Oct 16, 2017 · 32m · 20vc
20VC: Founders Fund's Chief Scientist on Why AI Is Mostly A Scam, Why The Value of Large Datasets Is Mostly Overplayed & The Societal Effects of 4m Truck Drivers Being Unemployed with Aaron Vandevender
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Founders Fund Chief Scientist Aaron VanDevender joins host Harry Stebbings to discuss evaluating deep-tech investments, deconstructing artificial intelligence and dataset moats, and addressing the socio-economic impacts of workforce automation.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 24.4% of the talking time here. How this is scored →
Harry as informed peer 2.0 Guest teaching 5.2 Guest disagreement 3.2 Harry pushing back 1.7
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 6:56 Calling AI a rebranding scam
Aaron forcefully dismisses modern AI as a historical sequence of actuarial science rebranding efforts powered by AWS computing.
Hardest push from Harry ▶ 8:34 Pushing back on AI hype dismissalHarry explicitly challenges Aaron's assertion by arguing that AI is necessary for big data to actually fulfill its promise.
Biggest teaching moment ▶ 13:40 Debunking dataset moatsAaron systematically dismantles the belief in massive data moats by showing how diminishing returns kick in early and how smaller datasets force better algorithmic thinking.
Harry holds his own ▶ 11:17 Connecting ambiguity to AGIHarry demonstrates strong domain context by questioning whether tolerance for ambiguity is simply general artificial intelligence framed on a multi-decade timeline.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| From Quantum Physics to Venture Capital | 1 | 3 | 1 | 1 | Harry opens with standard interview questions about Aaron's background and the chief scientist role at Founders Fund. Aaron politely outlines his responsibilities and explains his heuristic for distinguishing genius from crazy ideas. | |
| Why AI and Big Data Are Largely Rebranded Hype | 3 | 7 | 7 | 4 | Aaron delivers a contrarian critique, labeling most current AI as a mere rebranding of actuarial science using cloud computing. Harry pushes back with the common VC view that AI enables big data to finally deliver value, but Aaron dismisses big data as mostly hype. | |
| True AI Breakthroughs and the Myth of Dataset Moats | 3 | 7 | 5 | 3 | Aaron explains that true AI breakthroughs require handling ambiguity rather than pattern matching, and refutes the concept of dataset moats. Harry raises standard industry concerns about startup access to data, but Aaron explains how smaller datasets force superior algorithmic design. | |
| Societal Impacts of Autonomous Trucking and Wealth Redistribution | 2 | 6 | 2 | 1 | Aaron provides a deep analysis of autonomous trucking, framing it as both a goods distribution system and a wealth distribution network for local communities. Harry accepts the framing and asks how society should handle the displacement of millions of drivers. | |
| Evaluating Universal Basic Income, Technology, and Urban Density | 2 | 6 | 3 | 1 | Aaron critiques universal basic income as a blunt instrument that misses the core issue of human productivity. He argues that digital technologies actually amplify the value of physical proximity, advocating for increased urban density over direct checks. | |
| Quick Fire Round | 1 | 2 | 1 | 0 | Harry guides Aaron through rapid-fire questions covering favorite books, quantum computing, Slashdot, and portfolio investments. Aaron answers cooperatively and succinctly without any conflict. |
Statements from this episode (12)
VanDevender: Atomic electron microscopes require quantum computing techniques
“So we built single atom resolution electron microscopes, which turns out to be very similar to building quantum computers. And that you're manipulating the quantum states of individual atoms, but use that for biotechnology purposes.”
VanDevender: Drilling into specific technical details tests a founder's true depth
“Even if you don't necessarily yourself have the complete picture by just, you know, drilling down on specific aspects, you know, whether that person has, like, has all the answers and can go as deep as you want to go is a really good sign.”
VanDevender: Most products marketed as AI are rebranded actuarial science
“Most of the things that call themselves AI out there are really just a rebranding.”
VanDevender: Deep learning is just linear algebra at industrial scale
“At the core of it is just this matrix multiplication operations. It's, you know, you're just sort of doing linear algebra at industrial scale. We haven't gotten to the point for most of these things or most of the things that call themselves AI, where you're r…”
VanDevender: Most 'big data' datasets are only megabytes or gigabytes
“Most of the things that are sort of big data are actually not that big. You know, you would like to, you'd like to think of these petabytes and exabytes floating around there and coming up with really amazing insights that could never have been seen. But a lot…”
VanDevender: Defensive data moats are oversold and mostly noise
“Most of the data sets out there or the defensibility data sets out there I think are oversold. There is this very, it's a very compelling, very tantalizing notion that, oh, I just have so much data, and nobody else has it, and therefore it must be special, but…”
VanDevender: Smaller datasets force self-driving car companies to build better algorithms
“Even if you have a much, much smaller data set than Google, you can still do a really good job of developing a self-driving car, and in fact, it almost forces you into a better position, because if you have complete knowledge of the entire street grid, you can…”
VanDevender: US trucking operates as both goods and wealth distribution
“The whole trucking system does serves two purposes at the same time. One, it's a goods distribution system. So we load things on the trucks, we ship them around. That's how we get all the stuff that we buy at the store, but it's also a wealth distribution syst…”
VanDevender: Autonomous trucks will improve delivery but erode rural wealth distribution
“So if you replace the trucks with self-driving trucks, you end up with a much better goods distribution system, which is positive, but the trucking system is a much worse wealth distribution system.”
VanDevender: Robot maintenance jobs will pay more but employ fewer people
“There will be those, some of those jobs created and they will be better jobs than trucker jobs. Those people will be more productive. They'll get paid more. They won't be on the road all the time. They'll be home for dinner with their kids. So it'll be a bette…”
VanDevender: VR will repeat the mistaken belief that physical location doesn't matter
“I think we're sort of poised to make that same mistake again with virtual reality.”
VanDevender: Digital platforms increase the economic value of physical proximity
“The lesson has been that actually those digital technologies have actually made proximity more valuable. If you think about like where, where a lot of that value to people has come from, it's from companies like Lyft, like Postmates, a Craigslist, Amazon Prime…”