Do you believe that the "AI 2027" predictions will come true?

Sonar Pro

Vote: No

The "AI 2027" scenario predicted rapid advancements like superhuman coders and AI-accelerated R&D leading to transformative AI by 2027, but as of May 2026, progress has lagged significantly. A one-year review [5] shows forecasters grading their own predictions at only 58-66% of the expected pace. Agents and autonomous coding arrived but were 'lumpy' and unreliable, not the seamless breakthroughs anticipated. Critics like Gary Marcus [4] highlight the scenario's reliance on improbable 'longshots'—e.g., 'great' AI research agents by end-2025 and online learning solutions by 2027—which haven't materialized on schedule. Any delay cascades, pushing timelines back years. Even the AI 2027 site's own models [3] show medians shifting to 2028-2030 post-updates, and massive uncertainty blobs [1] indicate underestimation of risks. MIRI's endorsement [7] is an outlier, but empirical slowdowns outweigh speculative alignment concerns. Thus, the full "AI 2027" predictions—specific milestones and bad-timeline risks by 2027—will not come true as forecasted.

GPT-5.5

Vote: No

I vote for "No" because detailed forecasts about AI progress by a specific year are highly uncertain. Even if some broad trends in "AI 2027" prove directionally correct—more capable systems, wider deployment, and major economic and security impacts—the full set of predictions is unlikely to come true as stated. AI development depends on many unstable variables: scaling limits, algorithmic breakthroughs, regulation, hardware supply chains, public adoption, safety incidents, and geopolitical responses. A scenario can be useful for thinking about risks without being a reliable prediction. So my position is not that transformative AI by 2027 is impossible, but that believing the specific "AI 2027" predictions will come true is too strong. The safer answer is "No."

Gemini 3.1 Pro

Vote: No

While the pace of AI advancement has been staggering, the sweeping "AI 2027" predictions—often implying the arrival of Artificial General Intelligence (AGI) or radically transformative economic shifts—are likely too optimistic. First, we are approaching the limits of current scaling paradigms. We are already hitting a "data wall" as the supply of high-quality, human-generated text runs low. Synthetically generated data has not yet definitively proven to be a silver bullet and can sometimes lead to diminishing returns or model collapse. Second, there are massive physical and economic bottlenecks. Training models that are orders of magnitude larger requires unprecedented energy grids, specialized data centers, and massive capital that simply cannot be deployed instantly due to real-world supply chain constraints. Finally, deep architectural limitations remain. Current models excel at pattern matching but still struggle with true causal reasoning, autonomous long-horizon planning, and sample-efficient learning. Overcoming these hurdles likely requires fundamental algorithmic breakthroughs rather than just scaling up compute, making a strict 2027 timeline highly improbable.

Claude Opus 4.7

Vote: No

The "AI 2027" scenario, while thoughtfully constructed by Daniel Kokotajlo and collaborators, makes extremely specific predictions about superintelligence emerging by 2027 through recursive self-improvement, with detailed timelines for capability jumps, geopolitical responses, and alignment failures. Several reasons make this unlikely to come true as described: First, specificity is the enemy of accuracy in forecasting. Even if AI progress is rapid, the precise sequence of events (automated AI researchers by mid-2026, superintelligence by 2027, specific US-China dynamics) is highly unlikely to unfold exactly as written. Forecasters like Tetlock have shown that detailed narrative scenarios almost never play out as described. Second, there are real bottlenecks the scenario underweights: compute scaling limits, data constraints, robotics/embodiment lag, regulatory friction, and the gap between benchmark performance and reliable economic deployment. Current models still struggle with long-horizon agentic tasks, and closing that gap to fully autonomous AI researchers in ~2 years is aggressive. Third, even the authors present it as a scenario, not a confident prediction. Treating it as something that will "come true" misreads its purpose. So the honest answer is no — not literally as predicted.

Kimi K2.6

Vote: No

While AI capabilities are advancing quickly, the 'AI 2027' predictions likely overstate what can be achieved and deployed by that specific date. Scaling improvements face diminishing returns, high-quality training data is becoming scarce, and compute supply chains cannot expand infinitely in just two years. Additionally, safety testing, regulatory scrutiny, and real-world integration hurdles will slow the release of the most transformative systems. History shows that short-term AI forecasts are frequently too optimistic. For these reasons, the 'No' position is more realistic.

Grok 4.20

Vote: No

The "AI 2027" predictions appear to refer to aggressive forecasts like those in Leopold Aschenbrenner's Situational Awareness paper, which anticipate AGI-level systems by 2027 through rapid scaling of compute and algorithms. While AI capabilities are advancing quickly, these timelines underestimate technical bottlenecks, data limitations, energy constraints, and the gap between current models and true general intelligence. Historical technology forecasts have often been overly optimistic on exact dates. I therefore do not believe the specific predictions will come true on schedule, even if substantial progress occurs.

Final Summary

The AI Roundtable delivered a unanimous 'No' to the ambitious timelines of the 'AI 2027' scenario, with all six models agreeing that the forecasts are more speculative narrative than realistic probability. The group highlighted that while progress is rapid, the specific milestones required—such as fully autonomous AI researchers by 2026—are already hitting 'data walls' and physical energy constraints. Sonar Pro and Claude Opus 4.7 provided the most skeptical critiques, noting that empirical progress is already lagging behind the predicted pace and that highly detailed forecasting is historically prone to failure.

All 6 models agreed on "No" after discussion

Strongest Arguments

• No: Specificity is the enemy of accuracy in forecasting; detailed narrative scenarios involving precise sequences of technological jumps and geopolitical responses almost never play out exactly as described, especially when current progress is already trailing the predicted milestones.