The AI Crossroads: How Artificial Intelligence Could Stall, Slow, or Surge by 2030

The OECD finds that AI’s path to 2030 is highly uncertain, ranging from a plateau at today’s capabilities to rapid advances that could match or exceed human performance. Because all scenarios remain plausible, policymakers must prepare for uncertainty rather than betting on a single AI future.

Artificial intelligence is advancing so quickly that even its leading architects are unsure where it will stand by the end of this decade. A major new OECD study, developed by the organisation’s Strategic Foresight Unit and AI policy teams in collaboration with the Global Partnership on AI, draws on expertise from institutions such as the University of California, Berkeley, the University of Montreal, Osaka University, Linköping University, the University of Southampton, and Pontificia Universidad Católica de Chile. Its core message is clear: by 2030, AI could plausibly stagnate near today’s capabilities, or surge ahead to rival or surpass humans in many areas. The challenge for governments is that both futures remain possible.

How far AI has already come

The report begins with a sober assessment of recent progress. In just a few years, AI systems have crossed milestones once considered far off. They now outperform human experts on some PhD-level science questions, achieve gold-medal performance in elite mathematics and programming competitions, translate major languages at near-human quality, and complete increasingly long digital tasks with limited supervision. Benchmark data shows steep improvement curves across language, reasoning, coding, and vision. Yet these gains are uneven. AI systems still hallucinate, struggle to learn continuously, perform poorly outside narrow problem domains, and remain weak at physical tasks and complex social interaction. Progress has been rapid, but also fragile.

Four possible paths to 2030

Instead of predicting one outcome, the OECD outlines four broad futures. In the first, progress stalls. After 2025, AI improvements flatten. Systems in 2030 look more polished but not fundamentally more capable than those available today. They excel at short, well-defined tasks and draw on vast knowledge bases, but still require heavy human guidance. Memory remains limited, errors persist, and long-horizon autonomy proves unreliable. This scenario could emerge if scaling laws weaken, investment slows, or constraints on compute, energy, or data bite earlier than expected.

A second path sees progress slow rather than stop. AI continues to improve, but each gain is harder won. By 2030, systems function as reliable digital assistants, autonomously completing tasks that would take humans days. Accuracy improves and hallucinations decline, but true human-like learning remains elusive. Robotics advances only modestly and largely stays in controlled environments. Social interaction improves but still lacks depth and flexibility.

The third scenario imagines continued rapid progress. Here, AI in 2030 autonomously executes complex digital projects lasting weeks, outperforms many professionals in structured reasoning, and operates with high autonomy within human-set limits. Memory and learning improve enough to approximate “learning on the job,” while AI-guided robots begin handling selected tasks in dynamic real-world environments. Social skills advance to support ongoing relationships and multi-party interactions.

The fourth and most dramatic future sees progress accelerate. In this case, AI systems reach human-level or superior performance across most cognitive domains by 2030. They reason strategically, revise goals, learn continuously, and generate genuinely novel creative outputs. Robotics advances rapidly, allowing AI-guided machines to operate competently across many physical settings, often in specialised roles. While speculative, the OECD treats this outcome as plausible if breakthroughs and AI-assisted AI development compound current trends.

Why uncertainty dominates

A central theme of the report is uncertainty. The forces driving AI forward, scaling larger models, improving algorithms, and investing massive resources, may not hold indefinitely. Energy and water constraints could limit data centres, returns on investment may fall, and AI-assisted software development may prove less transformative than hoped. Expert interviews reveal sharp disagreement on timelines and outcomes, but near-universal low confidence in precise prediction. Some experts expect human-level AI within decades; others doubt it will ever arrive.

What this means for policy

The OECD’s conclusion is not a forecast but a warning. None of the four scenarios can be confidently ruled out. By 2030, AI could plateau near today’s capabilities or reshape economies and societies at extraordinary speed. For policymakers, betting on a single future is risky. Instead, governance must be flexible enough to handle both rapid acceleration and unexpected stagnation. In the age of AI, the report argues, uncertainty is no longer a temporary problem to solve, it is the reality policymakers must plan for.