With AI, work tends to move up a level
Investor Claire Zhou says AI judging other AI is her vision of AGI, and she bets on founders who adapt and decide well.
Fortune Magazine · This Creator Investor On The Truth About The AI Gold Rush | Term SheetClaire Zhou spent six years in investing, including as a partner at GSV, and was joining Lightspeed in a new dual role as creator and investor. She was asked what kinds of startups she would bet on for the next 10 or 20 years now that AI can create intelligence.
Work keeps moving to a higher level, she said. For a long time, she said, humans were manual computers, "just plugging wires in and out," and human alarm clocks. The machines took the repetitive task, and people moved to the judgment above it.
That is why people will still be needed because judgment is needed, in her view. In her telling, AI is now getting good enough to exercise a kind of judgment itself, to evaluate other AI systems, and that is how she thinks about the vision for artificial general intelligence, machines that can match people at most mental work.
If that sounds abstract, her day job makes it concrete. Zhou said she likes to bet at the pre-seed and seed stages, the earliest rounds of startup funding, because every founder faces so many decisions in a single day. The bet, as she put it, is on founders and how well they steer through the technology and, by compounding, choose well over and over.
She also looks at whether the team can adapt when a large AI lab or incumbent moves into its territory, in what she described as an exciting but uncertain time to build.
Questions people ask
Were there really human computers and human alarm clocks?
Yes. Before electronic computers, large groups of people, often called computers, did calculations by hand for astronomy, engineering, and the military. In industrial Britain, so-called knocker-uppers also went door to door to wake shift workers before alarm clocks were cheap and reliable.
What does AI judging other AI actually look like?
It means one AI system checks, scores, or supervises the work of another, rather than a person checking every output. AI developers already use this for tasks like rating answers, finding errors, and testing safety, because human review alone cannot keep up with the volume.
Does that mean jobs are safe from AI?
Not exactly. Zhou's point is that the type of work shifts upward toward choices, context, and responsibility. Some routine tasks can shrink or disappear, while new work grows around deciding what to build, which AI output to trust, and how to use it in a business.