According to Gizmodo, top AI CEOs are making bold predictions about superintelligent AI arriving imminently. Anthropic’s Dario Amodei says we’ll have AI systems “broadly better than almost all humans at almost all things” by 2026 or 2027, while Elon Musk claims we’ll have AI smarter than any single human this year and exceeding all human intelligence combined by 2030. OpenAI’s Sam Altman thinks artificial general intelligence will “probably get developed” before the end of Trump’s presidential term. But a new study from the Forecasting Research Institute reveals experts give these rapid timelines only a 23% chance of actually happening. The research surveyed computer scientists, economists, and AI researchers whom policymakers would consult, finding most expect AI to have significant impact by 2040 rather than within this decade.
Expert skepticism
Here’s the thing about these CEO predictions – they sound amazing, but the people actually building this stuff are way more cautious. The experts surveyed pointed to something crucial that tech leaders often overlook: implementation barriers. As one respondent put it, “Radical change in major systems just takes longer than 4-5 years.” Basically, even if the AI itself gets really smart, integrating it into real-world systems takes time. Think about how long it took electricity to transform society after its invention. The same principle applies here.
Another expert highlighted the bottleneck problem – we might have super-smart AI, but then we’ll discover “thousands (millions?) of potential bottlenecks in the economy which will only become legible as other processes are sped up.” It’s like having a Formula 1 engine but only country roads to drive it on. The infrastructure just isn’t there yet. And honestly, when you look at how even basic enterprise software implementations can take years, does anyone really believe we’ll overhaul entire industries in under five years?
Benchmark problems
There’s another elephant in the room that makes these predictions even shakier. A recent Oxford study found that many popular AI benchmarking tools are unreliable or misleading. So when companies claim their models are getting smarter, we might not even be measuring the right things properly. It’s like trying to predict Olympic performance based on training results that turn out to be flawed. This calls into question whether we’re actually seeing the rapid progress that tech leaders keep talking about.
And here’s where it gets really interesting – even the so-called “superforecasters” in this study have historically underestimated AI progress. Back in 2022, experts thought AI wouldn’t win a gold medal in the International Mathematical Olympiad until 2030, and superforecasters said 2035. But Google’s AI system actually did it this year. So forecasting AI is notoriously difficult both ways. Maybe the CEOs are right? Or maybe mathematical competitions are easier to conquer than real-world complexity?
Dissent in the ranks
Not every tech leader is drinking the superintelligence Kool-Aid either. Microsoft’s AI chief Mustafa Suleyman has called the pursuit of superintelligence “absurd,” while Salesforce CEO Marc Benioff recently described the AGI hype as a form of “hypnosis.” That’s pretty strong language coming from people who are deeply embedded in the AI world. It suggests there’s significant internal debate about how realistic these timelines really are.
Meanwhile, some experts who do believe superintelligence is coming aren’t exactly celebrating. More than 100,000 people, including Apple co-founder Steve Wozniak and AI “godfathers” Geoffrey Hinton and Yoshua Bengio, signed a statement calling for a prohibition on superintelligence development until certain conditions are met. So even among believers, there’s serious concern about what happens if we actually achieve this goal.
check”>The reality check
Look, the experts aren’t saying AI won’t be transformative – they absolutely believe it will be the “technology of the century” akin to electricity by 2040. They predict AI will provide daily companionship for about 15% of adults and assist in 18% of U.S. work hours by 2030. That’s still massive change, just not the overnight revolution the CEOs are promising.
So what’s really going on here? I think we’re seeing a classic case of Silicon Valley optimism meeting real-world constraints. Tech leaders have every incentive to talk up rapid progress – it drives investment, attracts talent, and keeps their companies in the headlines. But the people actually building these systems know that turning research breakthroughs into reliable, scalable technology is the hard part. The gap between lab demonstrations and real-world deployment is where most ambitious timelines go to die.
