Experts Challenge Narrow AGI Definitions at Santa Fe Institute
SANTA FE, N.M. (AP) — Cognitive scientists and AI experts convened at the Santa Fe Institute from March 31 to April 2, 2026, to scrutinize definitions of artificial general intelligence (AGI) through the lens of cognitive science. Organized by Melanie Mitchell and John Krakauer, the working group challenged the narrow benchmarks promoted by AI companies, emphasizing intelligence as multifaceted, including social interaction and open-ended curiosity. This approach contrasts with metrics that mimic IQ tests and capture only a fraction of natural abilities.
The closed-door event, hosted by the institute, drew participants from neuroscience, psychology, philosophy and AI. They highlighted a disconnect between industry hype and cognitive science, arguing that true intelligence involves rapid learning from others, task integration and physical interaction. No consensus definition of AGI exists, making timelines speculative, the group noted.
Broadening Perspectives on Intelligence
Experts at the meeting critiqued AI benchmarks that impress on paper but falter in real-world scenarios. "While they sound impressive, benchmark-based achievements often don’t hold up in the real world," said Mitchell, a Santa Fe Institute professor, in a news release. She added that cognitive scientists view intelligence as broader than mere task-solving.
The group drew on historical precedents, such as Alan Turing's 1950 imitation game, which questioned whether machines could equate to human reasoning. Participants favored comparative experiments over overarching metrics like IQ, underscoring elements like social interaction and curiosity in art or science.
This perspective aligns with broader skepticism in the field. A recent AAAI survey found that 76% of AI researchers doubt scaling large language models will achieve AGI, according to a Socos Academy report.
Clashing Views on AGI's Timeline and Arrival
In February 2026, University of California, San Diego, faculty claimed AGI has already arrived, asserting that current AI matches human-level reasoning, learning and problem-solving. They cited Turing's vision as fulfilled 75 years later in papers on arXiv and dev.what.it.is. These claims clash with the Santa Fe group's emphasis on definitional gaps and the AAAI survey's doubts.
A Nature journal study, using the PRISMA framework and BERTopic modeling, outlined five pathways for AGI development: societal integration, technological advancement, explainability, cognitive links to neuroscience and ethical alignment. The study avoided declaring AGI's arrival, focusing instead on balanced frameworks.
Discussions at the Santa Fe event touched on language-intelligence connections and risks like disinformation and goal misalignment. Participants noted a divide between industry and cognitive science, with varying opinions on existential threats.
Pathways Forward and Ethical Considerations
The Nature study stresses brain-inspired methods and ethical frameworks for AGI alignment, providing a counterpoint to contradictory claims. For instance, an arXiv paper proposes an "autonomy-supporting parenting" model for AGI, attributing a 30% chance of success under current paradigms to Stuart Russell in 2025. This advocates treating AGI like "child machines" to foster cooperation.
Broader trends indicate a shift from narrow AI to general systems, intersecting neuroscience and philosophy. The Santa Fe group discussed these intersections, highlighting the urgency of interdisciplinary scrutiny amid potential redefinitions of human-AI coexistence.
Bridging the Gap: Future Directions in AGI Discourse
The Santa Fe Institute plans to release a position paper synthesizing the group's views, which could reshape AGI debates. This follows the April 2026 event and aims to integrate cognitive science metrics into tech discussions.
Experts predict this effort will delay hyped timelines, urging policymakers to prioritize ethical "parenting" approaches and cognitive frameworks. By addressing the industry-science divide, such initiatives could mitigate risks like uncontrolled systems and disinformation, fostering a more grounded path toward AGI.