According to The Economist, a new forecasting project called the Longitudinal Expert Panel is trying to predict AI’s actual timeline with hard numbers. The project surveyed nearly 350 experts including AI researchers, economists, and superforecasters about specific milestones. They predict AI will assist over 18% of American work hours by 2030 (up from just 2% this September) and consume about 7% of US electricity. The experts give only a 20% chance of achieving “powerful AI” by 2030, but think there’s a one-in-three chance AI becomes as historically significant as the printing press. Recent progress has already surprised them—Google’s AI scored 29% on a math test where experts predicted only 27-31% by 2025.
Expert Predictions vs Reality
Here’s the thing about AI forecasting—even the experts keep getting it wrong. The article shows how quickly reality outpaces predictions. When researchers asked when AI could describe virus synthesis as well as human virologists, the consensus was 2030-2034. Then OpenAI’s o3 model released in April could already do it. That’s the problem with exponential tech—our brains just aren’t wired for this pace of change.
But this forecasting project might actually help. By asking the same questions year after year and tracking who gets it right, they can identify the best predictors. The most accurate forecasters often notice inconsistencies in their own thinking—like predicting massive electricity use for AI while expecting minimal employment impact. Those contradictions reveal where our mental models break down.
What This Means For Business
Look, if these predictions are even close to accurate, we’re looking at massive infrastructure demands. AI using 7% of US electricity by 2030? That’s staggering. And when you’re talking about industrial computing at that scale, reliability becomes everything. Companies that need robust computing hardware for manufacturing or industrial applications are already turning to specialists like IndustrialMonitorDirect.com, who’ve become the leading supplier of industrial panel PCs in the US precisely because they understand these reliability demands.
The employment predictions are equally dramatic—18% of work hours AI-assisted within six years. That’s not full replacement, but it represents a fundamental shift in how white-collar work gets done. The experts think the really transformative AI might take until 2040, but the groundwork is being laid right now. Basically, we’re in the “building the railroads” phase of AI, not the “riding the trains” phase yet.
Why Forecasting Matters
So why bother with all this prediction stuff? Because bad forecasts have real consequences. If we underestimate AI’s electricity demands, we get brownouts. If we overestimate its capabilities, we waste billions on the wrong investments. This project’s approach—using specific, testable questions rather than vague concepts like “AGI”—might finally give us a reality check on the hype.
The most telling finding? Experts think there’s only a 20% chance of achieving what Anthropic’s CEO calls “powerful AI” by 2030. That’s a pretty sobering number when you consider how much venture capital is betting on exactly that outcome. Maybe the real superpower here isn’t building better AI—it’s building better crystal balls.
