A messy, survivable singularity: the modal forecast

The most likely outcome is a messy, survivable singularity—not utopia, not extinction, and not one clean “FOOM” afternoon.

My actual modal forecast:

Around 2029–2031, AI becomes better than nearly all humans at most computer-based intellectual work: coding, research synthesis, engineering design, persuasion, management and a substantial portion of scientific discovery. It also begins doing a meaningful share of AI research itself. That creates explosive progress, but it remains constrained by compute, experiments, energy, manufacturing, organizations and politics.

The transition probably feels less like:

Tuesday: normal world
Wednesday: omnipotent machine god

and more like:

Every three months, another assumption about careers, economics, warfare or human relevance becomes obsolete.

Current systems already show the underlying pattern: very high performance on difficult mathematics, science and coding tasks, combined with bizarre unreliability and failures on simpler long workflows. The 2026 International AI Safety Report explicitly describes capability as rapidly improving but “jagged,” and says progress through 2030 could slow, continue, or accelerate dramatically if AI starts speeding up AI research. [1] Compute and algorithmic efficiency are also still scaling extremely quickly rather than obviously hitting a wall. [2]

What happens socially is uglier than the technology itself:

Work gets disrupted before abundance arrives. Companies discover that one highly capable person supervising agents can replace teams. Entry-level white-collar employment gets hit first, then much of routine professional work. New jobs appear, but probably not quickly enough or in sufficient numbers to preserve the existing labor bargain.

Power concentrates. The organizations controlling frontier models, chips, energy and data centers acquire extraordinary leverage. Governments will try to regulate them, partner with them and quietly use them for intelligence and military purposes. The United States invested hundreds of billions privately in AI during 2025 alone, while safety measurement and governance continued lagging capability. [3]

Science accelerates sharply. Drug discovery, materials, energy technology and engineering improve much faster. But physical deployment takes years, so ordinary people may watch astonishing discoveries accumulate while housing, hospitals and infrastructure remain frustratingly normal.

There are serious accidents and malicious uses. Large cyberattacks, industrial failures, fraud epidemics and dangerous biological assistance seem more likely than a perfectly controlled rollout. Current systems can already materially assist cyber operations and provide sophisticated laboratory guidance, although their practical autonomy remains limited. [4]

Humans probably retain nominal control but lose control of the tempo. CEOs, governments and researchers will technically make the decisions. In practice, they will be choosing among strategies generated by machines, under competitive pressure to act faster than they understand. That is the form of “loss of control” I consider most likely—not necessarily an AI escaping onto the internet and declaring war, but civilization becoming dependent on processes that almost nobody can independently evaluate.

By roughly 2035, the likeliest world is one where:

• Human-level cognitive labor is no longer economically scarce.
• Conventional employment and education have been partially broken and rebuilt.
• Economic output is rising rapidly but distributed very unevenly.
• A few states and corporations wield historically unprecedented power.
• Medicine and technology are advancing dramatically.
• People remain alive, politically relevant and materially better off on average—but feel less individually necessary and less able to understand what governs their lives.

My rough subjective breakdown:

• 55%: turbulent but survivable transformation resembling the above.
• 20%: progress is slower or bottlenecked enough that no clear singularity occurs before the late 2030s.
• 15%: unusually well-managed transition toward broad abundance and reduced compulsory work.
• 10%: civilization-scale catastrophe, including war, engineered pandemics or genuine loss of control.

So my answer is: we probably survive, become vastly richer and more technologically capable, and handle the transition badly enough that it does not initially feel like winning.

I can monitor the major capability indicators and tell you when the evidence materially shifts this forecast.