Elon Musk: Tesla Autopilot | Lex Fridman Podcast #18

1. 0:00 – 4:01

   Why autonomy matters: the two revolutions in cars

   1. LFLex Fridman0:00

      The following is a conversation with Elon Musk. He's the CEO of Tesla, SpaceX, Neurolink, and a co-founder of several other companies. This conversation is part of the Artificial Intelligence podcast. This series includes leading researchers in academia and industry, including CEOs and CTOs of automotive, robotics, AI, and technology companies. This conversation happened after the release of the paper from our group at MIT on driver functional vigilance during use of Tesla's autopilot. The Tesla team reached out to me, offering a podcast conversation with Mr. Musk. I accepted, with full control of questions I could ask and the choice of what is released publicly. I ended up editing out nothing of substance. I've never spoken with Elon before this conversation, publicly or privately. Neither he nor his companies have any influence on my opinion, nor on the rigor and integrity of the scientific method that I practice in my position at MIT. Tesla has never financially supported my research, and I've never owned a Tesla vehicle. I've never owned Tesla stock. This podcast is not a scientific paper. It is a conversation. I respect Elon as I do all other leaders and engineers I've spoken with. We agree on some things and disagree on others. My goal is always, with these conversations, is to understand the way the guest sees the world. One particular point of disagreement in this conversation was the extent to which camera-based driver monitoring will improve outcomes, and for how long it will remain relevant for AI-assisted driving. As someone who works on and is fascinated by human-centered artificial intelligence, I believe that if implemented and integrated effectively, camera-based driver monitoring is likely to be of benefit in both the short-term and the long-term. In contrast, Elon and Tesla's focus is on the improvement of autopilot, such that its statistical safety benefits override any concern with human behavior and psychology. Elon and I may not agree on everything, but I deeply respect the engineering and innovation behind the efforts that he leads. My goal here is to catalyze a rigorous, nuanced, and objective discussion in industry and academia on AI-assisted driving, one that ultimately makes for a safer and better world. And now, here's my conversation with Elon Musk. What was the vision, the dream of autopilot when, uh, in the beginning, the big picture system level, when, uh, it was first conceived and started being installed in 2014 in the hardware and the cars? What was the vision, the dream?

   2. EMElon Musk2:49

      I wouldn't characterize it as a vision or dream, simply that there are obviously two massive revolutions in, in the, uh, automobile industry. One is the transition to elect- electrification, um, and then the other is autonomy. And, uh, it became obvious to me that, in the future, any, any car that does not have autonomy, uh, would be about as useful as a horse. Which is not to say that there's no use, it's just rare and somewhat idiosyncratic if somebody has a horse at this point. So, it's obvious that cars will drive themselves completely, it's just a question of time, and if we did not participate in the autonomy revolution, then our cars would not be useful to people, relative to cars that are autonomous. I mean, an autonomous car is arguably worth five to ten times more than a non- a car which is not autonomous.

   3. LFLex Fridman3:53

      In the long term.

   4. EMElon Musk3:55

      Depends what you mean by long term, but let's say at least for the next five years, perhaps 10 years.

2. 4:01 – 5:11

   Making Autopilot legible: why Tesla shows what the car “sees”

   1. LFLex Fridman4:01

      So, there are a lot of very interesting design choices with autopilot early on. First is showing on the instrument cluster, or in the Model 3, on the center stack display, what the combined sensor suite sees. What was the thinking behind that choice? Was there debate? What was the process?

   2. EMElon Musk4:20

      The whole point of the t- display is to provide a health check on the r- the vehicle's perception of reality. So, the vehicle's, uh, taking in information from a bunch of sensors, primarily cameras, but also radar and ultrasonics, uh, GPS, and so forth. And then, uh, that, that information is then rendered into c- vector space, uh, and that, you know, with a bunch of objects with pr- with properties, like lane lines and traffic lights and other cars. Um, and then in vector space, that is re-rendered onto your display so you can confirm whether the car knows what's going on or not by looking out the window.

   3. LFLex Fridman5:02

      Right. I think that's a extremely powerful thing for people to get an understanding, sort of become one with the system and understanding what the system is capable of.

   4. EMElon Musk5:11

      Mm-hmm.

3. 5:11 – 7:10

   Uncertainty and debug views: what to show (and what not to)

   1. LFLex Fridman5:11

      Now, have you considered showing more? So, if we look at the computer vision, you know, like road segmentation, lane detection, vehicle detection, object detection underlying the system, there is at the edges some uncertainty. Have you considered revealing the parts that, uh, the- the uncertainty in the system, the sort of more-

   2. EMElon Musk5:33

      The probabilities associated with, with say image recognition or something like that?

   3. LFLex Fridman5:36

      Yeah. So right now it shows like the vehicles in the vicinity, a very clean, crisp image, and people do confirm that there's a car in front of me and the system sees there's a car in front of me, but to help people build an intuition of what computer vision is by showing some of the uncertainty.

   4. EMElon Musk5:53

      Well, I think it's, uh, yeah, my car, I always look- look at the sort of the- the debug view, and there's, there's two debug views, uh, o- one is...... augmented vision, uh, where, which I'm sure you've seen, where it basically, uh, we, we draw boxes and labels around objects that are recognized. And then there's, uh, what we call the visualizer, which is basically a vector-based representation summing up, uh, the input from all sensors. That, that does, does not show b- any pictures, but it shows, uh, all of the ... it basically shows the car's view of, of, of the world in vector space. Um, but I think this is very difficult for people to kno- normal people to understand. They would not know what the heck they're looking at.

   5. LFLex Fridman6:39

      So, it's almost an HMI challenge to... the current things that are being displayed is optimized for the general public understanding of what the system is capable of.

   6. EMElon Musk6:48

      Yeah. It, like, if you've no idea what, how computer vision works or anything, you can still look at the screen and p- and see if the car knows what's going on. And then if you're, you know, if you're a development engineer or if you're, you know, if you're, if you have the development build like I do, then you can see, uh, you know, all the debug information. But those would just be, like, total gibberish to most people.

4. 7:10 – 10:22

   Resource allocation: data advantage and the FSD computer hardware bet

   1. LFLex Fridman7:10

      Right. What's your view on how to best distribute effort? So, there's three, I would say, technical aspects of autopilot that are really important. So, it's the underlying algorithms, like the neural network architecture, there's the data, so that to train on, and then there's the hardware development. There may be others, but ... so, look, algorithm, data, hardware. You on- you only have so much money, only have so much time. What do you think is the most important thing to, to, uh, allocate resources to? Or do you see it as pretty evenly distributed between those three?

   2. EMElon Musk7:44

      We automatically get fast amounts of data, because all of our cars have eight external-facing cameras and radar, and usually 12 ultrasonic sensors, uh, GPS obviously, um, and, uh, IMU. And so we, we basically have a fleet that has, um ... and we've got about 400,000 cars on the road that have that level of data. I g- actually, I think you keep quite close track of it, actually.

   3. LFLex Fridman8:15

      Yes.

   4. EMElon Musk8:16

      Yeah. So, we're, we're approaching half a million cars on the road that have the full sensor suite.

   5. LFLex Fridman8:22

      Yeah.

   6. EMElon Musk8:22

      Um, the ... so this is ... I, I'm n- I, I'm not sure how many other cars on the road have this sensor suite, but I would be surprised if it's more than 5,000, which means that we have 99% of all the data.

   7. LFLex Fridman8:36

      So, there's this huge-

   8. EMElon Musk8:37

      Um-

   9. LFLex Fridman8:37

      ... inflow of data.

   10. EMElon Musk8:38

       Absolutely. Massive inflow of data. And then we ... it's d- it's taken us about three years, but now we've finally developed our full self-driving computer, which can process, uh, an or- an order of magnitude as much as the NVIDIA system that we currently have in the, in the cars. And it's really just a ... to use it, you unplug the Nvi- NVIDIA computer and plug the Tesla computer in, and that's it. And it's, it's, uh ... in fact, we're not even qu- we're still exploring the boundaries of its capabilities. Uh, but we're able to run the cameras at full frame rate, full resolution, uh, not even crop the images, and, uh, it's still got headroom, even on one of the, the systems. The hard d- full, full self-driving computer is really two computers, two systems on a chip that are fully redundant, so you could put a bolt through basically any part of that system and it still works.

   11. LFLex Fridman9:30

       The redundancy, are they perfect copies of each other, or ...

   12. EMElon Musk9:33

       Yeah.

   13. LFLex Fridman9:34

       Oh, so it's purely for redundancy as opposed to an arguing machine kind of architecture where they're both making decisions. This is purely for redundancy.

   14. EMElon Musk9:41

       I- I think of it more like it's ... if you have, uh, a twin engine aircraft, um, commercial aircraft, the system will operate best if both systems are operating, but it's, it's capable of operating safely on one. So ... but a- a- as it is right now, we can just run ... we're h- we haven't even hit the, the, the e- the edge of performance, so there's no need to actually distribute functionality across both SOCs. We, we can actually just run a full duplicate on b- on, on each one.

   15. LFLex Fridman10:17

       So, you haven't really explored or hit the limit of this-

   16. EMElon Musk10:20

       We have not yet hit the limit, no.

5. 10:22 – 12:58

   Learning from edge cases: disengagements, interventions, and “all input is error”

   1. LFLex Fridman10:22

      So, the magic of deep learning is th- that it gets better with data. You said there's a huge inflow of data, but-

   2. EMElon Musk10:30

      Yeah.

   3. LFLex Fridman10:30

      ... the thing about driving, the really valuable data to learn from is the edge cases.

   4. EMElon Musk10:35

      Mm-hmm.

   5. LFLex Fridman10:36

      So, how do you ... I mean, I've, I've heard you talk somewhere about, uh, uh, autopilot disengagements being an important moment of time-

   6. EMElon Musk10:45

      Yes.

   7. LFLex Fridman10:46

      ... to use. Is there other edge cases or perhaps can you speak to those edge cases, what aspects of them might be valuable? Or if you have o- other ideas how to discover more and more and more edge cases in driving?

   8. EMElon Musk11:00

      Well, there's a lot of things that are learnt. There are certainly, uh, edge cases where, uh, say somebody is on autopilot and they, they take over, um, and then w- okay, that, that will ... that, that's a trigger that goes up to our system that says, "Okay, did the tri- they take over for convenience, or do they take over because the autopilot wasn't working properly?" There's also ... like, let's say we're t- we're trying to figure out what is the optimal spline for traversing an intersection.

   9. LFLex Fridman11:27

      Right.

   10. EMElon Musk11:27

       Um, then, th- then, uh, the o- the ones where there are no interventions and, and w- uh, are, are the right ones. So you then say, "Okay, when it looks like this, do the following." And th- and th- and, and then you get the optimal spline for a complex, uh, navigating a c- a complex, uh, intersection.

   11. LFLex Fridman11:45

       So, that's for ... so, so there's kinda the common case-

   12. EMElon Musk11:49

       Mm-hmm.

   13. LFLex Fridman11:49

       ... so you're trying to, uh, capture a huge amount of samples of a particular intersection, how, when things went right, and then there's the edge case where n- uh, as you said, not for convenience, but something didn't go exactly right.

   14. EMElon Musk12:02

       Yeah. Some- some- somebody took over ... somebody asserted manual control from autopilot.

   15. LFLex Fridman12:05

       Hmm.

   16. EMElon Musk12:06

       And r- really, like, the way to look at this is view all input as error.If the user had to do input, it does something ... All input is error.

   17. LFLex Fridman12:13

       That's a powerful line to think of it that way, 'cause it may very well be error. But if you wanna exit the highway, or if you want to, uh, it's a navigation decision that all autopilot is not currently designed to do, then the, the driver takes over. How do you know the difference?

   18. EMElon Musk12:28

       Yeah. That, that's gonna change with navigate and autopilot, which we, we've just released. Uh, and, and with that still confirmed. So the navigation, like, lane, lane change based, uh, y- yeah, like asserting control in order to change, do, do a lane change or exit a freeway or, or doing a highway interchange, uh, the vast majority of that will go away with, um, the release that just went out.

   19. LFLex Fridman12:48

       Yeah, so that, that I don't think people quite understand how big of a step that is.

   20. EMElon Musk12:54

       Yeah, they don't.

   21. LFLex Fridman12:56

       So-

   22. EMElon Musk12:56

       If, if you drive the car, then you do.

6. 12:58 – 14:03

   Major capability leaps: Navigate on Autopilot, lane changes, and traffic lights

   1. LFLex Fridman12:58

      So you still have to keep your hands on the steering wheel currently when it does the automatic-

   2. EMElon Musk13:01

      Mm-hmm.

   3. LFLex Fridman13:01

      ... lane change? What are ... So there's, there's these big leaps through the development of autopilot, through its history, and what stands out to you as the big leaps? I would say this one, navigate in autopilot without, uh, confirm- without having to confirm is a huge leap.

   4. EMElon Musk13:21

      It is a huge leap.

   5. LFLex Fridman13:22

      What are the-

   6. EMElon Musk13:22

      And it, it also automatically overtakes slow cars. So it's, it's both navigation, um, and seeking the fastest lane. So it'll, it'll, it'll take, you know, overtake a slower car, um, and exit the freeway and take highway interchanges. And, and then, uh, we have, uh, traffic light re- traffic light, uh, recognition, which introduced initially as a, as a warning. I, I mean, on the development version that I'm driving, the car fully, fully stops and, and goes at traffic lights.

   7. LFLex Fridman13:57

      So those are the steps, right? You just mentioned something, sort of inkling of a step towards full autonomy.

   8. EMElon Musk14:02

      Mm-hmm.

7. 14:03 – 16:53

   What’s left for full self-driving: city streets and parking lots

   1. LFLex Fridman14:03

      What would you say are the biggest technological roadblocks to full self-driving?

   2. EMElon Musk14:10

      Actually, I don't think we ... I think we just ... The full self-driving computer that we just, uh, that the, the Tesla, what we call the FSD computer, uh, that, that's, uh, now in production. Uh, so if you order, uh, any Model S or X or any Model 3 that has the full self-driving package, you'll get the FSD computer. That, that was, that's important, uh, to have enough base computation. Uh, then refining the neural net and the control software, uh, which, but all of that can just be provided as an over-the-air update. The thing that's really profound and what I'll be emphasizing at the, uh, sort of what ... that investor day that we're having focused on autonomy, is that the cars currently being produced, or the hardware currently being produced, is capable of full self-driving.

   3. LFLex Fridman15:01

      But capable is an interesting word.

   4. EMElon Musk15:03

      Mm-hmm.

   5. LFLex Fridman15:03

      Because, um ...

   6. EMElon Musk15:04

      Like the hardware is.

   7. LFLex Fridman15:05

      Yeah, the hardware.

   8. EMElon Musk15:05

      And as we refine the software, it, the capabilities will increase dramatically. Um, and then the reliability will increase dramatically, and then it will receive regulatory approval. So essentially buying a car today is an investment in the future. You're, you're essentially buying a, a ca- you're, you're buying ... The, I think the most profound thing is that if you buy a Tesla today, I believe you are buying an appreciating asset, not a depreciating asset.

   9. LFLex Fridman15:33

      So that's a really important statement there because if hardware is capable enough, that's the hard thing to upgrade-

   10. EMElon Musk15:39

       Yes.

   11. LFLex Fridman15:39

       ... usually.

   12. EMElon Musk15:40

       Exactly.

   13. LFLex Fridman15:40

       So then the rest is a software problem.

   14. EMElon Musk15:43

       Yes.

   15. LFLex Fridman15:44

       Of-

   16. EMElon Musk15:44

       Software has like no marginal cost, really.

   17. LFLex Fridman15:48

       But what's your intuition on the software side? How hard are the remaining steps to, to get it to where, um, you know, uh, the experience, uh, not just the safety but the-

   18. EMElon Musk16:04

       Mm-hmm.

   19. LFLex Fridman16:04

       ... full experience is something that people would, uh, enjoy?

   20. EMElon Musk16:09

       I think people will enjoy it very much so on, on, on any- on the highways. It's, it's a total game changer for quality of life for using, you know, Tesla autopilot on the highways. Uh, so it's really just extending that functionality to city streets, adding in the, the traffic light, uh, traffic light recognition, uh, navigating complex intersections, and, um, and, and then, uh, being able to navigate complicated pa- parking lots. So the car can, uh, exit a parking space and come and find you even if it's in a, a complete maze of a parking lot. And, uh, and, and then if ... And then you can just ... It could just drop you off and find a parking spot by itself.

8. 16:53 – 20:08

   Supervision, regulation, and proving safety statistically

   1. LFLex Fridman16:53

      Yeah, in terms of enjoyability and something that people would, uh, would actually find a lot of use from the parking lot is a, is a really, you know ... It's, it's rich of annoyance when you have to do it manually so there's a lot of benefit to be gained from automation there. So let me start injecting the human into this discussion a little bit. Uh, so let's talk, talk about full autonomy. If you look at the current level 4 vehicles being tested on road, like Waymo and so on, they're only technically autonomous. They're really level 2 systems with just a different f- design philosophy, because there's always a safety driver in almost all cases and they're monitoring the system.

   2. EMElon Musk17:33

      Right.

   3. LFLex Fridman17:33

      Do you see Tesla's full self-driving as still for a time to come requiring supervision of the hu- the human being? So its capabilities are powerful enough to drive, but nevertheless requires a human to still be supervising just like a safety driver is in a ... other fully autonomous vehicles?

   4. EMElon Musk17:57

      I think it'll, it'll require detecting hands on wheel for at, at least, uh, six months or something like that from here. It, it really is a question of, like ...... from a regulatory standpoint, uh, what, h- how much safer than a person does autopilot need to be f- for it to, to be okay to not monitor the car? You know, and, and this is a, a debate that one can have. A- And then if you, but you need, you need a l- a, a large sample s- a l- large amount of data, um, so that you can prove with high confidence, statistically speaking, that the car is dramatically safer than a person, um, and that adding in the person monitoring does not materially affect the safety. So, it might n- need to be, like, two or 300% safer than a person.

   5. LFLex Fridman18:50

      And how do you prove that?

   6. EMElon Musk18:51

      Incidents per mile.

   7. LFLex Fridman18:52

      Incidents per mile?

   8. EMElon Musk18:53

      Yeah.

   9. LFLex Fridman18:53

      So, crashes and fatalities? So-

   10. EMElon Musk18:56

       Yeah. I mean, f- uh, fat- fatalities would be a factor, but there, there're, there are just not enough fatalities to be statistically significant, uh, a- at scale. But there are enough crashes. Th- you know, there are much, far more crashes than there are fatalities. So you can just assess what is the probability of, uh, of a crash. The- then there's a- another step which is probability of injury, then probability of permanent injury, then pro- probability of death. And all of those need to be, uh, much better than a person, uh, by at least, perhaps, 200%.

   11. LFLex Fridman19:33

       And you think there is, uh, the ability to have a healthy discourse with the regulatory bodies on this topic?

   12. EMElon Musk19:40

       I mean, th- there's no question that, um, that, uh, regulators pay dis- disproportionate amounts o- of tension to that which generates press. This is just an objective fact. Um, and T- Tesla generates a lot of press. So, the, you know, uh, in the United States, there's, I think, almost 40,000 automotive deaths per year. Uh, but if there are four in a Tesla, they'll probably receive a thousand times more press than anyone else.

9. 20:08 – 23:09

   Driver vigilance research meets Tesla’s philosophy: when humans make it worse

   1. LFLex Fridman20:08

      So, the, the psychology of that is actually fascinating. I don't think we'll have enough time to talk about that. But I'd l- I have to talk to you about the human side of things. So, myself and our team at MIT recently released a paper on functional vigilance of drivers while using autopilot. This is work we've been doing since autopilot was first released publicly over three years ago.

   2. EMElon Musk20:30

      Mm-hmm.

   3. LFLex Fridman20:31

      Collecting video of driver faces and driver body. So, I saw that you tweeted a quote from the abstract, so I can at least, uh, guess that you've glanced at it?

   4. EMElon Musk20:43

      Yeah, I read it.

   5. LFLex Fridman20:44

      Can I talk you through what we found?

   6. EMElon Musk20:46

      Sure.

   7. LFLex Fridman20:46

      Okay. So, it appears that in the data that we've collected, that drivers are maintaining functional vigilance such that we're looking at 18,000 disengagements from autopilot, 18,900, and annotating were they able to take over control in a timely manner. So, they were there, present, looking at the road, uh, to take over control. Okay. So, this, uh, goes against what, what many would predict from the body of literature on vigilance with automation. Now the question is, do you think these results hold across the broader population? So, ours is just a small subset. Do you think... Uh, one of the criticism is that, you know, there's a small minority of drivers that may be highly responsible where their vigilance decrement would increase with autopilot use.

   8. EMElon Musk21:40

      I, I, I think this is all really gonna be swept... Uh, I, I mean, uh, uh, the, the system's improving so much, so fast, that this is gonna be a moot point very soon. W- where vigilance is... L- like if something's many times safer than a person, then adding a person, uh, does... The, the, the effect on safety is, is limited. Um, and in fact, uh, it could be negative.

   9. LFLex Fridman22:10

      That's really interesting. So the, uh, the f- so, so the fact that a, a human may, some percent of the population may, uh, exhibit a vigilance decrement will not affect the overall statistic numbers of safety?

   10. EMElon Musk22:22

       No. In fact, I think it, it will become, uh, very, very quickly, maybe even towards the end of this year, but I'd say I'd be shocked if it's not next year a- at the latest, that, um, having the per- having a human intervene will decrease safety. Decrease. Uh, it's, it's... Like, imagine if you're in an elevator, and it used to be that there were elevator operators, um, and, and you c- you couldn't go in an elevator by yourself and, and work the, the lever to move between floors. Um, and now, uh, nobody wants an elevator operator because the automated elevator that stops at the floors is much safer than the elevator operator. And in fact, it would be quite dangerous to have someone with a lever that can move the elevator between floors.

10. 23:09 – 24:28

    Camera-based driver monitoring: benefits now vs irrelevance later

    1. LFLex Fridman23:09

       So, that's a, that's a really powerful statement and a really interesting one. Uh, but I also have to ask from a user experience and from a safety perspective, one of the passions for me algorithmically is, uh, camera-based detection of, uh, of just sensing the human, about detecting what the driver's looking at, cognitive load, body pose. On the computer vision side, that's a fascinating problem, but do you th- and there's many an industry who believe you have to have camera-based driver monitoring. Do you think there is, could be benefit gained from driver monitoring?

    2. EMElon Musk23:41

       If you have a system that's, that's at or b- that's at or below, uh, human-level reliability, then driver monitoring makes sense. But if, if your system is dramatically better, m- more reliable than, than a human, then driver monitining- monitoring is not f- does not help much. And, uh, like I said, you, you... Just like as, uh, you wouldn't want someone interv- like, you wouldn't want someone in an elevator... If, if you're in an elevator, do you really want someone with a big lever, s- some random person operating the elevator between floors? Th- they could ha... I wouldn't trust that. I would rather have the buttons. (laughs)

    3. LFLex Fridman24:17

       L- okay. You're, uh, optimistic about the pace of improvement of the system, from what you've seen with a full self-driving car computer.

    4. EMElon Musk24:25

       The rate of improvement is exponential.

11. 24:28 – 26:57

    Operational design domain (ODD): wide deployment vs constrained geofencing

    1. LFLex Fridman24:28

       So, one- one of the other very interesting design choices early on th- that connects to this is the operational design domain of autopilot, so where autopilot is able to be turned on. The- so contrast another vehicle system that we're studying is the Cadillac Super Cruise system. That's, in terms of ODD, very constrained to particular kinds of highways, well mapped, tested. It's much narrower than the ODD of Tesla vehicles.

    2. EMElon Musk24:58

       Mm-hmm.

    3. LFLex Fridman24:59

       What's- there's- there's-

    4. EMElon Musk25:01

       It's like ADD. (laughs)

    5. LFLex Fridman25:02

       (laughs) Yeah.

    6. EMElon Musk25:04

       (laughs)

    7. LFLex Fridman25:04

       (laughs) That was good. That's- (laughs) that's a good line.

    8. EMElon Musk25:06

       (laughs)

    9. LFLex Fridman25:07

       Uh, (laughs) what was a design decision, uh, w- what- in that different philosophy of thinking where there's pros and cons. What we see with, uh, a- a wide ODD is drive- Tesla drivers are able to explore more of the limitations of the system, at least early on, and they understand, together with the instrument cluster display, they start to understand what are the capabilities. So, that's a benefit. The con is you're go- you're letting drivers use it basically anywhere. Uh, so there's a ch-

    10. EMElon Musk25:38

        Well, anywhere that it can d- has- can detect lanes with confidence.

    11. LFLex Fridman25:41

        Lanes. Was there a philosophy, uh, design decisions that were challenging that were being made there? Or j- from the very beginning, was that, uh, done on purpose with intent?

    12. EMElon Musk25:54

        Well, I mean, I think it's p- frankly, it's pretty crazy giving- letting people drive a two-ton death machine, uh, manually. Uh, that's crazy. Like, I hear- like, in the future, people will be like, "I- I can't believe anyone was just allowed to drive one of these two-ton death machines," and they could just drive it wherever they wanted, just like elevators. You could just, like, move the elevator with the lever wherever you want. It could stop at halfway between floors if you want. It's pretty crazy. So, (sighs) it- it's gonna seem like a mad thing in the future that people were driving cars.

    13. LFLex Fridman26:33

        So, I have a bunch of questions about the human psychology, about behavior and so on.

    14. EMElon Musk26:37

        I don't know that-

    15. LFLex Fridman26:37

        That would become that-

    16. EMElon Musk26:39

        Yeah, mood- totally mood.

    17. LFLex Fridman26:40

        Uh, because, uh, you have faith in the AI system. Uh, not faith, but, uh, the- both on the hardware side and the deep learning approach from learning from data will make it just far safer than humans.

    18. EMElon Musk26:55

        Yeah, exactly.

12. 26:57 – 28:27

    Adversarial attacks on neural nets: confidence in defenses

    1. LFLex Fridman26:57

       Recently, there were a few hackers who, uh, tricked autopilot to act in unexpected ways with adversarial examples.

    2. EMElon Musk27:03

       Mm-hmm.

    3. LFLex Fridman27:04

       So, we all know that neural network systems are very sensitive to minor disturbances, these adversarial examples on input. Do you think it's possible to defend against something like this for the broader-

    4. EMElon Musk27:14

       Oh, yeah. No problem.

    5. LFLex Fridman27:15

       (laughs) - for the industry?

    6. EMElon Musk27:16

       Sure. (laughs)

    7. LFLex Fridman27:16

       So- (laughs)

    8. EMElon Musk27:18

       Yeah.

    9. LFLex Fridman27:19

       Can you elaborate on the- on the confidence behind that answer? (laughs)

    10. EMElon Musk27:23

        Um, well, the- you know, a neural net is just, like, basically a bunch of mat- matrix math. Like, you have to be, like, a very sophisticated- somebody who really understands neural nets and, like, basically reverse engineer how the matrix is being built and then create a- a- a little thing that's just exactly, um, causes the matrix math to be slightly off. But it's very easy to then block it- block that by- by having qu- basically anti-re- negative recognition. It's like if y- i- if the system sees something that looks like a matrix hack, uh, exclude it. The- so f- it's such an easy thing to do.

    11. LFLex Fridman28:02

        So, learn both on the- the valid data and the invalid data. So, basically-

    12. EMElon Musk28:06

        Mm-hmm.

    13. LFLex Fridman28:06

        ... learn on the adversarial examples to excl- be able to exclude them.

    14. EMElon Musk28:09

        Yeah. Y- like, you basically want to s- both know what is- what is a car and what is definitely not a car.

    15. LFLex Fridman28:16

        (laughs)

    16. EMElon Musk28:16

        You- you train for, this is a car and this is definitely not a car. Those are two different things. You know? I- uh, pe- people have no idea of neural nets really. They probably think neural nets involves like, you know, a fishing net or something.

13. 28:27 – 32:29

    Beyond self-driving: AGI, love, and the simulation question

    1. LFLex Fridman28:27

       (laughs) Uh, so as you know, d- so taking a step beyond just Tesla and autopilot, uh, current deep learning approaches still seem, in some ways, to be f- far from general intelligence systems. Do you think the current approaches will take us to general intelligence, or do totally new ideas need to be invented?

    2. EMElon Musk28:55

       I think we're missing a few key ideas for general intelligence, general- artificial general intelligence. (sighs) But it's gonna be upon us very quickly, and then we'll need to figure out, what shall we do, if we even have that choice. But- but it- it's amazing how people can't differentiate between, say, the narrower AI that, you know, allows a car to figure out what a lane line is and- and- and- and, you know, and navigate streets versus general intelligence. Like, these are just very different things. Like, your toaster and your- and your computer are both machines, but one's much more sophisticated than another.

    3. LFLex Fridman29:38

       You're confident with Tesla you can create the world's best toaster. Uh-

    4. EMElon Musk29:43

       The world's best toaster, yes. The world's- the world's best self-driving... I'm- I- yes. To- to me, right now, this seems game, set, match. I don't- I mean, that's how... I don't want us to be complacent or overconfident, but that's what it app- that is just literally what it- how it appears right now. I could be wrong, but it appears to be the case that Tesla is vastly ahead of everyone.

    5. LFLex Fridman30:10

       Do you think we will ever create an AI system that we can love and loves us back in a deep, meaningful way, like in the movie Her?

    6. EMElon Musk30:20

       I think AI will be capable of convincing you to fall in love with it very well.

    7. LFLex Fridman30:25

       And that's different than us humans?

    8. EMElon Musk30:29

       You know, we start getting into a metaphysical question of, like, do emotions and thoughts exist in a different realm than the physical? And maybe they do, maybe they don't, I don't know. But, but from a physics standpoint, I tend to think, I tend to think of things, you know, like physics was my main sort of training, and, and from a physics standpoint, essentially, if, if it loves you in a way that is, that you can't tell whether it's real or not, it is real.

    9. LFLex Fridman30:57

       That's a physics view of love.

    10. EMElon Musk30:59

        Yeah.

    11. LFLex Fridman31:00

        (laughs)

    12. EMElon Musk31:01

        If there's no... If you, if you cannot dis- if you cannot prove that it does not, if there's no test that you can apply that would make it... allow you to tell the difference, then there is no difference.

    13. LFLex Fridman31:17

        Right. And it's similar to, uh, seeing our world as simulation. There may not be a test to tell the difference between what the real world-

    14. EMElon Musk31:24

        Yes.

    15. LFLex Fridman31:24

        ... and the simulation, and therefore, from a physics perspective, it might as well be the same thing.

    16. EMElon Musk31:29

        Yes. Uh, and there may be ways to test whether it's a simulation. There might be, I'm not saying there aren't. But you could certainly imagine that a simulation could, could correct wha- that once an entity in the simulation found a way to detect the simulation, it could either restart, you know, pause the root simulation, start a new simulation, or do one of many other things that then corrects for that error.

    17. LFLex Fridman31:53

        So when maybe you or somebody else creates an AGI system, and you get to ask her one question, what would that question be?

    18. EMElon Musk32:17

        What's outside the simulation?

    19. LFLex Fridman32:21

        Elon, thank you so much for talking today. It was a pleasure.

    20. EMElon Musk32:24

        All right. Thank you.

Episode duration: 32:44