Forecasters who track the arrival of systems that can handle most cognitive work better than people have shifted their estimates repeatedly in recent years. Some pulled dates closer after the debut of models like ChatGPT. Others pushed them back when scaling seemed to hit snags. Then early 2026 brought fresh model gains from Anthropic that sent many estimates racing forward again.
Dan Schwarz charted these moves in detail. He examined repeated forecasts from researchers with strong track records who used roughly the same measure. That measure centers on automating most purely cognitive labor at superior quality, speed and cost. From 2023 into 2025 the majority shortened their timelines. Exceptions stood out. Tamay Besiroglu kept his outlook steadier.
But 2025 told a different story. Daniel Kokotajlo and Eli Lifland, authors of the detailed AI 2027 scenario, moved their expectations later. The Metaculus community followed. Dario Amodei of Anthropic and top forecaster Peter Wildeford did the same. Only Benjamin Todd shortened his median that year. The pattern aligned with a stretch of releases from xAI, Meta and Google that failed to deliver the leaps some anticipated.
Everything changed in the first months of 2026. Every forecaster who updated between January and April pulled their dates earlier. Schwarz included himself in that group. He pointed to rapid progress at Anthropic as the trigger that reversed the previous drift. The shifts appear tied to specific lab outputs rather than abstract theory.
Company leaders have offered their own views. Sam Altman stated in early 2025 that OpenAI now felt confident it knew how to build AGI. Demis Hassabis of Google DeepMind moved from a five-to-ten-year window to three-to-five years around the same time. Dario Amodei described the prospect of a data center full of genius-level systems arriving in two to three years. These public statements from CEOs at OpenAI, DeepMind and Anthropic fueled optimism even as some forecasters hesitated.
Yet not every update pointed the same direction. By late 2025 the AI Futures team behind the original AI 2027 projection revised its median to around 2030. Kokotajlo said his personal forecast had moved to that neighborhood while stressing large uncertainty remained. Eli Lifland noted 2027 stayed possible but the central expectation sat three years later. The group judged that actual progress through 2025 had reached only about two-thirds of the pace they once projected.
Prediction platforms captured the collective mood. At the start of 2025 Metaculus users placed the median for strong AGI in July 2031. One year later that figure had slipped to November 2033. The community had extended its outlook by two and a half years. Other aggregators told a similar tale. An analysis of nearly 10,000 predictions put median arrival somewhere between the late 2020s and early 2030s, with recent surveys showing acceleration after ChatGPT.
Broader expert surveys reflect the same compression. One review found that average forecasts dropped from a median of 50 years away in 2020 to much nearer dates. As of early 2026 a group of forecasters saw a 25 percent chance of AGI by 2029 and 50 percent by 2033. The direction of travel looks clear even if the exact year stays contested.
Definitions matter in these debates. Some focus on systems that outperform humans on most economically valuable tasks. Others emphasize passing rigorous benchmarks or achieving full automation of software engineering and research. Musk has described AGI as smarter than the smartest human, with arrival possible by 2026. Amodei has submitted formal comments to the White House expecting powerful systems in late 2026 or early 2027. Hassabis recently highlighted a 50 percent chance by 2030.
Progress on concrete problems keeps the discussion alive. DeepMind’s AlphaProof Nexus solved nine open mathematical problems that had resisted human efforts for decades. The system produced machine-checkable proofs at low cost. Such demonstrations prompt forecasters to reconsider their assumptions about remaining hurdles.
But gaps persist. Real-world robotic integration lags behind pure cognitive gains. Economic deployment faces regulatory, energy and data constraints. Many observers note that hype cycles can distort public perception while actual engineering milestones arrive more slowly than headlines suggest.
Markets have started to price in acceleration. Some prediction markets briefly showed 2026 as a live possibility for early AGI markers. Yet longer-term aggregates still cluster in the early 2030s. The distance between optimistic lab statements and community medians reveals how much disagreement remains.
Schwarz offered a simple frame for the pattern. The ChatGPT period drove forecasts earlier. The wave of 2025 models from several labs drove them later. Early 2026 advances from Anthropic reversed the move once more. He warned that strong Bayesians should not expect to predict their own next update. When intuition signals a coming shift, the right response may be to update immediately.
Industry insiders watch these numbers because the consequences stretch far beyond research papers. Automation of cognitive work at scale would reshape software development, scientific discovery, legal analysis and creative fields. Companies that bet correctly on timing could gain decisive advantages. Those that misjudge risk being left behind or overcommitting capital too soon.
Policy discussions have grown sharper. Some lawmakers focus on safety and misalignment. Others emphasize strategic competition and the risk of falling behind international rivals. European officials have raised concerns about American leadership in the technology. The interplay between capability gains and governance choices will likely determine how the next few years unfold.
Uncertainty reigns. No forecaster claims perfect foresight. The data show clear swings tied to specific model releases and experimental results. Recent demonstrations in mathematics and reasoning have reminded many why short timelines remain on the table. At the same time, repeated extensions by careful forecasters highlight the distance still to cover.
One thing looks consistent. The conversation has moved from abstract decades-away speculation to concrete arguments about the next two to five years. Labs continue to invest at record scale. Forecasters keep revising their charts. And the gap between today’s systems and full cognitive automation narrows with each breakthrough.
Whether the median lands in 2027, 2030 or later, the pace of adjustment itself signals something important. Expectations have compressed dramatically in a short period. That compression carries real weight for strategy, investment and planning across every sector touched by intelligent systems.
WebProNews is an iEntry Publication