Jan 2, 2019 · 37m · a16z
a16z Podcast | When Humanity Meets A.I.
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 episode of the a16z podcast, Sonal Chokshi and Frank Chen speak with Dr. Fei-Fei Li about artificial intelligence transitioning from theoretical laboratory research to real-world deployment. They explore compute hardware co-evolution, algorithm limitations, semi-autonomous safety ethics, and Dr. Li's humanistic approach to expanding AI diversity.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The host holds 16.5% of the talking time here. How this is scored →
The host as informed peer 5.3 Guest teaching 4.8 Guest disagreement 1.4 The host pushing back 2.5
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 5:34 Fei-Fei reframes chip tape-out premise
Fei-Fei directly corrects Frank's assertion that it is too early or risky to design specialized chips before winning algorithms are settled, explaining that algorithm and hardware R&D must happen concurrently.
Hardest push from the host ▶ 11:55 Sonal challenges AlphaGo creativity definitionSonal refuses Fei-Fei's framing of AlphaGo's optimal move-making as creativity, insisting on irrational, non-logical artistic creation like Jackson Pollock as the true benchmark.
Biggest teaching moment ▶ 7:01 Fei-Fei details full ML historyFei-Fei broadens the conversation beyond deep learning by educating the hosts on 60 years of AI foundation, explaining how non-deep statistical machine learning still powers most industrial applications.
The host holds their own ▶ 20:47 Frank analyzes explicit algorithmic choices in the Trolley ProblemFrank articulates a nuanced legal and technical distinction regarding split-second human reaction times versus calculated explicit decision-making in autonomous software.
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 |
|---|---|---|---|---|---|---|
| In Vitro vs. In Vivo AI & Drivers of AI Growth | 7 | 4 | 2 | 3 | Frank and Sonal demonstrate high technical expertise by bringing up specific hardware details like TPU development, ASIC tape-out costs, and parallelization. Fei-Fei clarifies that hardware and algorithm R&D must happen concurrently rather than sequentially, lightly reframing Frank's assertion. | |
| Machine Learning History & Limits of Deep Learning | 4 | 6 | 1 | 1 | Frank asks if deep learning is just a passing trend. Fei-Fei educates the hosts on the 60-year history of AI, covering symbolic logic, 30 years of non-deep statistical machine learning (SVMs, Bayesian nets), and the early roots of neural networks. | |
| Supervised Learning, Unsupervised Learning & AGI | 4 | 5 | 1 | 1 | Sonal correctly identifies the limitation of supervised learning compared to human child development. Fei-Fei expands on this, detailing why supervised paradigms fail for home robotics and differentiating task-driven AI from AGI. | |
| AI Creativity and Learning to Learn | 6 | 4 | 2 | 6 | When Fei-Fei cites AlphaGo's unexpected Go moves as AI creativity, Sonal explicitly pushes back, rejecting logical creativity in favor of irrational artistic expression like Jackson Pollock. Sonal references Kevin Kelly and artistic history to support her counter-position. | |
| In Vivo Real-World Deployment: Big Tech vs. Startups | 5 | 4 | 2 | 3 | Frank introduces Andrew Ng's critique of Tesla's Autopilot to challenge Fei-Fei's view on autonomous safety. Fei-Fei shares her perspective on consumer communication and machine ethics, while Frank and Sonal detail startup vs incumbent dynamics. | |
| Human-Machine Interaction & The Trolley Problem | 6 | 4 | 1 | 2 | Sonal demonstrates domain knowledge by referencing Nissan's staff anthropologist Dr. Melissa Cefkin and ethicist Patrick Lin. Frank details the algorithmic mechanics and legal liability implications of the Trolley Problem. | |
| Jackrabbit: AI Navigating Social Dynamics and Public Spaces | 5 | 5 | 1 | 2 | Sonal highlights Stanford's Jackrabbit project research on social navigation. Frank raises a sharp question about cross-cultural social norms across cities, prompting Fei-Fei to discuss incremental online learning. | |
| Humanistic AI & Bridging the Diversity Gap | 5 | 6 | 1 | 2 | Fei-Fei synthesizes existential AI anxiety and the CS diversity gap into a unified thesis on missing humanistic mission statements. Sonal contributes data on Stanford CS female enrollment and structural attrition across career stages. |
Statements from this episode (20)
Fei-Fei Li: AI Is Transitioning From Laboratories to Real-World Application
“But now going forward, we're entering what I call the AI in vivo time, which AI is entering real life.”
Fei-Fei Li: AI's Moment Is Driven by ML Algorithms, Big Data, and Hardware
“So the convergence of, I'd say mathematical foundations and statistical machine learning tools, the big data and the hardware is created this historical moment of AI.”
Fei-Fei Li: Edge AI inference chips offer huge market opportunity beyond GPUs
“GPUs are wonderful for training the deep learning algorithms, but I think there is still a lot of space in rapid testing or inference time Chips where it can be used in recognition, you know, in devices, in embedded devices.”
Fei-Fei Li: Dedicated AI Hardware Is Early as Algorithms Haven't Matured
“I think we're still in the exploratory stage because the algorithms are not matured enough yet.”
Frank Chen: Taping out a custom silicon ASIC costs $50 million
“It's fifty million dollars to tape out.”
Fei-Fei Li: Many Industry Applications Rely on Non-Deep Machine Learning
“In fact, many, many industry applications today still use some of the most powerful machine learning algorithms that are not deep.”
Fei-Fei Li: Deep Learning Originated in the 1960s and 1970s
“Deep learning is not the newest. It's actually developed in the sixties, seventies by people like Kunihiko Fukushima, then carried out by Jeff Hinton and Yang Lecun and their colleagues.”
Fei-Fei Li: Deep learning cannot yet enable interactive observational robot learning
“You want to just, you know, like show and talk about what tasks there is and have the robot observe and learn. That kind of training scenario we cannot do in deep learning yet.”
Fei-Fei Li: How to achieve artificial general intelligence remains largely unknown
“Specific task-driven applications are part of AI and important, but there is also the AGI, artificial general intelligence of reasoning, abstraction, Communication, emotional interaction, understanding of intention and purpose, formulation of knowledge underst…”
Fei-Fei Li: AlphaGo's unexpected moves against Lee Sedol demonstrate AI creativity
“If you look at the four or five matches of AlphaGo, there were multiple moments when AlphaGo made a movement. Master Li Sado was really surprised. And if you look at the Go community, people were just amazed by the kind of Creativity AlphaGo has in terms of ma…”
Fei-Fei Li: No existing AI math framework enables emotional and intuitive creativity
“The kind of creativity we're talking about blending our logical thinking, emotional thinking, and just, you know, intuitive thinking. And I haven't seen today's any work that builds on the kind of mathematical formulation that would enable that.”
Frank Chen: George Hotz built a self-driving car single-handedly at Comma.ai
“George at Comma.ai who built a self-driving car by himself, like one person.”
Fei-Fei Li: Google holds early self-driving advantage in data and algorithms
“Companies like Google, even though they didn't have cars at the beginning, they had algorithms. They started this early, so they now have both data and algorithms.”
Fei-Fei Li: I would never put my kids in early Tesla Autopilot
“So when Tesla's autopilot came out, I watched some of the YouTube videos. As a mom, I would never want to put my kids or myself into those cars.”
Sonal Chokshi: Nissan employs a full-time anthropologist for autonomous vehicle design
“Nissan has a, an anthropologist on staff, Dr. Melissa Shefkin, or Shefkin, I forgot how to pronounce her last name, but She's an anthropologist whose full-time job is to study these issues in order to build it into the actual design, and it's not just like sof…”
Fei-Fei Li: Human brain and motor speed are drastically slower than transistors
“Compared to computers, humans are extremely slow computing machines. The information transfer in our brain is very slow compared to transistors. And add on top of that, our motor system You know, from our brain to our muscles is even slower.”
Fei-Fei Li: Naive pedestrian yielding rules cause autonomous robots to stall
“We tested that. If we do that, the robot will never go anywhere because in a very crowded space, there's always people. If the robot just follows the most naive rule of I'll yield to people all the time, the robot would just be sitting there from the starting …”
Fei-Fei Li: Autonomous navigation robots currently require location-by-location training
“As of now, we have to train them Location by location. We have to gather data.”
Fei-Fei Li: Stanford CS undergrads are 25% to 30% women
“It's about 25 to 30% in undergraduate that we have women, and then this thing just goes down as you Oh, it goes down as you go higher up.”
Fei-Fei Li: Humanistic framing statistically increases high school girls' AI interest
“We conducted a rigorous evaluation project on this hypothesis. Can humanism increase the interest in AI? And we found a statistically significant difference From the beginning to the, to before and after for these girls thinking. And that particular paper is p…”