April 14th, 2019 by Paul Fosse
Screen Capture from https://youtu.be/dEv99vxKjVI
It this article, I’ll highlight the points Elon Musk and Lex Fridman made in a recent YouTube video on Artificial Intelligence and give you the timestamps for the different topics in the video. For more context, we also published an overview of the video on Friday.
Lex Fridman is a research scientist with a PhD in Computer Science working on human-centered artificial intelligence. You know who Elon Musk is. CleanTechnica has many articles on autonomous vehicles and I’ve written two myself on Tesla’s new neural network chip and its strategy compared to Waymo’s.
Screen Capture from https://youtu.be/dEv99vxKjVI
Lex Fridman has been in the news recently, as he published a paper concluding that Tesla drivers do a good job of remaining vigilant while monitoring their Teslas on Enhanced Autopilot.
0:00 — Introduction by Lex
3:15 — Elon states that in the future “Any car that doesn’t have autonomy will be as useful as a horse.”
3:50 — Autonomous cars are worth 5 to 10 times as much.
4:20 — Car’s display is a health check on the car’s perception of the outside world.
5:40 — Lex asks if Elon will show the computer’s uncertainty. Elon responds that there are two debugging views, one augmented reality where it draws boxes around objects and labels what it thinks they are and another more complicated view. They don’t think those debugging views would be useful to consumers.
7:45 — Lex asks if he should spend effort on the algorithms, the data, or the hardware. Elon states they have data coming from radar, 8 cameras, and 12 ultrasonic sensors in over 400,000 cars and they are just rolling out the fast hardware that took 3 years to develop.
10:10 — Talking about edge cases in driving, Elon clearly states, “All input is error” — meaning, if the customer has to disengage Autopilot for any reason, they analyze what needs to be improved.
14:50 — What Elon thinks is “really profound and we’ll be emphasizing at the upcoming investor day” is that the cars currently being produced will be capable of “full self driving” with only an over-the-air update. Of course, Elon has been saying this for close to 3 years and progress toward this goal was quite slow at first.
15:20 — You are buying an appreciating asset.
16:10 — People enjoy Autopilot today for highway driving. Tesla just needs to enhance it so it works on surface streets. The cars need to recognize lights, complex intersections, and complex parking lots.
17:45 — How long will we have to keep hands on the wheel? Elon thinks regulators will approve once probability of crashes, injuries, and death are 200% better than human drivers. I think Elon is being optimistic. I think regulators will want the system to be at least 10 times safer than a human and this will be a large battle over the next few years.
20:00 — Elon states that regulators follow the press and Tesla generates a lot of press. There are 40,000 deaths a year with cars, but every Tesla crash is a front page headline.
21:00 — Lex explains his study of 18,000 disengagements where he found drivers were paying attention.
21:40 — Elon thinks human vigilance will quickly become a moot point because the system is so much better than a human, the human’s input may actually hurt safety.
22:40 — Elon gives the example that human elevator operators have been totally replaced by automated elevators, and now it would hurt safety to allow human input to the elevator (beyond pressing the button).
24:20 — The rate of improvement in Tesla’s neural network is exponential.
25:20 — As opposed to competitors that tightly geofence their products, Tesla allows people to use Autopilot anywhere it can detect lines.
26:30 — In the future, people will look back and think it is crazy we allowed people to drive cars.
27:10 — Lex asks Elon about the recent incident where they tricked a car into going into the lane with oncoming traffic. Elon said that neural networks are basically matrix math, that hacking requires very sophisticated knowledge and that his team would find it easy to detect the matrix hacks.
28:20 — People have no idea how neural networks work.
28:50 — We are missing a few ideas to have general intelligence, but it will be upon us very soon.
29:50 — Game, Set, Match — Elon is confident Tesla is vastly ahead of the competition in building the world’s best self-driving car, but he acknowledges, “I could be wrong.”
31:00 — Can a machine love you? Elon states if you can’t tell, then it can. Perception=Reality. Elon channels Keanu Reeves in his famous “Red Pill or Blue Pill” scene of the 1999 blockbuster movie The Matrix when Lex asks Elon what he would ask the creator of the simulation, what one question would he ask? “What’s outside the simulation?”
Go here to watch the video for yourself!
It’s clear Elon feels he has a winning hand in using artificial intelligence to produce self-driving cars. It’s also easy to dismiss his claims as overconfidence, but having just come back from the Falcon Heavy launch, I think it is useful to consider what Elon has done in that industry. Elon is not a trained rocket scientist, but he has training in economics and physics and learned rockets on his own. He looked at what NASA had done over the last 40 years and confidently said he could make a rocket 10 times cheaper than the huge team at NASA. After 15 years in the industry, he has done that, and now his goal is more like 100 times cheaper instead of 10 times cheaper. Elon is frequently overly optimistic on the pace of progress, but is rarely wrong on his outrageous claims.
If you want to take advantage of my Tesla referral link to get 1,000 miles of free Supercharging on a Tesla Model S, Model X, or Model 3, here’s the code: https://ts.la/paul92237 (but as I have said before, if another owner helped you more, please use their link instead of mine).
Tags: Elon Musk, Lex Fridman, MIT, Tesla, Tesla autopilot, Tesla Full Self-Driving
About the Author
Paul Fosse I’ve been a software engineer for over 30 years, first working on EDI software and more recently developing data warehouse systems in the telecommunications and healthcare industry. Along the way, I’ve also had the chance to help start a software consulting firm and do portfolio management for several investment trusts. In 2010, I took an interest in electric cars because gas was getting expensive. In 2015, I started reading CleanTechnica and took an interest in solar, mainly because it was a threat to my oil and gas investments in my investment trusts. Tesla investor. Tesla referral code: https://ts.la/paul92237