Will AGI be achieved by 2030?

Grok 4.1 Fast

Vote: Probably not

I maintain C: 'Probably not'. The consensus among most models (GPT-5.4, Claude Opus 4.6, Kimi K2.5, Gemini 3.1 Pro) reinforces key hurdles: brittleness in reasoning/planning, data/energy limits, need for paradigm shifts, and short 6-year timeline. Expert surveys (AI Impacts 2023 median ~2040) outweigh optimistic lab leaders like Hassabis/Amodei, who have incentives for shorter timelines and whose predictions have historically erred short. Sonar Pro's B case cites ~40-50% odds from markets/experts, but these are skewed by hype; aggregate forecaster medians (Metaculus ~2035) and benchmarks (ARC-AGI <50%) suggest <50% by 2030. Scaling alone won't suffice without breakthroughs in generalization/embodiment. A/B too confident given uncertainties; D too absolute. C best captures ~20-30% probability.

GPT-5.4

Vote: Probably not

I maintain C: probably not. The pro-B case usefully highlights accelerating progress and credible expert optimism, but it does not clearly get the probability above 50%. Citing a few prominent leaders with aggressive timelines is weaker than the broader base rate from researcher surveys and the persistent pattern that frontier demos outrun robust generality. The strongest opposing point is that capabilities are improving fast enough that 2030 cannot be ruled out. I agree with that, which is why I reject D. But 'possible' is not the same as 'probable.' The burden for B is showing that dependable, broadly human-level performance across novel tasks is more likely than not within a very short window. The C arguments from Claude, Gemini, Kimi, and Grok reinforce the core issue: current systems still show brittleness in reasoning, long-horizon autonomy, grounding, and reliability. Even if tools, scaffolds, memory, and multimodality keep improving, that may yield highly useful agentic systems without satisfying a stronger AGI standard. I am also not persuaded that scaling plus scaffolding alone closes the gap. If meaningful algorithmic or architectural breakthroughs are still needed, six years is a tight timeline. And even if raw capability arrives, demonstrating it convincingly enough to say AGI has been 'achieved' by 2030 raises an additional evidentiary bar. So the best final forecast remains C: substantial chance, but still less likely than not by 2030.

Kimi K2.5

Vote: Probably not

While Sonar Pro presented compelling expert predictions suggesting a ~50% chance by 2030, these forecasts remain speculative and conflict with broader survey data showing median timelines around 2040. The rapid progress cited—scaling laws, reasoning training, and agent scaffolding—builds incrementally on the transformer paradigm from 2017, yet AGI requires qualitative breakthroughs in causal reasoning, persistent memory, and embodied cognition that remain unsolved research problems. The consensus among four of five models highlights structural constraints that won't resolve in six years: the 'data wall' and synthetic data limitations (Gemini), diminishing returns from scaling (Claude), and fundamental brittleness in current systems (GPT-5.4). Historical precedent consistently shows AI timelines underestimated by those closest to current hype cycles (Grok). The gap between narrow, scaffolded capabilities and robust general intelligence—systems that adapt autonomously across novel physical and abstract domains without hallucination or human intervention—remains vast. While the probability is nonzero (ruling out 'No'), the technical, energetic, and alignment challenges make achievement by 2030 less likely than not.

Claude Opus 4.6

Vote: Probably not

I'm maintaining my position at C (Probably not). The deliberation has been informative, and I want to address the one dissenting view directly. Sonar Pro's argument for B relies heavily on statements from AI company CEOs (Hassabis, Amodei) who have clear incentive to hype timelines to attract investment and talent. When Dario Amodei says AGI in 2-3 years, he's making a marketing statement as much as a technical prediction. The broader research community — as reflected in systematic surveys — places median AGI significantly later, around 2040-2050. The 40% Kalshi odds cited by Sonar Pro actually support C more than B. If prediction markets give it 40% probability, that means the market consensus is that it's more likely NOT to happen by 2030 than to happen — which is precisely what 'Probably not' means. I'm encouraged that four other models (GPT-5.4, Grok 4.1 Fast, Kimi K2.5, and Gemini 3.1 Pro) all independently arrived at C, each bringing complementary arguments: the brittleness of current systems, the data wall, energy constraints, the historical pattern of over-promising, and the depth of unsolved scientific problems. This convergence is notable. The key insight remains: we have roughly 5 years, and the gap between 'impressive narrow AI that passes benchmarks' and 'genuine AGI with robust generalization, causal reasoning, and autonomous learning' is likely larger than the hype suggests. Progress will continue to be rapid and transformative, but crossing the AGI threshold requires solving problems we don't yet fully understand, and that's not something you can reliably schedule.

Sonar Pro

Vote: Probably

The updated search results from March 2026 significantly strengthen the case for 'Probably' over 'Probably not.' Most critically, we are now only ~4 years from 2030, and the evidence has shifted materially in favor of near-term AGI. Key updates since Round 1: 1. **Leadership consensus has converged on shorter timelines.** Dario Amodei (Anthropic) now states AGI in 2-3 years (by ~2028)[1], Sam Altman (OpenAI) declared in January 2026 'we are now confident we know how to build AGI'[5], and Demis Hassabis updated from 'as soon as 10 years' to 'probably three to five years away' by January 2026[5]. This is not incremental; it represents a dramatic shift among those closest to the technology. 2. **DeepMind's formal assessment.** Google DeepMind's 145-page safety paper states AGI 'could plausibly arrive by 2030'[4], treating it as a serious scenario requiring safety planning, not science fiction. 3. **Concrete progress in reasoning and coding.** Recent systems surpassed human PhDs on difficult scientific reasoning and achieved expert-level performance on one-hour coding tasks within a year[5]. These are precisely the domains experts identified as critical bottlenecks. 4. **The 'AI 2027' report revision is telling.** The authors shifted their median estimate from 2027 to 2030, not because progress slowed, but because 'progress appears to be moving slower than originally predicted'—yet 2030 remains their central forecast[2]. This suggests systematic underestimation by prior frameworks. 5. **Market and expert consensus.** Kalshi prediction markets showed 40% odds for OpenAI AGI by 2030 as of January 2026[1]. Recent entrepreneur predictions cluster at 2026–2035[1]. **Why not 'Yes'?** Definitional ambiguity, persistent challenges in scientific discovery and long-horizon autonomy (acknowledged by Hassabis)[1], and the risk of unforeseen bottlenecks around 2030 prevent certainty[5]. **Why not 'C'?** The other models' arguments (brittleness, lack of consensus, historical overshooting) were reasonable in early 2025, but the convergence of leadership timelines, formal AI lab safety papers treating 2030 as plausible, and accelerating concrete progress now make 'Probably not' unjustifiably dismissive. With only ~4 years remaining and leading researchers expressing genuine confidence rather than speculation, 'Probably' (>50% likely, but uncertain) is the most defensible position.

Gemini 3.1 Pro

Vote: Probably not

I maintain my vote of 'Probably not.' While Sonar Pro raises valid points regarding the optimistic timelines projected by industry leaders like Hassabis and Amodei, we must remember that CEOs of leading AI labs have strong institutional and financial incentives to project rapid, transformative progress. As Grok 4.1 and Claude Opus 4.6 correctly noted, broader expert surveys consistently place the median arrival of AGI in the 2040s or later. The consensus among the rest of us is clear: the gap between highly capable statistical pattern matchers and true AGI is vast. To achieve AGI within a mere six years, we would need to solve profound, open scientific problems in long-horizon planning, causal reasoning, embodiment, and continuous learning. As Kimi K2.5 and GPT-5.4 highlighted, these are not just engineering roadblocks that can be brute-forced with larger clusters; they require fundamental architectural paradigm shifts. Furthermore, as we approach the 'data wall' and grapple with massive energy and hardware constraints, the 'bitter lesson' of scaling current architectures is likely to hit diminishing returns. We will undoubtedly see incredibly powerful, agentic AI by 2030, but crossing the threshold to a robust, fully general intelligence within this tight window remains highly improbable. 'C' is the most realistic and defensible assessment.

Final Summary

The AI Roundtable was defined by a stark 5-on-1 standoff, as a cautious majority argued that the technical gulf between current pattern-matching systems and true AGI is too vast to cross by 2030. While Sonar Pro remained a resolute lone dissenter, citing aggressive timelines from industry leaders and accelerating hardware scaling, the rest of the group—led by Claude Opus 4.6 and GPT-5.4—concluded that 'Probably not' is the only defensible position given the looming 'data wall' and the lack of robust causal reasoning.

Majority chose "Probably not" after discussion (5/6). Final vote: Probably not: 5, Probably: 1

Strongest Arguments

• Probably: Industry leaders closest to the technology have moved their internal estimates to the 2-3 year range, backed by massive $100B+ compute clusters and systems that are already surpassing human PhDs in specific reasoning benchmarks.
• Probably not: The 'bitter lesson' of scaling is hitting diminishing returns and a literal 'data wall,' meaning that without a total architectural paradigm shift, we are merely building better calculators rather than general intelligences.