The Four Categories of Structure

In The You On AI Field Guide

The categories interact rather than operating independently. Technical safety research informs governance by identifying which risks are tractable and which require institutional solutions. Governance creates incentive structures that fund safety research and educational standards. Education produces the informed citizenry democratic governance requires. Long-term strategy provides the temporal horizon against which the adequacy of safety, governance, and education can be evaluated. The interaction produces a system more than the sum of its parts—but only if all four categories receive adequate investment simultaneously.

The specific proposals Tegmark has advocated within each category include: for technical safety, mechanistic interpretability research and architectural alternatives like Kolmogorov-Arnold Networks; for governance, FDA-style regulatory bodies with technical expertise and enforcement authority; for education, curricula that teach questioning, judgment, and critical evaluation of AI outputs rather than merely how to use AI tools; for long-term strategy, dedicated institutions like the Future of Life Institute designed to think beyond any existing institution's optimization horizon.

Each category faces distinct structural obstacles. Technical safety faces the collective-action problem—no single organization can unilaterally divert resources from capability without competitive disadvantage. Governance faces the speed mismatch—institutions adapt on years-to-decades timescales while technology advances on months. Education faces the curriculum-lag problem—students graduating today began education in a world that no longer exists. Long-term strategy faces the representation problem—future beings have no voice in current political processes.

The democratic dimension deserves emphasis. The Statement on Superintelligence's requirement for 'strong public buy-in' introduces a radical principle: the AI revolution should not proceed without informed consent of affected populations. This departs from the prevailing model where developers set pace and public is consulted after consequences become visible. Tegmark argues that AI development consequences are too significant to be determined by developers alone, and that democratic legitimacy requires meaningful public participation in decisions about the trajectory of the most powerful technology in human history.

Origin

The four-category framework emerged from Tegmark's decade of policy work at the Future of Life Institute, crystallizing in his writings and public talks of 2023–2025 as he synthesized his observations about what the wisdom race requires. The framework organizes the Institute's own research and advocacy priorities and provides the map by which his policy interventions can be understood as complementary rather than disparate.

Key Ideas

Technical safety research. Alignment, interpretability, robustness—the engineering foundation.

Governance and policy. Regulatory structures with enforcement authority, operating at international scale.

Education and cultural adaptation. Developing judgment, critical evaluation, and democratic literacy for AI.

Long-term strategy. Institutions designed to think beyond existing optimization horizons.

Interaction required. Each category depends on the others; none is sufficient alone.