Jan 2, 2019 · 27m · a16z
a16z Podcast | Machine Intelligence, from University to Industry
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this a16z summit panel, host Frank Chen and guest Cameron Schuler explore how artificial intelligence transitions from university labs to commercial applications. They cover reinforcement learning through game environments, the challenges of academic brain drain, organizational models for AI adoption, and the societal and ethical implications of machine intelligence.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The host as informed peer 4.9 Guest teaching 4.3 Guest disagreement 1.1 The host pushing back 2.3
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 11:10 Cameron rejects big tech corporate vision thesis
Cameron counters Frank's pushback directly, noting that only a tiny handfull of companies fit that model and that gaming companies remain strictly bottom-line driven.
Hardest push from the host ▶ 10:37 Frank presents corporate R&D counterargumentFrank directly challenges Cameron's warning about talent drain by arguing that visionaries like Zuckerberg and Page are better positioned to fund long-term scientific research than universities.
Biggest teaching moment ▶ 2:20 Cameron breaks down poker versus board gamesCameron educates the host on imperfect information environments, explaining that poker requires inferring obfuscated data compared to perfect-information games like Chess or Go.
The host holds their own ▶ 19:03 Frank shares internal tech stack data from FacebookFrank steps out of his moderator role to demonstrate deep technical domain knowledge, citing Facebook's FB Learner Flow platform and specific developer metrics.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Why AI Researchers Focus on Games | 6 | 5 | 1 | 5 | Frank demonstrates solid domain knowledge by citing specific achievements like Atari, Checkers, Poker, and AlphaGo, while challenging Cameron on whether game-based AI is just a toy. Cameron explains human learning paradigms, petri dishes for decision-making under ambiguity, and how poker differs fundamentally from board games due to imperfect information. | |
| Academic-Industry Collaboration Models | 3 | 4 | 0 | 1 | Frank asks exploratory questions regarding Amy's corporate partnerships, funding models, and past operational successes. Cameron educates the host on Canadian university IP negotiation flexibility and how internal corporate data science teams often feel threatened by external machine learning experts. | |
| Tech Giants and Academic Talent Drain | 5 | 5 | 2 | 4 | Frank raises the issue of big tech firms like Uber and Google draining university computer science talent. Cameron elaborates using a football analogy ( millions play, few pros ) to explain the difference between using ML tools and doing foundational research, emphasizing the risk short-term corporate thinking poses to long-term innovation. | |
| Corporate R&D Vision vs. Innovation Constraints | 7 | 6 | 3 | 6 | Frank delivers a strong counterargument that visionaries like Zuckerberg and Page are capable of funding long-term fundamental R&D. Later, Frank challenges Cameron on Geoff Hinton's controversial stance regarding replacing radiologists, noting the black box nature of deep learning. | |
| Reinforcement Learning Concepts with Rich Sutton | 6 | 6 | 0 | 1 | Frank prompts Cameron to discuss Rich Sutton, interjecting with sharp historical context about Marvin Minsky and symbolic AI derailing neural network progress for a generation. Cameron explains RL foundations including on-policy/off-policy learning and temporal difference learning. | |
| Audience Q&A: Commercializing AI & Executive Education | 8 | 2 | 0 | 2 | In response to an audience question about industrializing AI, Frank explicitly breaks moderator decorum to share deep operational insights, highlighting Facebook's FB Learner Flow workflow system enabling 25 percent of all developers to write ML code. | |
| Audience Q&A: Corporate R&D, Genetic Algorithms & Self-Improving AI | 7 | 5 | 1 | 1 | Cameron critiques C-suite risk-aversion and lack of true innovation cultures. Frank highlights literature recommendations and shares details about Google AI agents autonomously discovering emergent encryption to illustrate self-improving code. | |
| Audience Q&A: AI Ethics and Superintelligence | 2 | 6 | 3 | 1 | An audience member asks about superintelligence ethics and compares human-AI power dynamics to human-ant relations. Cameron draws on Rich Sutton's views regarding treating AI as members of society rather than indentured servants, responding to the ant comparison with a human-monkey intelligence analogy. | |
| Podcast Session Conclusion | 0 | 0 | 0 | 0 | Brief session wrap-up and closing thank yous. |
Statements from this episode (15)
Schuler: 20-25% of DeepMind staff at acquisition were University of Alberta students
“When DeepMind got bought, half the people there were actually Canadian-trained, and roughly 20 or 25% were our students.”
Schuler: 45% of citations in DeepMind's AlphaGo paper were from U of Alberta
“AlphaGo, I think it was roughly 45% of the research cited on their AlphaGo paper came from the University of Alberta.”
Schuler: U of Alberta AI algorithms run four times faster than DeepMind's
“We're actually meeting some of the deep mind algorithms that they have right now. We're roughly four times faster, I believe.”
Schuler: Yoshua Bengio deliberately chose to remain in academia over industry
“There's another Canadian professor named Yoshua Bengio at University of Montreal, who's consciously decided not to leave for this, for that very reason.”
Schuler: Google will eventually be disrupted in advertising
“From my perspective, their risk is, 80% of the revenue is generated by advertising. They are going to get disrupted in that at some point.”
Schuler: Poaching academic AI talent risks leaving no experts to train students
“I think there's a huge problem where there's a lot of risk that to get the people that can solve the difficult problems, there won't be anyone to train them.”
Schuler: Gaming companies like Electronic Arts could win in AI
“I think the people who could really win at AI are the games companies, the electronic arts and groups like that.”
Chen: Deep learning will outperform trained radiologists within five years
“It takes five years to train a radiologist, and in five years, deep learning will get better results than a trained radiologist, so we should stop training them right now.”
Schuler: Rich Sutton's textbook is computer science's most influential publication
“According to the Alan Soufre AI Semantics Scholar, he's the most highly cited researcher in reinforcement learning, 11th most influential researcher in all of competing science. And his textbook on reinforcement learning was ranked as the single most influenti…”
Schuler: Distributed computing enabled practical deep learning applications
“So the industrial project, even in capital markets, a lot of stuff we've done, you could have done 20, 25 years ago, deep learning, right? It was really distributed computing that made the big difference in that.”
Chen: 25% of Facebook developers write deep learning code
“If you look at something like FB Learner Flow, which is Facebook's Automation workflow system for artificial intelligence. They've gotten it so good that 25% of their total software developer universe is writing deep learning. 25%, right?”
Schuler: Aside from tech giants, major corporations lack an AI strategy
“Pretty much everyone has missed the boat on an AI strat, like everybody. There's a handful of companies, Google, Microsoft, Facebook that are doing it, but Everybody, I mean, I talk to companies with 15, twenty billion in revenue, they go, we don't have an AI …”
Schuler: Genetic algorithms will never succeed in capital markets
“If you take a look at capital markets, the domain space is, I mean, the dimensionality of data is so huge that a genetic algorithm is never going to get there.”
Schuler: Heavy backend AI processing won't work for mobile-first markets
“The way I see the world moving certainly is more mobile, and if you are relying on some big back end where it has a lot of processing power, it's not going to work, so it's really about training systems and bringing them into mobile and things like that, espec…”