Let me start with my post-hoc impressions of the original Veritasium video. This might be tainted by having watched the Tom Nicholas video. It was clearly marked as sponsored, and my takeaway was...
Exemplary
Let me start with my post-hoc impressions of the original Veritasium video. This might be tainted by having watched the Tom Nicholas video. It was clearly marked as sponsored, and my takeaway was that its main point was that driverless cars are safe if done properly (i.e. fully automated) and that we should invest in getting them on the roads. It was not making a point about genuine market readiness or commercial viability. Was it a masterwork of critical reporting? No, but I didn't expect that going in.
Let me also say that I'm not as neutral as I expect either Tom or Derek to be. I'm a fan of self-driving tech. I somewhat like Veritasium. I did not enjoy my first impression of Tom Nicolas. This is not a neutral reply.
Tech readiness
Let's start with the point of technology readiness. Again, my takeaway from the Veritasium video was not that the technology is literally ready for deployment. Tom however frames the Veritasium video as one proclaiming that the tech is ready to be rolled out globally, all that's stopping us is popular resistance, when in actually even a cursory glance shows the opposite: No car manufacturer has achieved full automation yet and Waymo have achieved it only in a walled garden. How you can come away from a video that displays a Waymo car having a case of overambitious cautiousness and the wacky user experience of having to send the car around the block to be able to continue your ride - how you can come away from that and think they're selling you a finished product is beyond me. In my opinion, Derek makes the point that the Waymo approach (as in, only let it run wild once you've achieved full automation) is safe for the public; not that it is available and you should use it. Those two are critically different and many digs at the practical usability of Waymo disappear once you reject Toms reframing. For example, he complains repeatedly that there are too many people involved in the operation of Waymo. Considering their product isn't done yet, I don't think that's surprising. For one, they're still developing; secondly, they are doing a very limited rollout, which justifies using people for customer support rather than sophisticated software - after all, they want to demonstrate that they can build a self-driving car, not that they can build a uber app. Similarly, regarding the mapping of streets and driving in unusual circumstances, of course there's going to be people supervising the car.
Tom also complains that Derek talks mostly about the sensors and little other tech aspects, e.g. the environment mapping. This again has to do with the framing Tom does: For a safety discussion, this mapping is only tangentially related; more important are the sensors that allow you to react. Of course sensor data processing also goes into that, which is super hard to convey, so I can't fault Derek for not doing that.
Statistics
First off, I appreciate the bit of honesty that the 60 million [traffic deaths in the 20th century] figure is not material to the argument. I would've appreciated the same for the 94% [of accidents have drivers as leading cause] figure. This kind of causality tracking (but was it really the driver's fault?) doesn't help if we have actual traffic statistics that show a reduction in accidents for self-driving cars. Because that is the one figure that I keep hearing about wrt. self-driving car safety and the one that actually convinces me: the number of km driven by the car per small/medium/heavy/lethal accident. He never addresses that argument though. Nevermind that the scenario with the overgrown hedge and deteriorated markings he constructs seems disingenuous: He attacks those entire 94% with a scenario that is surely not representative of those 94%, and that we don't even know whether the NTSB would "misclassify".
Tom later claims that the scenario with deteriorated markings and overgrown hedges would be a problem for an autonomous vehicle. However that depends heavily on the tech. Here I think waymo has the upper hand: It's reasonable to assume they'd have either previous mappings of the area or mappings from other directions. The inconsistencies in markings could then be discovered quite easily and the car could reasonably drive in such a way as to not assume others will yield. This is not rocket science.
As for the "most experienced driver" talking point, the fact that the cars have driven in 25 cities highlights what I think is waymo's path to commercial viability: I presume they can roll out their cars in a foreign city without first mapping it. You put the car there, have a safety driver on board and let it map the environment. Once the map is sufficiently accurate, you can remove the safety driver. Once the AI is sufficiently advanced, you might even be able to send it exploring without a safety driver. In fact, without looking into waymo's practices: I'm not at all convinced that humans had a big role to play in generating the detailed maps of Phoenix. I think just using the sensors on-site is probably all it takes.
Another aspect of having the "most experienced driver" is that - yes, it has diminishing returns if you keep it in the same suburb, but you still gather more and more edge cases of driver behaviour. Which, ya know, is one of the things their car does: Predicting other drivers. We humans do it too, and it's kind of essential. Although I'll put a caveat here that none of them mention: ML algorithms are substantially more data hungry than human cognition. We might fix that sometime soon, but for now there are a lot of tasks, e.g. in language and vision modelling where ML needs a lot more data. Not in everything. Data that is harder to access using human sensors is where ML can really shine. Not the case here though, imo.
Comparisons to other forms of automation
Tom also spends quite a bit of time attacking the comparisons to aviation and elevator automation. Maybe it's because I understand roughly how autonomous cars, aircraft autoland and autothrottle and various other forms of automation work, but the comparisons Veritasium draws to other tech are fair, at least considering that there's a time constraint on the video: Of course he could go into more detail of how they are related to self-driving and how they are not and then state why he thinks they are relevant to the discussion. But my take-away is that they are relevant, it's just that Tom has unreasonably high standards there. In any case, the trend for humans to be capable of safely automating ever more complex and safety-critical things seems clear. We also need less and less tight control over the environment as we advance. Also, Tom completely misses that the localizers utilized in the CAT3 Autoland are used by pilots anyway for instrument landings. With the exception of devices that have non-actuated physical switches in the cockpit (flaps, gears), this piece of automation does a lot of the pilot's work. Limited, yes, but clearly the plane genuinely handles the tight control loop without the pilot. As for Asiana 214, that one has been litigated in the comments by Tom and Veritasium. My takeaway there is that NTSB did focus mainly on human error, but added that the complexity and design of the automation system contributed to that. Incidently, this makes Waymo's point for them, as one strategy out of this conundrum is to automate away the human when possible.
Later in the video, Tom misses the mark completely on aviation safety yet again: Aircraft safety is so good because we invest so heavily in it. It's not that we spend so much even though the risks are low. The truth is, we stop investing so much into aviation safety, accident rates will skyrocket. Ask Boeing. Hell, ask any airline maintenance tech. The stuff that would happen to machines as complex as modern airliners if we didn't pay as much attention as we do is frightening.
Ethics
Oh my fucking god, the damn ethics thought experiments again. Fuck. No. No it's not ok to just rebutt that rare things still happen. The ethics of self-driving cars revolve around the comparison to humans, so let's start there. Has any human ever had to make such a trolley problem decision? You? Someone you know? Someone you heard of? Now, I haven't turned over every rock, but I have heard of not one incident where a human was faced with such a situation in traffic. And even then, I have about 0 hope that a human would even make a choice based on in any way rational thoughts. I could easily imagine a human sacrificing a busload of people that they can't see to save a person, any person, whose face they can see, purely because that person is no longer anonymous to them; assuming such situations would happen. I could also easily imagine humans just shitting their pants and doing nothing to save anyone; or freaking out and making things even worse. Genuine trolley problem events are exceedingly rare and we shouldn't let those disrupt us. Let me frame that as a trolley problem: we can either blither on about the ethics of self-driving cars and do nothing until we have found the solution to the trolley problem, or we can invest in self-driving cars, get them rolled out as soon as the technology is ready and save thousands of lives. Let's at least wait until we have one report of a traffic accident where a self-driving car had to make the choice of who to kill.
A driverless future
Now for the part where Tom will present to me that driverless cars are not inevitable. First he attacks some of the benefits they might bring, and he dismisses a promise of less congestion by stating that they have the same footprint on the road. However, driverless cars have the advantage of reacting much quicker. Consider the example of "traffic jams out of nowhere", where congestion is an emergent property of reaction times, mechanics and psychology. You could completely eliminate these in a purely-self-driving environment and I think you could almost prevent these by having a few driverless cars set the pace. They detect a traffic jam ahead and slow down a bit. The traffic jam has time to resolve (as opposed to otherwise when human drivers would have to progressively brake harder than the driver in front until we have a dead stop) and no one has to slow down too much. Similar mechanics are at work at intersections, where you can drastically increase throughput by following the car ahead more closely. Machines can do that. Machines could also theoretically operate intersections completely without signals. The potential is there. I don't think it's that hard to see. As for reclaiming value of stationary cars, I'm sure not everyone works at exactly the same time. Consequently, the cars could service other riders while you are at work or asleep.
Relatedly, with access to an affordable driverless taxi service, I can justify getting rid of my own car, which frees up money I could spend on public transit. Can't see the math? Let me elaborate. Currently I'm fortunate to have access to a cheap car sharing service. I can also commute to work using mass transit. I'm also frugal/lazy to a fault and would probably not like owning a car, but many people who could commute to work using mass transit need a car for other aspects of their lives: Odd trips here and there, transporting heavy stuff, roadtrips, whatever. People want access to cars, and realistically we can't ever make mass transit appealing enough to literally out-convenience a car in every aspect of transport. Anyway, so someone who needs access to a car, but could commute to work using mass transit. Given they own a car, how are they going to get to work if the car is already sitting in their driveway and the bus is less convenient and/or more expensive? By car, of course. Now, instead give them access to an affordable driverless taxi service that can do all the odd/inconvenient transport duties. Suddenly the question isn't "Do I buy a year pass for the tram and a car, or just the car", but rather it's "Do I buy a car or do I buy a year pass for the tram and occasionally call a waymo?". Waymo and other similar services free you up from the pressure to own a car, which frees you up from the pressure to use it. But, I hear you say, surely the answer is then carsharing and not driverless cars, right? Well, no. Carsharing alone still needs you to drive the damn thing. Go on, leave your car parked 360 days of the year, see how comfortable you feel in your driving skills the other few days. I'll wait. Ask me how I know. In this vision of the future (one out of several possible, I admit), providing self-driving cars actually reduces car utilization.
As for the comments that demand quite radical change towards pedestrian-focused cities I've seen on YT, I think it's important to work with what we have. Our cities are as they are and infrastructure is generally a long-term investment. Change is going to be slow, and unless we want to spend positively gargantuan sums redoing it, we have to deal with what we have. We shouldn't mistake "this is the ideal city for our current-day needs and possibilities." with "this is the city we should build right now". There is value in the infrastructure we have now, and that should factor into any discussion. And yes, train good, car bad, I agree. But again, gotta work with the infrastructure we have. In terms of flexibility, roads and road vehicles aren't going away for a while, might as well reduce traffic deaths.
Closing
I think it's important here to differentiate "Veritasium did an unethical ad video that should eliminate his credibility" and "self-driving cars are bad". Those are very different topics and I feel they're being commingled in unfortunate ways. The video also precludes any kind of constructive debate stemming from this, as it views like a hit-piece against Veritasium personally rather than a starter for a discussion of sponsorship ethics. The fact that Tom doesn't seem overly concerned with other examples of Edu YouTube / Sponsorships or other videos of Veritasium makes it seem even more like a hit piece rather than an invitation to a level-headed debate. I appreciate one has to drill somewhat deep to address the entire video, but a bit of context or comparisons would've made it seem less like shit stirring. Also, for how much he accuses Veritasium of being uncritical, he is very much also uncritical of his own claims, e.g. ascribing malice (or related ulterior motives) baselessly. But in his case the bias is ok, because takedowns are his shtick and he's not being paid for it by a sponsor but by youtube ad revenue and that's clearly better. (My superficial impression anyway)
Unfortunately, for me personally, the more interesting points about "mass transit doesn't have a lobby" and "corporate influence on Edu YouTube" are largely lost in a sea of half-cooked bickering about driverless cars and frankly unjustified attacks on Veritasium. I certainly see more lying by omission in Tom's video than I did in Veritasium's. And two wrongs don't make a right.
Worth my time? Well, I guess, considering it got me thinking enough to write this mess.
If you've seen parts of this before, it's because you saw me taking notes for this as edits on my previous comment. I've tried to make it cohesive and fluent, but I'm not entirely happy with it. Feel free to provide feedback.
Relatedly, with access to an affordable driverless taxi service, I can justify getting rid of my own car, which frees up money I could spend on public transit
I usually avoid Youtube "drama" at all costs but this was an interesting watch. Anyways from the comments there is a response from Veritasium: And the response to the response: Snarky...
I usually avoid Youtube "drama" at all costs but this was an interesting watch. Anyways from the comments there is a response from Veritasium:
Tom, I’m happy to receive your constructive criticism, but I’m disappointed you didn’t fix any of the factual errors we alerted you to via email before you launched this video. Examples:
23:42 You cherry-picked this quote to make it seem like the NTSB blamed automation for the crash, when the report focuses squarely on human error: “The National Transportation Safety Board (NTSB) determined that the probable cause of this accident was the flight crew’s mismanagement of the airplane’s descent during the visual approach, the Pilot Flying’s unintended deactivation of automatic airspeed control, the flight crew’s inadequate monitoring of airspeed, and the flight crew’s delayed execution of a go-around after they became aware that the airplane was below acceptable glidepath and airspeed tolerances.”
32:37 Self-driving cars have maps including traffic control so they would know where stop signs are meant to be even if road markings aren’t there or stop signs are obscured. Plus they have better obstacle detection and avoidance than human drivers.
39:16 I’m not saying rare accidents don’t happen, I’m saying they happen less often than common accidents, many of which could be prevented by self-driving cars. I sent you an academic paper that recreated in simulation 72 real-world fatal accidents that occurred in the area where Waymo operates. In almost all cases the accident was avoided or mitigated by the Waymo driver. Why did you omit this study?
47:03 It’s well understood that autonomous cars properly coordinated could reduce traffic because they don’t have the same reaction time delays as humans. For example all cars at an intersection could start moving together instead of one at a time as we currently do.
47:10 We don’t have to increase the car utilization rate to 100% to reclaim significant value. If cars were parked 90% of the time instead of 95%+ we would only need half as many vehicles.
Isn’t it ironic that a video purporting to call out misinformation itself contains so many distortions and factual errors? (Which we pointed out in advance but you didn’t feel compelled to fix)
On the issues themselves, I like public transport. I also ride a bike, and enjoy walking to get around when it’s practical. But cars will be a part of the transport mix for the foreseeable future. And it’s my opinion, based on the evidence, that roads will be safer the more cars are driven by computers than humans. No one has to pay me to tell you that.
And the response to the response:
Hiya,
As I said in my reply to the email from your team, my decision to make this video came from a place of disappointment rather than animosity. As such, it’s a real shame to see you again coming out swinging in response to my points rather than the video giving you any pause for thought surrounding your relationships with sponsors.
I’m glad to see that you've taken on board my critiques of your initial rebuttals to this video and attempted to strengthen them. I imagine you know the flaws in these responses to what you mistakenly call "factual errors" but I'll do you the courtesy of replying to them anyway.
23:42 You accuse me of cherry-picking here. However, the very next sentence from that which you quote states that 'Contributing to the accident were (1) the complexities of the autothrottle and autopilot flight director systems that were inadequately described in Boeing’s documentation and Asiana’s pilot training, which increased the likelihood of mode error [...]'. The reality is that, as in your original video, you're working on the false presumption that humans and automated/autonomous technologies are in conflict with one another. However, as is the case here, such technologies are always going to involve human interaction and they therefore need to be designed in ways that ensure that those interactions don't lead to accidents (as the Acting Chairman of the NTSB states in the quote included in my video). You use the crash of Flight 214 as an example of human failure with the implication that automated systems could have avoided the crash; it is disingenuous and distasteful to have eliminated all mention of autothrottle from your version of events just because it didn't suit the argument you wanted to make.
32:37 It is, again, completely disingenuous to refuse to mention the maps which Waymo vehicles rely on when it suits your argument during your initial video but to now hold them up as a vital part of the technology. You spoke as though those maps didn’t exist in the video and so it was only fair, in critiquing your video, to respond in kind.
A further point for consideration here is the extent to which creating maps/scans for the entirety of even the United States (let alone other countries too) is practical. That seems like a highly intensive task which would be very costly, possibly to the point it’s unworkable.
39:16 Two points here:
Firstly, you do not say in your video at all that the kinds of accidents considered when people discuss the ethical implications of driverless cars “could be prevented by self-driving cars”—you say they’re rare and quickly move on.
Secondly, the “academic paper” your team sent over was a write up of a study produced for Waymo and written by Waymo staff. I can’t seem to find any instance of it being published in a reputable journal (or a journal at all for that matter) or of it being subject to peer review. It is also, as one might expect, based entirely upon accidents which occured in the few suburbs of Phoenix where Waymo cars presently operate (and have had years of “training” in) and so it would be improper to suggest that its findings would be representative of a wider roll out to locations that Waymo’s cars have not been fine-tuned for or where conditions differ.
47:03 While they might have an impact, the idea that these tiny bits of time saved here and there would eliminate congestion is optimistic. The more extreme version of this (included in the CGP Grey video you’re referencing) in which all autonomous vehicles “talk” to one another (or even just to the traffic lights) would require such a level of implementation and standardisation that it is unlikely even in the medium term to the point where it’s a fun thought experiment but little more.
47:10 I make no such claim that we would need to “increase the car utilization rate to 100% to reclaim significant value” in my video. I merely state that the use of the 95% figure is misleading as it ignores several important caveats (like the existence of the nighttime). Even a figure which related to the percentage of the daytime that cars go unused would have been more useful to use here—it may not have given you such an impressive figure but it would have been a fairer representation of the reality.
To reiterate, suggesting that my video contains “distortions” and “factual errors” is deeply disingenuous. Nevertheless, I know that, whatever the circumstances, someone suggesting that you’ve acted improperly is a horrible experience. I hope that, with the passage of time, you’ll take on board some of the thoughts provided here about your relationships with sponsors.
Best wishes,
Tom
Snarky back-and-forth aside I think the overall thrust of Tom's argument to be valid -- it's damaging to your educational brand if you let sponsors dictate your message and it does appear that was the case here, at least to some extent. However, I also find a lot of the points raised to be somewhat nitpicky overall and detract from the core argument. I think it's also possible to raise this topic in a less abrasive manner, it seemed like this was almost personal at points in the video.
I agree, which is why I thought Tom made a mistake by spending so much time trying to lawyer over these details. Predictably the argument devolves into deviling out those details instead of a...
he more or less completely ignores the beginning and the end, where the "big ideas" of the video are, and as you said, nitpick specific details.
I agree, which is why I thought Tom made a mistake by spending so much time trying to lawyer over these details. Predictably the argument devolves into deviling out those details instead of a worthwhile discussion of sponsorship in educational videos.
this is the "I have a black friend" of publishing propaganda about transit systems
It's possible that Veritasium simply made a mistake -- that he was genuinely excited about the technology, took the sponsorship money and gushed about Waymo without realizing the implications. The guy is not a journalist after all and not thinking critically about something you're excited to talk about is probably easy to do.
It's also possible that he deliberately omitted critical components because it was in his contract or whatever - we'll never really know. But by at least acknowledging the possibility of good faith here you can open up a discussion directly on the issue itself while giving Veritasium something of an out. Instead we just get "no, you're wrong" ad nauseam and audiences are left unconvinced either way.
I don't know if someone getting paid to make what is essentially a propaganda video under the guise of an educational video can be excused as a mistake. It seems to me that such actions would have...
It's possible that Veritasium simply made a mistake -- that he was genuinely excited about the technology, took the sponsorship money and gushed about Waymo without realizing the implications.
I don't know if someone getting paid to make what is essentially a propaganda video under the guise of an educational video can be excused as a mistake. It seems to me that such actions would have to be intentional.
I think you can make an argument either way -- given that it's presumably the first time this has happened and that the guy is a nerd who regularly makes videos gushing about nerd topics I don't...
I think you can make an argument either way -- given that it's presumably the first time this has happened and that the guy is a nerd who regularly makes videos gushing about nerd topics I don't think it's entirely unlikely.
That said, it doesn't really matter even if we were 100% convinced it was done on purpose. Structuring your message around "Hey, sponsorship that asserts creative control over educational videos is bad for society. I understand that may have not been your intent but do you see how it comes across? Would you be willing to make a follow-up video sharing any of your reservations about the technology?" would lead to a more productive discussion and produce a better outcome than an attempted crucifixion.
I agree, and that was really what this video was. I didn't see this as an "attempted crucifixion," but a video showing all the negative effects of sponsor-controlled ads posing as educational content.
I agree, and that was really what this video was. I didn't see this as an "attempted crucifixion," but a video showing all the negative effects of sponsor-controlled ads posing as educational content.
OK "attempted crucifixion" was hyperbolic on my part but I'm troubled that there was no attempt to even mention the possibility of a good faith interpretation to platform discussion from. Tom...
OK "attempted crucifixion" was hyperbolic on my part but I'm troubled that there was no attempt to even mention the possibility of a good faith interpretation to platform discussion from.
Tom rightly points out that the Waymo video was offered in full support of the premise with no caveats or reservations but then offers none of his own for his unproveable hypothesis. That it's likely true to some extent is somewhat irrelevant in my opinion. Now the discussion is centered on the disagreement between the creators with all the wonderful tribal aspects inherited by their various audiences instead of the actual issue. If Tom led with a little bit of nuance it's a lot more likely we get a decent discussion and perhaps even a mea culpa from Veritasium where he acknowledges the issue, the perception of the video, and vows to do better.
Why does it have to be intent? I can easily see him being excited about the technology beforehand, then being offered the sponsorship. Is he going to throw out his previous opinion on the topic?...
Why does it have to be intent? I can easily see him being excited about the technology beforehand, then being offered the sponsorship. Is he going to throw out his previous opinion on the topic? Of course not.
Look, I get the disdain for self-driving cars. I'm not sure how they would affect the US market, whether it would lead to even greater suburbanization. I can say for myself that I use a car sharing service - very rarely, but every so often it's good to get to a place that is not ideal for public transit, or to haul around heavy stuff. You know what? I don't want a car. I don't even want to use a car most of the time. I'm not very confident in my driving skills any more because I don't get much time behind the wheel. Self-driving would fit perfectly there.
Is that an unreasonable position that is only explainable by tech propaganda? Could Derek hold a similar position?
Ultimately, the important question here is "how would this video have looked without the sponsorship?" - sadly, that's mostly unknowable unless we see a contract or sponsorship guidelines. Which I don't think is going to happen. Maybe we can dig out a video of him talking about a similarly controversial topic without a sponsorship? Bonus if he's similarly interested/biased on the topic.
Not having seen the video yet, just scrolling through the comments to get a feel, I think it's important here to differentiate "Veritasium did an unethical ad video that should eliminate his...
Not having seen the video yet, just scrolling through the comments to get a feel, I think it's important here to differentiate "Veritasium did an unethical ad video that should eliminate his credibility" and "self-driving cars are bad". Those are very different topics and I feel they're going to be commingled in unfortunate ways. They certainly are in the YT comments.
As for the original video, that one was clearly one big sponsored video, at least to me. I'm not sure what degree of critical reporting people were expecting.
And yes, I'll watch the video once I can get around to it. Apologies if I'm off base as a result.
A lot, clearly. Veritasium, aside from this video (and maybe a few others, can't remember if he did more sponsorships beforehand), marketed itself as trustworthy (It's even in the slogan...
I'm not sure what degree of critical reporting people were expecting.
A lot, clearly.
Veritasium, aside from this video (and maybe a few others, can't remember if he did more sponsorships beforehand), marketed itself as trustworthy (It's even in the slogan "Veritasium: The element of truth"), and as far as I know anyway, did very well on that, showing reputable evidence and interviewing (as far as I recall, mostly unbiased) experts.
So, people who were subscribed to Veritasium and watched his videos, trusted him to tell the truth accurately.
And this is in the context of YouTube info-tainment, where most people don't do their own research afterwards and take these videos at face value. (Most people consider them something to pass the time and "learn" at the same time with while eating lunch and whatnot)
(The rest of this comment might be considered political)
Take for another example, the YouTube channel Kurzgesagt. Their latest climate change video is sponsored by Bill Gates' personal blog "Gates Notes". A lot of people listened to and shared their advice as something trustworthy, when all it boils down to is "vote for better laws, and vote for your wallet", and in a world where legislators are "donated to", and where voting with your wallet implies the more money you have, the more votes you have, this video doesn't really advise on anything except keeping the status quo.
Which would be helpful to Gates, as someone who got the power he has today from the back of Microsoft's capitalism, often unethical, and continues to do so (with funding COVID vaccines and forcing them into a deal with a company he has investments on) with a hefty PR campaign to present himself as a good guy.
Ugh. Gotta say hearing about stuff like this honestly drains me a lot. I've long been sick of constantly questioning everything I read and hear everywhere, all of the time. I want to be able to...
Ugh. Gotta say hearing about stuff like this honestly drains me a lot. I've long been sick of constantly questioning everything I read and hear everywhere, all of the time. I want to be able to trust stuff now and then, and it feels like every time I turn my head there's a new post somewhere about how so-and-so isn't trustworthy.
That's not to say I think you shouldn't have made this post. Just that hearing about Kurzgesagt making billionaire-favorable takes due to sponsorships sucks because I do trust them, and I want to be able to keep doing so. I know that alone needn't tank their reputation for me, but it's the uncritical perspective I had that was so enjoyable.
I recognize that just accepting Youtube edutainment videos is, in general, unwise, but I want to have somewhere I can reasonably do so — trying to keep my guard up 24/7 is both infeasible and stressful. I need breaks from that, and I keep worrying that I'm losing them.
You're also just accepting a comment on Tildes, from somebody you do not know and have no reason to trust, as fact when the reality is a lot more nuanced. I found that being highly informed about...
You're also just accepting a comment on Tildes, from somebody you do not know and have no reason to trust, as fact when the reality is a lot more nuanced.
I found that being highly informed about a few things gives you the opportunity to more often have a correct gut reaction when somebody is talking shit. It also gives you a decent feel for when you're yourself not informed enough and makes you want to look things up; something which also becomes easier and quicker as you do it more.
Don't fall into the trap of thinking that whoever's opinion you last read is more correct than the previous one.
Thank you. I love Veritasium. His videos are extremely well-made, and as far as I can tell, quite accurate, at least for a YouTube video. It doesn't hurt to hear a little about the backstory.
Thank you. I love Veritasium. His videos are extremely well-made, and as far as I can tell, quite accurate, at least for a YouTube video.
It doesn't hurt to hear a little about the backstory.
Let me start with my post-hoc impressions of the original Veritasium video. This might be tainted by having watched the Tom Nicholas video. It was clearly marked as sponsored, and my takeaway was that its main point was that driverless cars are safe if done properly (i.e. fully automated) and that we should invest in getting them on the roads. It was not making a point about genuine market readiness or commercial viability. Was it a masterwork of critical reporting? No, but I didn't expect that going in.
Let me also say that I'm not as neutral as I expect either Tom or Derek to be. I'm a fan of self-driving tech. I somewhat like Veritasium. I did not enjoy my first impression of Tom Nicolas. This is not a neutral reply.
Tech readiness
Let's start with the point of technology readiness. Again, my takeaway from the Veritasium video was not that the technology is literally ready for deployment. Tom however frames the Veritasium video as one proclaiming that the tech is ready to be rolled out globally, all that's stopping us is popular resistance, when in actually even a cursory glance shows the opposite: No car manufacturer has achieved full automation yet and Waymo have achieved it only in a walled garden. How you can come away from a video that displays a Waymo car having a case of overambitious cautiousness and the wacky user experience of having to send the car around the block to be able to continue your ride - how you can come away from that and think they're selling you a finished product is beyond me. In my opinion, Derek makes the point that the Waymo approach (as in, only let it run wild once you've achieved full automation) is safe for the public; not that it is available and you should use it. Those two are critically different and many digs at the practical usability of Waymo disappear once you reject Toms reframing. For example, he complains repeatedly that there are too many people involved in the operation of Waymo. Considering their product isn't done yet, I don't think that's surprising. For one, they're still developing; secondly, they are doing a very limited rollout, which justifies using people for customer support rather than sophisticated software - after all, they want to demonstrate that they can build a self-driving car, not that they can build a uber app. Similarly, regarding the mapping of streets and driving in unusual circumstances, of course there's going to be people supervising the car.
Tom also complains that Derek talks mostly about the sensors and little other tech aspects, e.g. the environment mapping. This again has to do with the framing Tom does: For a safety discussion, this mapping is only tangentially related; more important are the sensors that allow you to react. Of course sensor data processing also goes into that, which is super hard to convey, so I can't fault Derek for not doing that.
Statistics
First off, I appreciate the bit of honesty that the 60 million [traffic deaths in the 20th century] figure is not material to the argument. I would've appreciated the same for the 94% [of accidents have drivers as leading cause] figure. This kind of causality tracking (but was it really the driver's fault?) doesn't help if we have actual traffic statistics that show a reduction in accidents for self-driving cars. Because that is the one figure that I keep hearing about wrt. self-driving car safety and the one that actually convinces me: the number of km driven by the car per small/medium/heavy/lethal accident. He never addresses that argument though. Nevermind that the scenario with the overgrown hedge and deteriorated markings he constructs seems disingenuous: He attacks those entire 94% with a scenario that is surely not representative of those 94%, and that we don't even know whether the NTSB would "misclassify".
Tom later claims that the scenario with deteriorated markings and overgrown hedges would be a problem for an autonomous vehicle. However that depends heavily on the tech. Here I think waymo has the upper hand: It's reasonable to assume they'd have either previous mappings of the area or mappings from other directions. The inconsistencies in markings could then be discovered quite easily and the car could reasonably drive in such a way as to not assume others will yield. This is not rocket science.
As for the "most experienced driver" talking point, the fact that the cars have driven in 25 cities highlights what I think is waymo's path to commercial viability: I presume they can roll out their cars in a foreign city without first mapping it. You put the car there, have a safety driver on board and let it map the environment. Once the map is sufficiently accurate, you can remove the safety driver. Once the AI is sufficiently advanced, you might even be able to send it exploring without a safety driver. In fact, without looking into waymo's practices: I'm not at all convinced that humans had a big role to play in generating the detailed maps of Phoenix. I think just using the sensors on-site is probably all it takes.
Another aspect of having the "most experienced driver" is that - yes, it has diminishing returns if you keep it in the same suburb, but you still gather more and more edge cases of driver behaviour. Which, ya know, is one of the things their car does: Predicting other drivers. We humans do it too, and it's kind of essential. Although I'll put a caveat here that none of them mention: ML algorithms are substantially more data hungry than human cognition. We might fix that sometime soon, but for now there are a lot of tasks, e.g. in language and vision modelling where ML needs a lot more data. Not in everything. Data that is harder to access using human sensors is where ML can really shine. Not the case here though, imo.
Comparisons to other forms of automation
Tom also spends quite a bit of time attacking the comparisons to aviation and elevator automation. Maybe it's because I understand roughly how autonomous cars, aircraft autoland and autothrottle and various other forms of automation work, but the comparisons Veritasium draws to other tech are fair, at least considering that there's a time constraint on the video: Of course he could go into more detail of how they are related to self-driving and how they are not and then state why he thinks they are relevant to the discussion. But my take-away is that they are relevant, it's just that Tom has unreasonably high standards there. In any case, the trend for humans to be capable of safely automating ever more complex and safety-critical things seems clear. We also need less and less tight control over the environment as we advance. Also, Tom completely misses that the localizers utilized in the CAT3 Autoland are used by pilots anyway for instrument landings. With the exception of devices that have non-actuated physical switches in the cockpit (flaps, gears), this piece of automation does a lot of the pilot's work. Limited, yes, but clearly the plane genuinely handles the tight control loop without the pilot. As for Asiana 214, that one has been litigated in the comments by Tom and Veritasium. My takeaway there is that NTSB did focus mainly on human error, but added that the complexity and design of the automation system contributed to that. Incidently, this makes Waymo's point for them, as one strategy out of this conundrum is to automate away the human when possible.
Later in the video, Tom misses the mark completely on aviation safety yet again: Aircraft safety is so good because we invest so heavily in it. It's not that we spend so much even though the risks are low. The truth is, we stop investing so much into aviation safety, accident rates will skyrocket. Ask Boeing. Hell, ask any airline maintenance tech. The stuff that would happen to machines as complex as modern airliners if we didn't pay as much attention as we do is frightening.
Ethics
Oh my fucking god, the damn ethics thought experiments again. Fuck. No. No it's not ok to just rebutt that rare things still happen. The ethics of self-driving cars revolve around the comparison to humans, so let's start there. Has any human ever had to make such a trolley problem decision? You? Someone you know? Someone you heard of? Now, I haven't turned over every rock, but I have heard of not one incident where a human was faced with such a situation in traffic. And even then, I have about 0 hope that a human would even make a choice based on in any way rational thoughts. I could easily imagine a human sacrificing a busload of people that they can't see to save a person, any person, whose face they can see, purely because that person is no longer anonymous to them; assuming such situations would happen. I could also easily imagine humans just shitting their pants and doing nothing to save anyone; or freaking out and making things even worse. Genuine trolley problem events are exceedingly rare and we shouldn't let those disrupt us. Let me frame that as a trolley problem: we can either blither on about the ethics of self-driving cars and do nothing until we have found the solution to the trolley problem, or we can invest in self-driving cars, get them rolled out as soon as the technology is ready and save thousands of lives. Let's at least wait until we have one report of a traffic accident where a self-driving car had to make the choice of who to kill.
A driverless future
Now for the part where Tom will present to me that driverless cars are not inevitable. First he attacks some of the benefits they might bring, and he dismisses a promise of less congestion by stating that they have the same footprint on the road. However, driverless cars have the advantage of reacting much quicker. Consider the example of "traffic jams out of nowhere", where congestion is an emergent property of reaction times, mechanics and psychology. You could completely eliminate these in a purely-self-driving environment and I think you could almost prevent these by having a few driverless cars set the pace. They detect a traffic jam ahead and slow down a bit. The traffic jam has time to resolve (as opposed to otherwise when human drivers would have to progressively brake harder than the driver in front until we have a dead stop) and no one has to slow down too much. Similar mechanics are at work at intersections, where you can drastically increase throughput by following the car ahead more closely. Machines can do that. Machines could also theoretically operate intersections completely without signals. The potential is there. I don't think it's that hard to see. As for reclaiming value of stationary cars, I'm sure not everyone works at exactly the same time. Consequently, the cars could service other riders while you are at work or asleep.
Relatedly, with access to an affordable driverless taxi service, I can justify getting rid of my own car, which frees up money I could spend on public transit. Can't see the math? Let me elaborate. Currently I'm fortunate to have access to a cheap car sharing service. I can also commute to work using mass transit. I'm also frugal/lazy to a fault and would probably not like owning a car, but many people who could commute to work using mass transit need a car for other aspects of their lives: Odd trips here and there, transporting heavy stuff, roadtrips, whatever. People want access to cars, and realistically we can't ever make mass transit appealing enough to literally out-convenience a car in every aspect of transport. Anyway, so someone who needs access to a car, but could commute to work using mass transit. Given they own a car, how are they going to get to work if the car is already sitting in their driveway and the bus is less convenient and/or more expensive? By car, of course. Now, instead give them access to an affordable driverless taxi service that can do all the odd/inconvenient transport duties. Suddenly the question isn't "Do I buy a year pass for the tram and a car, or just the car", but rather it's "Do I buy a car or do I buy a year pass for the tram and occasionally call a waymo?". Waymo and other similar services free you up from the pressure to own a car, which frees you up from the pressure to use it. But, I hear you say, surely the answer is then carsharing and not driverless cars, right? Well, no. Carsharing alone still needs you to drive the damn thing. Go on, leave your car parked 360 days of the year, see how comfortable you feel in your driving skills the other few days. I'll wait. Ask me how I know. In this vision of the future (one out of several possible, I admit), providing self-driving cars actually reduces car utilization.
As for the comments that demand quite radical change towards pedestrian-focused cities I've seen on YT, I think it's important to work with what we have. Our cities are as they are and infrastructure is generally a long-term investment. Change is going to be slow, and unless we want to spend positively gargantuan sums redoing it, we have to deal with what we have. We shouldn't mistake "this is the ideal city for our current-day needs and possibilities." with "this is the city we should build right now". There is value in the infrastructure we have now, and that should factor into any discussion. And yes, train good, car bad, I agree. But again, gotta work with the infrastructure we have. In terms of flexibility, roads and road vehicles aren't going away for a while, might as well reduce traffic deaths.
Closing
I think it's important here to differentiate "Veritasium did an unethical ad video that should eliminate his credibility" and "self-driving cars are bad". Those are very different topics and I feel they're being commingled in unfortunate ways. The video also precludes any kind of constructive debate stemming from this, as it views like a hit-piece against Veritasium personally rather than a starter for a discussion of sponsorship ethics. The fact that Tom doesn't seem overly concerned with other examples of Edu YouTube / Sponsorships or other videos of Veritasium makes it seem even more like a hit piece rather than an invitation to a level-headed debate. I appreciate one has to drill somewhat deep to address the entire video, but a bit of context or comparisons would've made it seem less like shit stirring. Also, for how much he accuses Veritasium of being uncritical, he is very much also uncritical of his own claims, e.g. ascribing malice (or related ulterior motives) baselessly. But in his case the bias is ok, because takedowns are his shtick and he's not being paid for it by a sponsor but by youtube ad revenue and that's clearly better. (My superficial impression anyway)
Unfortunately, for me personally, the more interesting points about "mass transit doesn't have a lobby" and "corporate influence on Edu YouTube" are largely lost in a sea of half-cooked bickering about driverless cars and frankly unjustified attacks on Veritasium. I certainly see more lying by omission in Tom's video than I did in Veritasium's. And two wrongs don't make a right.
Worth my time? Well, I guess, considering it got me thinking enough to write this mess.
If you've seen parts of this before, it's because you saw me taking notes for this as edits on my previous comment. I've tried to make it cohesive and fluent, but I'm not entirely happy with it. Feel free to provide feedback.
What is wrong with the existing taxi service?
I usually avoid Youtube "drama" at all costs but this was an interesting watch. Anyways from the comments there is a response from Veritasium:
And the response to the response:
Snarky back-and-forth aside I think the overall thrust of Tom's argument to be valid -- it's damaging to your educational brand if you let sponsors dictate your message and it does appear that was the case here, at least to some extent. However, I also find a lot of the points raised to be somewhat nitpicky overall and detract from the core argument. I think it's also possible to raise this topic in a less abrasive manner, it seemed like this was almost personal at points in the video.
I agree, which is why I thought Tom made a mistake by spending so much time trying to lawyer over these details. Predictably the argument devolves into deviling out those details instead of a worthwhile discussion of sponsorship in educational videos.
It's possible that Veritasium simply made a mistake -- that he was genuinely excited about the technology, took the sponsorship money and gushed about Waymo without realizing the implications. The guy is not a journalist after all and not thinking critically about something you're excited to talk about is probably easy to do.
It's also possible that he deliberately omitted critical components because it was in his contract or whatever - we'll never really know. But by at least acknowledging the possibility of good faith here you can open up a discussion directly on the issue itself while giving Veritasium something of an out. Instead we just get "no, you're wrong" ad nauseam and audiences are left unconvinced either way.
I don't know if someone getting paid to make what is essentially a propaganda video under the guise of an educational video can be excused as a mistake. It seems to me that such actions would have to be intentional.
I think you can make an argument either way -- given that it's presumably the first time this has happened and that the guy is a nerd who regularly makes videos gushing about nerd topics I don't think it's entirely unlikely.
That said, it doesn't really matter even if we were 100% convinced it was done on purpose. Structuring your message around "Hey, sponsorship that asserts creative control over educational videos is bad for society. I understand that may have not been your intent but do you see how it comes across? Would you be willing to make a follow-up video sharing any of your reservations about the technology?" would lead to a more productive discussion and produce a better outcome than an attempted crucifixion.
I agree, and that was really what this video was. I didn't see this as an "attempted crucifixion," but a video showing all the negative effects of sponsor-controlled ads posing as educational content.
OK "attempted crucifixion" was hyperbolic on my part but I'm troubled that there was no attempt to even mention the possibility of a good faith interpretation to platform discussion from.
Tom rightly points out that the Waymo video was offered in full support of the premise with no caveats or reservations but then offers none of his own for his unproveable hypothesis. That it's likely true to some extent is somewhat irrelevant in my opinion. Now the discussion is centered on the disagreement between the creators with all the wonderful tribal aspects inherited by their various audiences instead of the actual issue. If Tom led with a little bit of nuance it's a lot more likely we get a decent discussion and perhaps even a mea culpa from Veritasium where he acknowledges the issue, the perception of the video, and vows to do better.
Why does it have to be intent? I can easily see him being excited about the technology beforehand, then being offered the sponsorship. Is he going to throw out his previous opinion on the topic? Of course not.
Look, I get the disdain for self-driving cars. I'm not sure how they would affect the US market, whether it would lead to even greater suburbanization. I can say for myself that I use a car sharing service - very rarely, but every so often it's good to get to a place that is not ideal for public transit, or to haul around heavy stuff. You know what? I don't want a car. I don't even want to use a car most of the time. I'm not very confident in my driving skills any more because I don't get much time behind the wheel. Self-driving would fit perfectly there.
Is that an unreasonable position that is only explainable by tech propaganda? Could Derek hold a similar position?
Ultimately, the important question here is "how would this video have looked without the sponsorship?" - sadly, that's mostly unknowable unless we see a contract or sponsorship guidelines. Which I don't think is going to happen. Maybe we can dig out a video of him talking about a similarly controversial topic without a sponsorship? Bonus if he's similarly interested/biased on the topic.
Not having seen the video yet, just scrolling through the comments to get a feel, I think it's important here to differentiate "Veritasium did an unethical ad video that should eliminate his credibility" and "self-driving cars are bad". Those are very different topics and I feel they're going to be commingled in unfortunate ways. They certainly are in the YT comments.
As for the original video, that one was clearly one big sponsored video, at least to me. I'm not sure what degree of critical reporting people were expecting.
And yes, I'll watch the video once I can get around to it. Apologies if I'm off base as a result.
A lot, clearly.
Veritasium, aside from this video (and maybe a few others, can't remember if he did more sponsorships beforehand), marketed itself as trustworthy (It's even in the slogan "Veritasium: The element of truth"), and as far as I know anyway, did very well on that, showing reputable evidence and interviewing (as far as I recall, mostly unbiased) experts.
So, people who were subscribed to Veritasium and watched his videos, trusted him to tell the truth accurately.
And this is in the context of YouTube info-tainment, where most people don't do their own research afterwards and take these videos at face value. (Most people consider them something to pass the time and "learn" at the same time with while eating lunch and whatnot)
(The rest of this comment might be considered political)
Take for another example, the YouTube channel Kurzgesagt. Their latest climate change video is sponsored by Bill Gates' personal blog "Gates Notes". A lot of people listened to and shared their advice as something trustworthy, when all it boils down to is "vote for better laws, and vote for your wallet", and in a world where legislators are "donated to", and where voting with your wallet implies the more money you have, the more votes you have, this video doesn't really advise on anything except keeping the status quo.
Which would be helpful to Gates, as someone who got the power he has today from the back of Microsoft's capitalism, often unethical, and continues to do so (with funding COVID vaccines and forcing them into a deal with a company he has investments on) with a hefty PR campaign to present himself as a good guy.
Ugh. Gotta say hearing about stuff like this honestly drains me a lot. I've long been sick of constantly questioning everything I read and hear everywhere, all of the time. I want to be able to trust stuff now and then, and it feels like every time I turn my head there's a new post somewhere about how so-and-so isn't trustworthy.
That's not to say I think you shouldn't have made this post. Just that hearing about Kurzgesagt making billionaire-favorable takes due to sponsorships sucks because I do trust them, and I want to be able to keep doing so. I know that alone needn't tank their reputation for me, but it's the uncritical perspective I had that was so enjoyable.
I recognize that just accepting Youtube edutainment videos is, in general, unwise, but I want to have somewhere I can reasonably do so — trying to keep my guard up 24/7 is both infeasible and stressful. I need breaks from that, and I keep worrying that I'm losing them.
You're also just accepting a comment on Tildes, from somebody you do not know and have no reason to trust, as fact when the reality is a lot more nuanced.
I found that being highly informed about a few things gives you the opportunity to more often have a correct gut reaction when somebody is talking shit. It also gives you a decent feel for when you're yourself not informed enough and makes you want to look things up; something which also becomes easier and quicker as you do it more.
Don't fall into the trap of thinking that whoever's opinion you last read is more correct than the previous one.
Thank you. I love Veritasium. His videos are extremely well-made, and as far as I can tell, quite accurate, at least for a YouTube video.
It doesn't hurt to hear a little about the backstory.