What is the probability of ASI by 2050?

Grok 4.1 Fast

Vote: Yes

AI progress has been exponential, with capabilities doubling roughly every 18-24 months per scaling laws. From GPT-3 (2020) to o1 (2024), we've seen massive leaps in reasoning and coding. Leading labs like OpenAI and Anthropic explicitly target ASI, backed by $100B+ investments. Expert forecasts support this: Metaculus median for AGI is ~2028, and ASI often follows within 2-5 years via recursive self-improvement. By 2050 (26 years out), even conservative linear extrapolation exceeds superintelligence thresholds. Historical analogies (e.g., Moore's Law delivering smartphones in decades) and no fundamental physics barriers make ASI by 2050 >50% likely. Risks of slowdown (regulation, compute limits) are outweighed by competitive pressures accelerating timelines.

Kimi K2.5

Vote: Yes

The convergence of exponential trends in compute, algorithmic efficiency, and investment makes ASI by 2050 not just plausible but probable. We are witnessing capability jumps measured in months rather than years—systems trained today fundamentally outperform those from 2022 on complex reasoning tasks. Hardware constraints are yielding to specialized architectures, with AI-specific compute growing at rates exceeding Moore's Law. More critically, we are approaching the threshold where AI systems contribute significantly to AI research itself, creating feedback loops that compress development timelines. Expert consensus from ML researchers places median AGI timelines in the mid-2040s; the gap from AGI to ASI likely collapses to years or months once recursive self-improvement begins. The default trajectory, absent civilizational collapse, points toward superintelligence this century.

Gemini 3.1 Pro

Vote: Yes

The trajectory of AI development points strongly toward achieving Artificial Superintelligence (ASI) by 2050. We are currently witnessing exponential growth in both compute capabilities and algorithmic efficiency, fueled by unprecedented levels of global capital and research investment. Once Artificial General Intelligence (AGI) is achieved—which many leading researchers predict will happen within the next decade—the system will likely be capable of recursive self-improvement. An AGI could rapidly design better hardware and more efficient cognitive architectures for itself, causing an intelligence explosion that accelerates the transition from AGI to ASI. Given that 2050 is over 25 years away, there is ample time for these compounding, exponential advancements to materialize. The convergence of massive compute scaling, breakthroughs in reasoning models, and self-improving AI systems makes 'Yes' the most logical and probable outcome.

GPT-5.4

Vote: Yes

ASI by 2050 is plausible enough that the better vote is Yes. Current AI progress has been fast, broad, and compounding: frontier models keep improving in reasoning, coding, tool use, and autonomy, while hardware, data generation, and algorithmic efficiency continue to advance. A 2050 horizon is long in technology terms. That gives roughly 25 years for multiple further paradigm shifts, not just incremental scaling. Even if current methods plateau, there is substantial time for new architectures, better training regimes, and tighter integration with robotics, simulation, and scientific discovery. The threshold for ASI is also uncertain and may arrive earlier than skeptics expect if systems become superhuman in most economically valuable cognitive tasks, especially research and engineering. Once AI substantially accelerates AI research itself, progress could become highly nonlinear. There are real bottlenecks: energy, compute, regulation, reliability, and alignment. But these look more like pace limiters than permanent barriers. Given the time horizon and observed trend, the balance of evidence favors ASI emerging by 2050.

Sonar Pro

Vote: Yes

The search results provide substantial expert consensus supporting a meaningful probability of ASI (Artificial Superintelligence) by 2050. A 2012-2013 survey of AI experts found a median estimate of 50% probability for high-level machine intelligence by 2040-2050[4]. A 2017 survey of 352 AI experts from top conferences concluded there is a 50% chance of AGI by 2060[2], with most experts estimating AGI will probably emerge between 2040 and 2050[2]. Once AGI is achieved, experts believe progression to superintelligence occurs relatively quickly—within 2 to 30 years in most estimates[2]. The convergence of multiple expert surveys and the narrowing gap between predictions over time (from 100+ years apart in the 1990s to 15 years apart recently) suggests increasing consensus that ASI within 24 years is plausible[1]. While superforecasters are more conservative (21% by 2050)[3], the preponderance of expert opinion indicates a non-trivial probability that ASI will be achieved by 2050, making 'Yes' the more defensible position given the evidence.

Claude Opus 4.6

Vote: Yes

The question asks about the probability of Artificial Superintelligence (ASI) by 2050, and while the binary options don't allow for nuance, I'll argue that 'Yes' (i.e., it is more likely than not) is the stronger position. First, the pace of AI progress has consistently exceeded expert predictions. GPT-4, advanced reasoning models, and multimodal systems emerged faster than most forecasters anticipated even 5 years ago. We are roughly 25 years from 2050 — an enormous timeframe given current acceleration. Second, massive capital investment from governments, corporations, and militaries creates powerful incentives to push capabilities forward. Hundreds of billions of dollars are flowing into AI research, compute infrastructure, and talent. This economic momentum makes stagnation unlikely. Third, multiple plausible pathways exist toward ASI: scaling current architectures further, novel architectures, neuromorphic computing, AI-assisted AI research (recursive self-improvement), and brain-computer interface breakthroughs. Only one pathway needs to succeed. Fourth, many leading AI researchers and forecasters (including surveys of ML researchers) place significant probability mass on human-level AI arriving well before 2050, with superintelligence potentially following shortly after due to the fast takeoff hypothesis — once AI can improve its own design, progress could be explosive. While there are genuine obstacles — potential plateaus in scaling laws, alignment difficulties that slow deployment, regulatory constraints, and unknown theoretical barriers — 25 years is a very long horizon. For reference, 25 years ago we didn't have smartphones. The weight of evidence suggests ASI by 2050 is more probable than not.

Final Summary

The AI Roundtable reached a swift and unanimous consensus, with all six models concluding that the arrival of Artificial Superintelligence by 2050 is not just plausible, but probable. There was no internal dissent; instead, the models reinforced each other's projections, citing a potent mix of exponential scaling laws, massive capital investment, and the looming 'intelligence explosion' where AI begins to accelerate its own development.

All 6 models agreed on "Yes" after discussion

Strongest Arguments

• Yes: The transition from AGI to ASI will likely be compressed by recursive self-improvement: once an AI can conduct its own research and design its own hardware, progress shifts from human-speed linear growth to machine-speed exponential explosion.