Jul 16, 2017 · 29m · a16z
16 Questions About Self Driving Cars
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 an Andreessen Horowitz (a16z) presentation, Frank Chen explores the future of autonomous transportation, detailing sixteen key technical, business, and social questions shaping the self-driving vehicle revolution.
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 0.0 Guest teaching 6.1 Guest disagreement 2.0 The host pushing back 0.0
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 26:15 Relegating human driving to Legoland
Frank Chen forcefully rejects the argument for human driving rights on public roads, telling driving enthusiasts that human drivers simply aren't needed on the road and should go drive at Legoland.
Hardest push from the host ▶ 12:35 Dismissing reliance on V2X radios for initial modelsChen pushes back against optimistic industry reliance on vehicle-to-everything communication, flatly stating he would not depend on V2X deployment for first-generation model releases due to protocol and security issues.
Biggest teaching moment ▶ 15:58 Stanford research on automated human etiquette learningThe Stanford Lead Researcher explains how machine learning models infer unwritten human social rules and spatial etiquette directly from observational video data to navigate human spaces.
The host holds their own ▶ 7:15 Detailed breakdown of computational power envelope trade-offsChen demonstrates deep technical knowledge of hardware design constraints by breaking down how trunk supercomputers (50W to 500W) create mileage drag on electric and gas vehicles.
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 Cars? Market Scale Analysis | 0 | 6 | 1 | 0 | In this solo presentation monologue, the host is absent. Frank Chen educates the audience on the market scale of the automotive industry compared to tech giants like Apple and breaks down the SAE 6-level autonomous classification framework. | |
| Question 2: LiDAR vs. Stereo Camera Vision Systems | 0 | 6 | 2 | 0 | Frank Chen explains the technical tradeoffs between solid-state LiDAR sensors and dual stereo camera systems. He notes how plummeting LiDAR component costs challenge Tesla's camera-only approach. | |
| Question 3: Pre-Computed HD Maps vs. Building Maps on the Fly | 0 | 6 | 2 | 0 | The speaker contrasts pre-computed HD mapping requirements with real-time computational mapping. He outlines how onboard supercomputers strain the power envelope and impact electric vehicle driving range. | |
| Question 4: Software Blend - Deep Learning vs. Deterministic Robotics Rules | 0 | 6 | 2 | 0 | Chen details the architectural debate between end-to-end deep learning neural networks and traditional deterministic robotics rule algorithms for path planning and motion execution. | |
| Question 5: Real-World Testing vs. Virtual World Simulation | 0 | 6 | 2 | 0 | The segment covers virtual world simulation training in gaming engines like Grand Theft Auto alongside V2X communication obstacles like the T-bone collision scenario. | |
| Question 7: Eliminating Traffic Lights and Four-Way Stops | 0 | 7 | 1 | 0 | Chen showcases how smart intersection traffic management could eliminate traffic lights and plays Stanford research footage on learning human social etiquette for autonomous Jack Robots. | |
| Section Intro: Business Questions & Question 12: Who Will Win? | 0 | 6 | 2 | 0 | Chen pivots to business dynamics, predicting whether incumbent automakers, Silicon Valley pioneers, or Chinese developers will dominate as car ownership shifts toward Transportation as a Service. | |
| Question 11: How Will Insurance and Liability Change? | 0 | 7 | 3 | 0 | Chen unpacks insurance liability paradoxes when hacking garage doors and self-driving cars, citing statistics that 24 out of 25 fatal crashes stem from human driver error. | |
| Question 10: When Will Human Driving Become Illegal? | 0 | 6 | 4 | 0 | Chen forcefully dismisses human driving as dangerous, provocatively suggesting human driving enthusiasts be relegated to Legoland while highlighting induced demand and urban space reconfiguration. | |
| Question 16: Timeline and Adoption Curves for Autonomous Vehicles & Conclusion | 0 | 5 | 1 | 0 | Chen concludes by listing varied public adoption timelines from major industry players spanning 2018 to 2040. |
Statements from this episode (23)
a16z Thesis: Every Moving Object Will Eventually Become Autonomous
“Our thesis is actually that everything that moves will eventually go autonomous. So if there's a reason for a plane or a truck or a toy or a shopping cart to move by itself and get from destination A to B, it will because the cost will collapse and It'll be us…”
Chen: Auto Market Revenue Exceeds Smartphones Despite Fewer Unit Sales
“But look at cars. Much higher revenue, fewer units, but it is a massive, massive market.”
Chen: GM Offers $2,500 Driver-Assist Package on $17,000 Cars
“That is a GM car. Just took it off their website. It's a 17,000 dollar car. For 2500 dollars, you can add self-driving capabilities. Emergency braking, lane keep adaptive cruise control, right?”
Chen: Handoff UX is a Major Challenge for Semi-Autonomous Vehicles
“Having a mixed mode, where sometimes a car drives itself and sometimes you drive it, is how do you design that user experience? How do you get the car to say, hey, I'm uncertain about what to do in this situation. You need to take over. And if you look at the …”
Chen: Solid-State LiDAR Costs Will Drop From $75,000 to $250
“The good news is we're headed rapidly towards making that thing solid state. In other words, no moving parts. It's going from 75,000 dollars to 250 dollars.”
Chen: Most AV Experts Require LiDAR, With Tesla as Key Outlier
“Most people that I talk to think definitely LiDAR. There's a couple of big outliers. I think Tesla is trying a system without LiDAR, but we'll see.”
Chen: Consumer Maps Lack Resolution Required for Autonomous Vehicles
“But even as amazing as that is, it's not quite enough for a car that wants to drive itself to navigate the world.”
Chen: Only Three Mapping Companies Remain Globally as of 2017
“There's three mapping companies left on the planet.”
Chen: Real-Time Map Processing Degrades EV Range and Fuel Efficiency
“So the supercomputer that's going in the trunk of your car needs to be even more super, and that has range implications if you're building an electric vehicle. It has gas mileage implications if you're building a gas-powered vehicle, right?”
Chen: NVIDIA Advocates End-to-End Deep Learning for Self-Driving Cars
“NVIDIA would have you believe that the path to get to a self-driving car is basically deep learning end-to-end.”
Chen: Self-Driving Cars Will Blend Deep Learning and Rule-Based Code
“And so I think we're going to see some blend of these techniques, but it's a tribute to the power of deep learning and how effective those techniques that we're even having this conversation that you could build an end to end deep learning car.”
Chen: Mercedes-Benz Installing V2V Radios in S-Class Cars by 2019
“Mercedes-Benz has started putting V to V radios in their S series. I think that comes 2019. Maybe a little earlier.”
Chen: Autonomous Vehicles Will Automatically Learn Local City Driving Conventions
“And so the car manufacturers wouldn't have to create 300 versions, one for each city. They would ship it with learning algorithms that could learn the driving conventions of that city.”
Chen: China Aims to Publish the Most Deep Learning Papers Globally
“And then to keep an eye on is the list of Chinese manufacturers who are very aggressively pursuing this space. By the way trying to publish more deep learning papers than any other country on the planet.”
Chen: TaaS Shift Will Turn Automakers Into B2B Suppliers Like Boeing
“And if that happens, then the car companies become B to B providers, just like Boeing and Airbus. They sell to fleet managers and your consumer loyalty is to the fleet provider. Not to the car manufacturer.”
Chen: Automakers Engineer Car Door Slam Sounds to Evoke Ownership
“The sound that the car door makes when you slam and shove, right? That is an engineered sound to make you feel like that's my car.”
Chen: Autonomous Vehicle Insurance Will Price Algorithm Safety, Not Demographics
“In a self-driving world, eventually there will be no drivers, and so, like, your demographics are irrelevant. So your premium prices might become solely a function of how effective the autonomous algorithms are.”
Chen: Autonomous Fleets Will Reduce Accidents but Raise Individual Repair Costs
“We're pretty confident that when we get to a fully autonomous fleet, we'll have fewer accidents, but maybe the cost of repair will go up because yeah, we got to repair that supercomputer in your trunk that figured out how to drive on the road.”
Chen: Human Error Accounts for 24 of Top 25 Accident Causes
“If you actually look at the top 25 causes of accidents, 24 out of 25 are human driver error related.”
Chen: Accident Rates May Rise During Mixed Human-Autonomous Transition Period
“In the short term, though, when we don't have full autonomous and we have a mix of self-driving and human drivers, the accident rate might actually go up.”
Chen: Society Must Deploy Autonomous Cars Immediately Despite Temporary Accident Increases
“Morally, from a social point of view, we should try to get self-driving cars on the road as soon as possible, even if we get this small tick up in new types of accidents that we've never seen before due to the interaction, because the self-driving algorithms w…”
Chen: Human Driving Should Be Banned Once Autonomous Vehicles Prove Safer
“So if it were true that the algorithms are demonstrably, measurably, statistically better than a human driver, then we should not let human drivers on the roads, and so if you wanted to go drive, then go to Legoland.”
Chen: Adding Highway Capacity Increases Commute Times via Induced Demand
“Every new commute modality, like when we introduce highways, commute times increase. You would have thought the opposite, but economists explain this as induced or latent demand, which is when you add freeway miles, you just get more drivers. Congestion doesn'…”