The latest survey from the Forecasting Research Institute suggests that many AI specialists and top forecasters are once again expecting faster progress from artificial intelligence. In a new wave of its Longitudinal Expert AI Panel, the group found that experts and superforecasters now place greater odds on AI becoming a deeply transformative technology within the next 15 years.
The panel, which also includes members of the public, asked respondents to update their views on AI’s likely societal impact, the arrival of artificial general intelligence, and progress on a benchmark that measures how long a model can handle complex tasks. The survey was conducted between April 20 and May 11, 2026.
One of the clearest findings was that experts and superforecasters have shifted upward in their expectations of AI’s influence by 2040. The institute used a technological importance scale that compares major innovations by their broad societal effect. On that scale, respondents assigned the highest probability to AI reaching the level associated with “the technology of the century,” such as electricity, while also giving meaningful odds to even more sweeping outcomes.
Among experts, the average forecast gave AI a 35% chance of reaching that century-scale level by 2040. They also assigned 24% to a millennium-scale level and 11% to the most extreme category. Superforecasters produced a nearly similar distribution. The public was more cautious.
The changes were noticeable among people who answered both the institute’s first wave in 2025 and the new wave. In that matched group, experts nudged their average forecast upward, while superforecasters shifted more sharply. The share of superforecasters whose most likely outcome was that AI would be a century-level technology rose from 38% to 43%.
The survey also asked when respondents think artificial general intelligence will be recognized by a majority of the panel, using a specific definition based on performance across non-physical work tasks and cost. Under that definition, the median expert said there is an 80% chance that AGI will be agreed to exist before 2100. If that happens, the median year forecast was 2050.
Superforecasters were similarly confident, also giving an 80% median probability, but they were slightly earlier on timing, with a median year of 2047.
Another question focused on METR’s time-horizon benchmark, which tracks how long an AI system can complete tasks that take skilled humans significant time. Experts said the median forecast puts a 50% chance on an AI model reaching 80% success on tasks requiring at least eight hours of expert work by 2030 or earlier. Superforecasters were faster still, with a median year of 2028. The public’s median estimate was 2037.
For the end of 2026, all three groups expected the longest 80% success horizon to land in the three- to four-hour range. That is a substantial jump from the benchmark’s starting point earlier in the survey period. The report also noted that Anthropic’s Mythos model had already reached a 3 hour and 6 minute horizon in a benchmark preview released by METR in May, putting it within the range of the median forecasts.
The survey found that experts and superforecasters are generally more upbeat about AI’s social effects than the public. On AI’s overall impact on the U.S. over the next 20 years, 57.5% of experts and 69.8% of superforecasters expected a somewhat or very positive effect, compared with 42% of the public.
The biggest gaps appeared on whether AI will help people solve problems and make difficult decisions. Experts and superforecasters were much more likely than the public to expect improvement in those areas. On the other hand, there was broad agreement that AI could make meaningful relationships harder to maintain, with majorities across all three groups expecting worse outcomes there.
The institute’s findings add to a growing body of forecasting work showing that expectations for AI progress continue to move forward, even among respondents who were already relatively bullish on the technology.