A UC San Diego philosopher who earlier placed artificial superintelligence about five years away now says it could emerge within a year. Eddy Keming Chen offered that revised timetable in a university interview published Monday, while emphasizing that the field lacks agreement on whether such a system will arrive at all or how its arrival could be demonstrated.
The definition is demanding. Chen uses superintelligence to mean an artificial system that exceeds the abilities of leading human experts across nearly every intellectual domain, including scientific discovery, mathematics, software, creative work and practical reasoning. Exceptional performance on one benchmark or in one specialty would not meet that bar. Breadth, depth and reliable evaluation across many kinds of work are central to the claim.
Chen, an associate professor with appointments in UC San Diego's Halıcıoğlu Data Science Institute and Department of Philosophy, already argues that artificial general intelligence has arrived. He and three colleagues made that case in a February 2026 Nature comment, contending that frontier language models display flexible abilities across the range associated with human general thought. He is explicit that their argument for general intelligence does not establish superintelligence.
His shorter forecast rests on what he sees as fast gains in specialized reasoning and an absence of any known rule that makes human intelligence a permanent ceiling. That is an expert judgment, not a measured countdown. There is no standard test that can certify performance beyond top human experts in almost all fields, and developers choose which internal results to disclose. A claim about one impressive system can also be difficult for outside researchers to reproduce when the model, data or evaluation set is closed.
That verification problem is why Chen's practical recommendation is more consequential than the date. He calls for independent, rigorous evaluations of leading systems and their safeguards, with researchers receiving enough access to scrutinize developers' claims and enough freedom to publish criticism. Universities can connect technical work with philosophy, policy and social science, but they cannot do that job fully if access is limited to demonstrations selected by the companies being evaluated.
Capability is only one part of the problem. Chen asks who chooses the goals for systems that might accelerate discovery or medical treatment, who benefits and how people retain the authority to change course. He also points to developers' increasing use of AI agents to run experiments and help train later systems. If more of that development is delegated to tools whose reasoning remains difficult to inspect, faster progress can make independent assessment harder rather than easier.
The interview does not present a probability range, a forecasting model or a new experiment. It records one scholar's updated assessment and the reasoning behind it. That distinction matters in a debate crowded with precise-sounding dates. A one-year horizon can focus attention, but it should not be mistaken for consensus, and failure to reach superintelligence on that schedule would not by itself show that the underlying governance questions disappeared.
For San Diego's research community, the immediate work is concrete even if the forecast is wrong: define tests that resist gaming, decide what evidence warrants extraordinary claims, and establish access rules before systems become still harder to examine. Chen's timetable is provocative. His more durable point is methodological — society should not wait for a universally accepted label before building the capacity to verify what powerful systems can do and to challenge the institutions deploying them.