Jon Krohn: 00:00 This is episode number 942 on LEAP, the longitudinal expert AI panel. Welcome back to the SuperDataScience Podcast. I am your host, Jon Krohn. Today I’m distilling what hundreds of experts think is actually likely to happen with AI near term milestones, longer term transformational impact, and what all that means for work and for infrastructure. The backbone here is a new effort called the Longitudinal Expert AI Panel, or LEAP for short, a monthly survey of over 300 experts and super forecasters general experts in forecasting the future across a range of subjects. And LEAP is designed to replace hand wavy claims with specific testable predictions. As a interesting kind of relatively near term example. By 2030, the median LEAP forecast says AI will assist roughly 18% of US work hours. That’s a huge jump from low single digits, about 2% today in the same timeframe. With all of that extra AI assisting people, the panel expects AI training and deployment to consume about 7% of US electricity more than all data centers combined in the US today on historical impact, more longer term, LEAP asks experts to score AI on Nate Silver’s technological Richter scale.
01:28 That’s a 10 point scale where the higher you are on the scale, the more crazy transformational the technology is. By 2040, the experts from LEAP modally, so their most frequent prediction is that they expect AI to be a technology of the century. That’s a level eight out of 10 on Nate Silver’s technological Richter Scale. Think the level of electricity or the automobile and these same experts from LEAP assign about a one in three probability that AI reaches level nine by 2040 in 15 years, and that’s a tier. Level nine is a tier reserve for technologies like the printing press or the industrial revolution that changed the course of human history. That would be pretty mega indeed. So why believe these numbers? Well, a key strength of LEAP is short horizon forecasting, asking questions that resolve within minutes so we can identify which forecasters are most accurate and weigh them more over time.
02:30 That matters because progress is lumpy and sometimes outpaces expectations. As a vivid example on Frontier Math, a benchmark of expert level problems, Google’s Gemini 2.5 deep think set a new record last month in October with 29% on tiers one to three of the Frontier Math benchmark vaulting past many earlier guesses about near term performance. We’re also seeing frontier models pick up Practical laboratory know-how in sensitive domains, which is kind of a recent virology capabilities test found. Top models outperforming PhD level virologists on troubleshooting complex wet lab tasks. Reporting around those results highlights open AI’s O three as a top performer and has already sparked calls for stronger guardrails because of how dangerous it is to be able to manufacture viruses. The takeaway, however, probably shouldn’t be panic, it’s the capabilities are advancing in ways that policy and governance have got to keep pace with. Now as you’re listening to these kinds of transformative technologies, 18% of US work hours expected to be assisted by AI in just five years and a one in three chance of a level three out of 10 crazy transformative technology at the level of the printing press or industrial revolution just 15 years from now.
03:55 What does all of that mean for jobs? Large organizations like the IMF estimate, about 60% of jobs in advanced economies are exposed to AI tasks, roughly half likely complimented and half of those jobs at risk of substitution LEAPs. Numbers suggest slower growth in some white collar categories rather than a sudden collapse, which aligns with a policy focus on re-skilling task redesign and AI augmented workflows rather than one for one job replacement for leaders, and you listeners, whoever you are, for me, the message is clear. Treat 18% AI assisted work by 2030 as a planning baseline as an individual or as someone managing people invest in skills and process change. Now assume material pressure on power and infrastructure and watch the short horizon signals. So because LEAP is monthly, you can watch it and see how it’s performing against benchmarks and validation cycles.
04:59 You can follow this LEAP trend over time, not just the sensationalist headlines that you see out there frequently in news and on podcasts. I’m sure as well forecast will update, some will overshoot, others will be conservative, but the direction of travel is unmistakable. AI’s impact is going to be broad, accelerating and increasingly measurable. Giddy up. Alright, so if you want to read more on any of the key things that I talked about in this episode, like the longitudinal Expert AI panel or Nate Silver’s Technological Richter Scale, I’ve got links for you in the show notes. That is it for today’s episode. I’m Jon Krohn and you have been listening to the SuperDataScience Podcast. If you enjoyed today’s episode where you know someone who might consider sharing this episode with them, leave a review of the show on your favorite podcasting platform. Tag me in a LinkedIn post with your thoughts and if you haven’t already subscribed to the show, most importantly, I just hope you’ll keep on listening. Until next time, keep on rocking out there and I’m looking forward to enjoying another round of the SuperDataScience Podcast with you very soon.
