also produces dependable results. A system may accelerate experimentation, but reproducibility still depends on whether other researchers can verify the methods, assumptions, and findings.
Competition in AI may extend beyond commercial systems into national scientific capacity. If AI can shorten parts of hypothesis generation and experimentation, countries with greater access to compute and advanced research systems may be able to test more ideas within the same period.
The Kyoto Vision does not establish how governments will provide that access or how they will measure the scientific return. It does, however, signal that compute policy and science policy are becoming more closely connected as AI takes a larger role in the discovery process.
