The center of gravity among AI researchers is not “nothing changes” and not a fully automated economy by 2036. The modal view is: AI systems will become much more capable, more agentic, and deeply embedded in software, science, education, medicine, media, and office work; many specific expert-level tasks will be automated or co-produced with AI; governance and safety will matter more because misuse, concentration of power, cyber/bio/misinformation risks, and alignment failures become more plausible. But researchers remain sharply divided on whether this decade produces broadly human-level or superhuman systems, and especially divided on whether those systems translate quickly into full occupational automation or explosive economic growth.
A concise forecast consistent with the survey evidence: by 2036, AI is likely to be a general-purpose cognitive infrastructure comparable in social importance to the internet or electrification, with highly capable agents able to complete multi-step digital work. It is plausible, but not consensus, that machines outperform humans at nearly every task by then. It is unlikely, even in the optimistic automation surveys, that all human occupations are fully automatable by 2036; the median forecast for that is far later.
1. Capabilities will keep advancing rapidly, especially for digital tasks.
The strongest direct evidence is the 2024 survey “Thousands of AI Authors on the Future of AI,” which gathered forecasts from 2,778 researchers publishing at top AI venues. The aggregate forecast gave at least a 50% chance that AI systems would achieve several concrete milestones by 2028, including building a payment-processing website from scratch, creating a song indistinguishable from a new song by a popular musician, and autonomously downloading and fine-tuning a large language model. That implies that most surveyed researchers expected major capability gains well before the 10-year mark.
2. There is a real chance of broadly human-level performance within the decade, but it is not a settled consensus.
The same survey estimated a 10% chance by 2027 and a 50% chance by 2047 that unaided machines outperform humans in every possible task, assuming science continues undisrupted. Interpolating from those numbers, 2036 is inside the serious-probability window, not the median expectation. In plain English: many researchers think transformative AI could arrive within 10 years; the aggregate median still puts full “outperform humans at every task” somewhat later.
3. Full job/occupation automation is expected to lag capability benchmarks.
The AI-authors survey forecast only a 10% chance by 2037 and 50% by 2116 that all human occupations become fully automatable. That gap is important. Researchers distinguish “AI can do many hard tasks” from “the labor market, institutions, liability regimes, robotics, workflows, and trust all adapt enough for complete occupational automation.”
4. Researchers are more worried than the general public narrative often suggests.
In the AI-authors survey, 68.3% thought good outcomes from superhuman AI were more likely than bad outcomes, but many still assigned non-trivial probability to catastrophe. Between 38% and 51% gave at least a 10% chance to outcomes as bad as human extinction from advanced AI. More than half said substantial or extreme concern was warranted for scenarios including misinformation, authoritarian control, and inequality. The broad view is therefore “high upside, high uncertainty, non-negligible tail risk,” not simple optimism.
5. Safety and governance researchers broadly agree risk-reduction work should be prioritized more.
The International AI Safety Report, led by Yoshua Bengio and written by over 100 independent experts with an international advisory process, describes advanced general-purpose AI as a fast-moving area with capabilities and risks that require shared scientific assessment. It emphasizes that current safeguards remain incomplete and that real-world effectiveness is uncertain. This is consistent with the survey finding that there was broad agreement that research aimed at minimizing potential AI risks should receive more priority.
6. Economic impacts are expected to be large, but there is no consensus on “explosive growth.”
A more skeptical macroeconomic view comes from Daron Acemoglu’s 2024 paper “The Simple Macroeconomics of AI,” which estimates AI-driven total factor productivity gains over 10 years at no more than 0.66%, and possibly under 0.53%, using task exposure and cost-saving assumptions. This is a minority counterweight to more transformative forecasts: even if AI is technically impressive, the economic payoff may be bottlenecked by adoption, complementary organizational change, regulation, hard-to-measure tasks, and human institutions.
Most AI researchers appear to believe the next 10 years will bring very large AI capability gains and meaningful social disruption. The median researcher view is not that all jobs disappear by 2036, and not that superhuman AI is guaranteed by then. The best summary is: powerful AI agents and broad task automation are likely; full occupational automation is unlikely; human-level-or-better AI across nearly all tasks is plausible but uncertain; and the risk distribution is unusually wide, with both extraordinary benefits and serious catastrophic-risk concerns treated as credible by substantial parts of the field.