
Under three minutes on what the AI leveling studies measured. They score output, the writing and the polish, and for some people thinking and writing are two different skills. AI removes that translation tax. My hypothesis, labeled as one: part of the leveling is AI un-masking people whose thinking was strong all along.
Narrated in my own voice, cloned with ElevenLabs. The dot field in the video is an illustration, with no data behind it.
Transcript
For most of my life, my writing read like a machine. Structured, competent, a little stiff. I used to think that was a flaw. Turns out it was compensation. Words are my second language. My first is pictures. So I build sentences the way I build systems, and so does AI. My family thinks it's hilarious that the machine writes like me, and not the other way around.
You've seen the headline. Give people AI, and the weakest performers gain the most. At BCG, the below-average consultants improved 43%. AI, the great leveler. I'm not here to fight the data. I want to know what it measured.
Every one of those studies scores output. The writing, the polish. But thinking and getting it onto a page are two different skills. Some people see the answer clearly and pay a translation tax on every sentence. Score them on output, and they look average.
Look closer at the Science paper. AI mostly substituted for effort, and moved the work toward ideas, away from rough drafts. It ate the drafting. The exact tax. That low score? For some of those people, it was a toll.
Talent research has seen this channel for decades. Vanderbilt followed gifted 13-year-olds for 30 years. Spatial reasoning, holding a whole system in your head, predicted their patents and papers on top of math and verbal scores. And standard talent searches miss about half of the top 1% in it.
So here's my hypothesis, and I'm calling it that on purpose. Part of the leveling is AI lifting people to adequate. The other part is AI un-masking people whose thinking was strong all along. Nobody has measured conception apart from articulation yet, so the averages cannot tell the two apart.
Everyone agrees AI multiplies a mess. Point it at someone who sees the whole board, and it multiplies the signal. You don't need the study to act. Stop treating polish as capacity. If you run a team, look again at the person whose memos were always a little rough, and whose instincts were always a little right. The tax that hid them just dropped to zero.
Most companies will spend the next two years admiring the productivity bump. Underneath it, the people they filed as average are the most mispriced talent on the board. That window closes the moment everyone else notices. Who on your team was unmasked recently? Let's talk.
Sources
Noy, S. & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science.
Dell'Acqua, F. et al. (2023). Navigating the Jagged Technological Frontier. Harvard Business School / BCG working paper.
Kell, H., Lubinski, D., Benbow, C. et al. (2013). Creativity and technical innovation: spatial ability's unique role. Psychological Science.
Vanderbilt University news release, Jul 15 2013 (Lubinski: talent searches miss about half of the top 1 percent in spatial ability).