Elon Musk: Self-driving is way harder than I thought | Lex Fridman Podcast Clips
ฝัง
- เผยแพร่เมื่อ 28 พ.ค. 2024
- Lex Fridman Podcast full episode: • Elon Musk: SpaceX, Mar...
Please support this podcast by checking out our sponsors:
- Athletic Greens: athleticgreens.com/lex and use code LEX to get 1 month of fish oil
- ButcherBox: butcherbox.com/lex to get offers & discounts
- InsideTracker: insidetracker.com/lex and use code Lex25 to get 25% off
- ROKA: roka.com/ and use code LEX to get 20% off your first order
- Eight Sleep: www.eightsleep.com/lex and use code LEX to get special savings
GUEST BIO:
Elon Musk is CEO of SpaceX, Tesla, Neuralink, and Boring Company.
PODCAST INFO:
Podcast website: lexfridman.com/podcast
Apple Podcasts: apple.co/2lwqZIr
Spotify: spoti.fi/2nEwCF8
RSS: lexfridman.com/feed/podcast/
Full episodes playlist: • Lex Fridman Podcast
Clips playlist: • Lex Fridman Podcast Clips
SOCIAL:
- Twitter: / lexfridman
- LinkedIn: / lexfridman
- Facebook: / lexfridman
- Instagram: / lexfridman
- Medium: / lexfridman
- Reddit: / lexfridman
- Support on Patreon: / lexfridman - วิทยาศาสตร์และเทคโนโลยี
Thanks for watching this clip. Full podcast with Elon Musk and Lex Fridman is here: th-cam.com/video/DxREm3s1scA/w-d-xo.html
@Lex Clips It's difficult to believe Musk was so foolish about this. Did he really not sit down of an evening and scratch out on a couple of whiteboards the computing power and speed necessary for a self-driving vehicle---the number of bits and flops that would have to be crunched? I did that of an hour at the old Baseball Think Factory in 2008 and it was easy to see it wouldn't be remotely possible on a significant scale prior to 2020.
yeah but what about this MIT presentation about car doors?
Nice podcast, one thing and I dunno eh it bugs me lol but please either have a glass of water or open ur mouth a little more when talking
Sorry hate criticising and probably only me that's bug bearing haha
funny how you both look like women :)
Humans don't drive by converting every observable object into a vector and then determining it's relevance. That's backwards.
Humans drive by determining the relevance of each object at any given moment and determining the vector and likely future position of the most relevant object. This is a loop and the most relevant object can change with every cycle. Even a human can do that several times in a second and the real problem is determining the most relevant object. That is why more experienced drivers are better drivers than younger drivers who have far better reaction times. The older drivers identify the most relevant objects far faster then the younger drivers.
When you face a situation of object occlusion it goes into your short-term memory until you either see the object again or it becomes irrelevant because you have passed the object. If you're sensible and a cautious driver, you slow down as you approach the point of the road where the no longer visible object is likely to be. You also pay attention to that point far more closely than you do other points.
When you enter a school zone, as mentioned, you react appropriately by slowing down. When you leave the school zone you just increase your speed. You're not thinking, "I'm in a school zone, I have to drive slower," every moment you're in the school zone. You just drive 20mph until you see the sign that tells you that you're leaving the school zone.
What you really have to do is have a system that keeps the vehicle between the lines unless it is overridden by an 'interrupt' from the part of the system that is watching the most relevant objects.
It would be so interesting to hear a bunch of the programmers talking about how they solved various problems on this topic.
th-cam.com/video/6hkiTejoyms/w-d-xo.html
You should check out this AI podcast.
There is. Tesla AI day th-cam.com/video/j0z4FweCy4M/w-d-xo.html
@@saxo689 which one?
@@arpit743 it was a joke. this one
Elon in 2035: Going to Mars is much harder than I thought
too real man :(
All of his projects have massive speculation behind them. He talks about these very advanced and futuristic visions but we are not even close to having the technology or the support of government(because of greed) to achieve them just yet.
Someone will pick up where he left off, that's the whole point of innovation
@@YourSlogan seem like the whole point of human life to pass information down
True. But I think he knew it’s hard and he is deliberately overoptimistic
I've been a delivery driver for years and I've thought this for years it's not that self driving wouldn't work great but it's the roads and the world around it that we've created that the computers are almost confused how to approach it, it's so un optimized and flawed
Look at what google is doing for it. Only need to map about 10 certain cities in the world to capture most of the taxi work.
@@bennyl7224 can you link me to some video about it? thanks
@@richardkerckhove not in the country
@@richardkerckhove a separate lane on a hwy you could have conga lines of EVs connected with FSD AI.
But that requires government to allocate lanes.
@@Ergooo th-cam.com/video/gVsJvLneNS0/w-d-xo.html
It's been obvious for at least 4 years that self driving is way harder than Elon Musk thought.
Snake oil salesman is gonna say whatever it takes to make you buy his garbage.
@@attractivegd9531 He over-promises on autonomy but the rest isn't snake oil. There's a lot of haters out there. Sad really. The guy has transformed global space launch and instead of resting on his laurels upped the ante and went for Starship. It's amazing and fun to watch unlike most industry in automotive or aerospace.
@@dorbie You are definitely a fanboy, he has been debunked many times now and by the highest science authorities like CNRS.
@@attractivegd9531 Wrong, you replied to me criticizing Elon and his long overhyped self-driving so your claim about me being a fanboy is absurd and lacks any credibility. Just the facts, SpaceX has landed over 100 first stage rockets with affordable refurb, and cut the cost of space launch to 25% of the competition, forcing ULA to halve their prices and come up with their own engine reuse plan. They've returned the USA to manned space launch as Boeing still struggles to even test Starliner. So no, you can't debunk stuff that's literally already happened, like millions of electric vehicles sold. In automotive and Aerospace Musk's companies have pioneered technologies that the competition actually killed instead of leading with. Now they're all left playing catchup but they lack the culture to compete IMHO. I'll be delighted if Blue Origin actually makes some progress on their orbital efforts, but their lawsuits hold humanity back. Musk isn't above criticism, but your ilk are a joke with your stupid "debunked" claims, you're the flat Earthers of tech innovation (let me guess you're a Thunderf00t fanboy).
@@dorbie lt;dr sorry fanboy!
It feels like the Full Self Driving development is only on the vehicle side to make vehicles be able to behave like human and read the signs on the road. In the future traffic and road infrastructure will also need to be upgraded to facilitate full scale autonomy.
Careful, don't go putting responsibility on municipal governments, we'll never get self driving😅
That doesn't scale. Self driving cars need to be able to drive where humans can without special infrastructure.
if you treat vehicles like horses back in the days, where the horse is the main driver and humans were passengers of it (on its back), then we just design roads for the horses. There wouldn't be anything silly in the far future as like a human riding on the back of another human in long distances. So, although it is fun for petrol heads to go around vroom vrooming, the far future remains to be functional evolution. And then there will people who treasure beautiful cars as to beautiful horses. That's their thing, but the main infrastructure is going to be autonomous vehicle when it gets solved. And the engineering way to do it is "divide and conquer", which has been the most successful approach. And we are dividing and revising constraints of the system, so that it would be successful. (removing chaos of snow, windy conditions, tornadoes, drunk drivers, random moose on the road, etc.)
It could be like all protected tunnel ways for autonomous vehicles like the bots in Amazon or Alibaba warehouses. Or could be that we have better monitoring systems on the highways and roads to send standardized signals to ev vehicles (wirelessly about the road conditions ahead). Further yet, we have to go back to those classical road congestion equations to figure out better regulated intersections, it could all be like a semiconductor chip, so that the populated high traffic areas have gates that direct traffic into that direction over priority of the reverse oncoming traffic, in other words, one way road, without turns, the other slower paths for the lower traffic cars, this ultimately facilitates deliveries and major travels, but hinders minor travels as in going to a friend's house versus going to the airport. It would be slower to go to that friend's house or party somewhere. And so these gated intersections that direct traffic are time compensated to switch open these intersections allowing faster routes to the night club at night, for example.
Completely agree, smart roads need to be created for smart cars. Inventing human AI to drive cars on asphalt roads is overkill.
@@superheaton most of your points are bad but I like the passion
Elon in 2022: Twitter is way harder than I thought
Twitter itself isn't hard... All the haters and targeted media is hard...
@@blondymonk1535 The users are part of what Twitter is. Twitter isn’t just a code base or other companies could replicate it.
Living in bith Europe and the US, here's a fun take: Europe has much more traffic rules and signs which would make it much easier for a self driving car to determine rules on the go (e.g. there are no four way stops). However, US roads are much wider, more geometrical and with better visibility. Now, moving one system to another environment will be tricky and that's why there will be multiple winners in self driving cars race.
Cars aren’t the future
I’m 56 years old and I can tell that my brain isn’t able to process as well as it used to when I’m riding high speed on my mountain bike so it’s truncating the data in order to keep up.
How do you know it's truncating data?
@@arnavrawat9864 are u a mtn biker? Ive ridden for 25 years applying that knowledge as I pick lines, timing of braking, body memory, body position are all mental intensive then you add more speed and things coming up quicker makes me less fluid. I find myself having to thinking less and react more as processing can’t keep up but I still love the thrill.
@@andrewrivera4029 bruhhhhh.
Mad respect to you, really. But I think you are mistaking the integration of the experience into your nervous and entire body with getting old.
I think your brain has to work less and in different ways because the act of riding a bike for you is becoming part of every fiber of your body.
Muscle memory. Intuition. Etc.
When I was a young boy I had to think just to walk. As I grew up tho less thinking was required. But that wasn’t because my brain was atrophying.
@@lukemarlowe6930 absolutely, my perceived decreased high speed brain function is due to age. I’m not a optical or neural expert but the pics we see with our eyes have been corrected by the brain to cover blind spots so its becoming more apparent that my brain no longer does it as quickly, I just used the word truncated to describe that.
It's likely that your brain just has more self preservation based predictive modeling that is slowing things down. A thicker tangle of data to sift through before action can be taken. A young brain doesn't have as much data to work with. We used to call it wisdom.
What I think he's running into with FSD that making it harder than he thought it would be is the other things that influence our driving that are intrinsically human. Everyone has close calls when learning to drive that we never forget that affect how we drive from that point on. Like, when that guys comes to a stop in a two lane road unexpectedly to make a left turn, after you rear end one, you never drive on a two lane road with no turn lanes the same again, you always cover your brakes from that point. successful FSD is more than having reaction times milliseconds faster than humans but having intuition that something might happen and being ready for it. How often when your in heavy traffic that small movements of the cars around you tell you what that person is contemplating doing and because you can interpret the other divers little "tells" give you a clue to what he's going to do in the future. So with the neural net self learning programing these sorts of things are probably required to make FSD 5th level.
and right there you nailed it, its the intuition and what that allows humans to do that computers will take years and years if not decades to be able to do, road structure is literally different on every road in america and yet our intuition allows to seamlessly make sense of it all like its all the same. that is really really really really hard for computers
@@jjg1501 right, and I think computers may never achieve this because they don't care if they die or end up a vegetable. In order to simulate that which is in humans the nural net will need to simulate the hypothalamus pituitary adrenal axis, or HPA as its known.
FSD will always be a "good enough". People are just being sold software that performs well enough. What "good enough" means is the real question. As of right now, where FSD is now, I think it is reaching that point. It's not much different from a videogame being sold to consumers in terms of the software development process.
@@johnnyb8629 reinforcement learning simulates that pretty well
Classification of objects (instanciations), context of objects, and knowing the objects behaviour types or modes.... fascinating
Anyone knows/managed to find the conference Lex mentions, about car doors, by Andrej ? Can't find it anywhere, maybe it was not streamed, or made by someone else ?
It will be interesting to watch self driving vehicles navigating through drunk people. I can imagine people standing in the road or placing street cones in the middle of the road.
Model 3 owner here, I don’t have full self driving, but I do use autopilot and 1 thing I can say is that the preservation and safety of the car trumps all other functions, if there was a human stood in the road, alarms would triggered in the car and even if you didn’t action it, the car would stop way before potential collision
Better than this guy: th-cam.com/video/nxTdwpz0nFw/w-d-xo.html
@@SPECCommanderShepard How does it perform approaching a red light? Especially a red light that hasn't stopped any other cars in front of you.
Self driving cars seem like a problem that would be easier to solve if all cars became autonomous at the same time.
They could be networked on like a Battlenet type system where paths, speeds, destinations, etc. are shared between all vehicles within an 'air-space' - and decision making is shared between the nodes (slowing down or changing routes to avoid convergences, etc.)
From a privacy perspective this sounds quiet horrible to be honest.
@@peterschmidts8245 meanwhile the phone in your pocket is tracking every word you say, and every move you make.
Privacy and 'off the grid' is relegated to antiquity at this point.
Yep, however self driving will have to be shown to be good enough on the roads as they are before mass adoption. Mass adoption will open a new path for optimization.
"harder than I thought"...story of my life
"it's harder than I thought"
That's not what she said
Story of Elon
What Lex said about concepts really interesting, and has some interesting implications.
I wonder if the hard problem to consciousness was solved by evolution as a side effect of the compression it needs to do because of memory limits. Because a concept seems like it could be utilized by a brain as a sort of way to boil things down to the smallest possible operatable unit, and the best algorithm to process a concept is a conscious one. I don't think consciousness is possible without concepts. At least this idea suggests a possible origin of consciousness.
This was suggested by Yann Lecun on a podcast, it might have also been on Lex's im not sure
Try to notice every thought that goes through your head for a few minutes and you start to realize what consciousness really is.
@@jimj2683 What's really strange is that a state of awareness without any subject of focus or orientating concept is perceptually indistinguishable from acute unawareness. The only difference is that the former allows for consciously passive observation while the latter is more akin to a sleep state where the conscious mind struggles in vain for recollection. This of course can only be felt by latching onto our perceptual extremeties, for lack of a better term, as a reference point as opposed to whatever passes through them like normal. It's truly a bizarre thoughtspace. Just attempting to describe it is absurd. It's akin to attempting to look directly at one's own eyes unassisted while at the same time gazing through a mirror to see their reflections.
@@trajectoryunown you just couldn't wait to write this pile of bs up. Come on bro
Yep did my dissertation on deep reinforcement learning, imitation learning and offline learning with computer vision for autonomous vehicles (using Carla) and fuck me it was difficult just for simulation of the problem.
Secondary insight basically.
A computer would have to be faced with every single scenario from every possible camera angle with every single available outcome in order to avoid accidents aswell as roads and marking simply not of high enough quality that cause issues with navigation also.
For accurate vector-space don't only rely on one cars sensor's. All vehicles should be in constant network with each other, satellite 🛰️ imagery, and sensors from posts, curbs, and perhaps even the streets themselves. Reimagine driving altogether don't just confine it to current manual driving.
Is there somewhere a source for the MIT car door talk?
What about the windshield wiper problem? Can’t solve self driving until the wipers work!
Love how honest Elon is about challenges….falcon doors, self driving etc….just says it like it is.
He's not a liar.
now he tells its hard :D he is a liar that said its easy to make it to mars and the moon, but NASA pays him to do stuff, because he failed. Just like every other stuff he made. Maybe Elon thinks its all like a video game or movie!
@@jarniskat You're employing the straw man argument logical fallacy.
No, he hyped all his projects all those years, stock grew(and his wealth) mainly on this hype, now when it's obvious to anyone that he didn't deliver on time and is already years behind his delivery estimates, he says basically obvious things (it's hard, it will take much longer time etc). Soon he will be saying the same about Neuralink and Mars. He is good at marketing. This was and is his thing. Others who are working on similar tech were honest, weren't hyping and are more or less in the same place as Tesla is in terms of technology development(autopilot, EV quality). Eventually his companies will deliver but other companies also will, and they will do it without all those ridiculous dates and hype to increase their stock and make themselves rich fast.
@@Mr__Singularity Very well said. I hate his exaggerations but at the same time, that hype, those exaggerations propelled the electric car industry forward. Other companies would not be making electric cars if it wasn't for him. Governments would not have passed laws against regular cars without his success.
The hardest part of this problem to solve will be the moral decisions humans have to make. Until a person can get in a car and feel like the car will make the same decisions they make, they will never feel entirely safe. They don't want the most efficient math for human survivability, they want the car to always do what they would do.
This means there won't be a one size fits all, and I think some sort of learning algorithm that understands how people drive and profile different approaches will have to be an eventuality.
Cool point. Although I think our morals are highly susceptible and certain difficult decisions might take more than a split second or even forever to be made in accordance to ourselves
We definitely just need a rail system.
First step will probably be replacing truckers with the technology. That's where the biggest financial incentives are.... by the time it's available for consumer use it should be well tested. Assuming they ever manage to build a full featured system.
Been recently doing my license in Germany. Thought if tesla could encode the German driving rules, they would have FSD in no time 😄
Do the German driving rules have anything about deer at night, or Moose standing in the road, or rockslides?
@@ka9dgx actually they do. If you See an Animal on the road, there are rules how to behave 😄
@Top Lobster You are advised to NOT BREAK and basically try to keep a straigh line while ramming trough whatever there is on the road..
@@Labix98 thats not true
@@Labix98 You are advised to HIT THE BREAKS and NOT TRY TO DODGE the obstacle (e.g. deer) since that often results in worse accidents.
When he is talking about forgetting the most things possible. He is not talking about long term memory. He is talking about working memory, yes the brain has working memory that was discovered with the help of what happens when there is a brain that does not work correctly, through observation of different mental disabilities.
You can test how much working memory you have by trying to remember a random string of numbers in a set amount of time the amount of digits you can remember without special technique is around 5-8 for most people. If one wants to remember more you have to forget something else or get that into longtem memory.
Those 5-8 memory slots can be any abstract concept, so if you have an abstract concept that is the number "123" then that number only takes up 1 slot in memory.
That's how the brain can work on complicated things by abstracting from them so forgeting as much as possible is forgetting the details and replaceing them with good abstract concepts,
those take less space (in working memory) so you can remember more that is why people often can't remember the details off stuff that are not in long term memory, because they just think about the concept.
For example it does not matter what shape the tiles of the sidewalk have.(you will just remember there was a sidewalk and if it was just and ordinary side walk you likely won't remember how big the Tiles were unless you Fokus on them in which case You put the shape of the Sidewalk Tiles into longtem Memory)
@@M00SEN420 did a little bit of reformatting hope that helps
I think Musk is underplaying the control problem aspect in likening it to a video game. In a game like GTA or Cyberpunk the world’s control logic is primarily self-consistent and is designed to respond well with respect to its own mechanics. In such games traffic is very well managed because AI is reacting *to* other AI *with* ideal vector knowledge. If you solve the vector knowledge gap, you still have to resolve the issue that not every other agent on the road can be safely assumed to operate under logical principles that are consistent with your AI’s driving behaviors (and often not consistent with the law). In a game like GTA there’s not much impetus on the AI to have a good reaction to a player on a motorcycle driving into oncoming traffic, if there’s any impetus it is for the AI’s response there to be either (1) entertaining or (2) a realistic approximation of how a human driver might respond. An autonomous vehicle (even with ideal vector information) should neither optimize for entertainment nor for the emulation of human behavior. That’s a much deeper problem than Musk has credited and you can see why Lex mentions the game theoretic element (though this also still suggests the flawed assumption that all other agents on the road act rationally).
He's saying that the problem of programming traffic rules is trivial when you have ideal vector knowledge. He is not saying that self driving is comparable to video game AIs
I think that as long as the vehicles behave according to the law of physics they can at least partially be predicted
@@olemorud8362 He directly alluded to Cyberpunk and GTA and how these games manage traffic, which is the exact reason I cited them. Even if he hadn't, it wouldn't nullify the points I raised. There's more than the codification of systemically consistent traffic rules. One has to account for agents within the system which are not operating under the same codified sets of behavior.
@@francescogiacomelli403 the laws of physics won't tell you what someone means when they flash their high beams at you and vector data alone won't always clarify intent. The other operator may be signaling (for example) that you may merge or enter an intersection even though they have right of way. Detecting this is easy, disambiguating this intent from all the other candidate options is non-trivial and requires context. Moreover there is the implication that all of this is timed (if the other vehicle's operator sees no response within a few seconds they will generally reneg on their offer and begin moving). Kinematics isn't sufficient to ensure operation of the vehicle that is both safe and effective.
@@mattreigada3745 Even with accounting for accidents and bad drivers it's fairly trivial to hard-code traffic rules assuming you have perfect information about the environment. Getting this information is the non-trivial problem.
I think the issue here is one of expectation - an older driver gets the pedals mixed up and plows through a restaurant, and it doesn't get a huge reaction, what are you going to do, but if a self driving car gets into an accident, it's treated quite differently, as a reason to stop trying. If the number of just lowered, that's seems good to me, but people expect 0, perfection.
totally agree.... this is a barrier for progress everywhere!
Or, when we let clueless teens get behind the wheel when learning - people barely think about that. In reality FSD could already be considered safer because there's someone at the controls who can step in at all times.
@@ultimatewick yep. Elon says straight out FSD is already insanely safer than humans. it's just we expect it to be 0
I think his plan is to sell them to China or more censored/risk taking societies...where they will be seen as investments for taxi services.
@@timb4248 it would be very american to farm all the risk out, bring it back after
I wish the car would remember construction zones better.
My car refuses to remember that a lane is closed for construction and keeps wanting to get into the lane. I drove this route every day.
I don't think they've solved the memory issues. Hopefully in the future cars will be able to communicate with each other and give a heads up if, say, someone is crossing a street.
It would be cool and efficient for cars to help each other out so they aren't constantly going blind into a location. Nothing becomes a sudden realization. Like an ambulance a few blocks down heading towards you. Just let everyone know the general direction and have the cars start clearing the path.
Anybody got a link to the greatest talk on car doors lex is referring to?
what is a vectorspace?
I personally can’t see self driving cars, with little to no human eminent, becoming mainstream. The only way is if every car changed to this way of driving simultaneously therefor removing the human factor/variable from the road grid.
Yeah those motorized cars... who needs them? Horses are way more reliable!
@@FuriousImp I personally like the idea of self driving cars. The problem will always be the human factor out on the road. The unpredictability of humans will always cause problems for an A.I.
Even though I drive for a living, I’d love to see a simultaneous rollout of automated cars. It’d take the hundreds of idiots I see off the road instantly.
You've hit the main problem, the roads are completely an uncontrolled environment.
Those painted lines on the ground isn't going to stop anything from crossing into it.
Musk is using the term vector space incorrectly.
Often, humans use other senses like hearing to make decisions on the road. For example, if I hear a siren of an ambulance or fire truck, I get ready to do different things depending on the situation, like stopping on green light to let the emergency vehicle pass, or pulling to the side and slowing down (or even completely stopping) if that vehicle is moving in the same direction. There are multiple other situations like that. Is Tesla's computer going to be able to do that? There is also a lot of human factor involved in driving, like being able to predict what people could do on the road. It's not just about following the distance, the traffic rules, and seeing objects on the road.
So why are they still marketing it as "Full Self Driving" - When it clearly cannot and is not allowed to fully self drive?
It’s a set of features that is available to whoever buys it, when software evolve or new ones become available they get sent for free to whoever purchased FSD.
The completely autonomous feature, you got to admit, would be mesmerizing and we could argue if that is gonna happen. There’s also a way to bet on this by shorting or buying TSLA stock, when it goes up fsd is becoming reality and when it tanks it’s losing credibility
@@francescogiacomelli403 Imagine paying 40k to be a crash dummy in a Beta test. Pass.
Vehicles are an extension of ourselves, You have to be able to read body language to be a decent driver. If I see an elderly person in front of me at a stop light, I probably know what to expect when the light turns green, versus, a younger person... To many examples to mention, but it is way more complicated than programming some software about distance, and space etc.
You have to read the body language of a car...
But the beauty of using an artificial neural network is we don't program the software explicitly for distance, space etc... It learns them. As well as it can learn to interpret body language for predicting possible outcomes, just like you did.
@@TheSCPStudio not just the car, but the millions of variables around you as well as an ever changing environment. A cone, a speed bump, a cat, a billboard, a puddle, a fallen down tree branch, a person, a cardboard box, its just say too much to process and program, especially a computer.
@@DensityMatrix1 they are programmed to learn right? So what direction they learn in still has to be programmed. Or am i mistaken
Hey man just wanted to say like the car in your case would treat the situation of the older gentleman jus tteh same as you. When a car is going a certain speed it will follow behind a certain distance this could be coded and it would be universal for 100% of siuations where u are pulling away from a stop with someone in front of you.
Do NOT take a shot every time Elon says vector space
9:40 in other words, driving wisdom
This was cool and interesting.
Nice clip
lex fridman's demeanor is like that of a character in a film.
Advancing causal learning and combining it with statistical learning.
The more self-driving cars on the road and the more we transition our roads and "road world" to fit these cars, the better and safer they'll be. There will likely come a time in only the next couple decades when cars on the road are constantly communicating with each other as well as features of the road environment also. The rate of car crashes will drop to nearly, and travel times will reduce due to increased "speed limits" and increased efficiency of intersections, on/off ramps, etc.
7:38 that's a very good point. Your mind is doing prediction error for most part
If only there was a way to get accurate measurements of how far an object is which can help make a more accurate vector space, that uses light and other companies like google and apple use for their self-driving models.
Nice. I think this will be why Tesla's vision system never gets out of beta. I believe they cannot upgrade to LIDAR because of Elon's public promise that all currently made Tesla's could and will reach FSD using current hardware, otherwise class action lawsuits and a terribly hurt ego. This stupid gaff is shooting him in the foot.
This is the first time I understood clearly what a vector space is.
Don’t you love it when something finally clicks? Double slit experiment blew my mind for a hot minute, now I get it.
Except that's not vector space is.
@@kari8187 I had a double slit experiment in college
@@oblomist yea lol like that's the description that got him to get it? Haha such a bad explanation.
A vector is a moving object, so a vector space is a space with moving objects
Fascinating stuff
If the engineers of self driving cars could also engineer the roads they operate on, excluding manual drivers, it would be easy. I don't think you should mix self driven cars with manually driven cars, period.
Agreed, so lets keep the man operated vehicles and ditch the "self" driving nonsense.
@@canoninmunaone I can see a purpose for self driving vehicles, but I see the need as more niche than the norm. I am 60 now and HATE, HATE, HATE flying. I would opt for trains except trains are not comfortable, expensive and have limited service. A self driving vehicle could provide a nice option against airlines. But I would not ever want to trade my manual truck and the freedom it provides for convenience. We can have both without having both on the same roads. Also, creating separate roads for self driving cars, as well as recharging stations, hotels, restaurants, etc, would create a huge amount of jobs.
Almost like a automated driver-less mass-transit rail network (that already exists). Just for resource wasting single moron? Cool!
@@mt3ccd only in very, very limited areas. Not a realistic option for MOST people. Plus I am not in favor of giving up my autonomy completely. We CAN have both easily. EVs are not a reliable option to every circumstance. Forcing compliance is facist behavior. Pass.
Many lidar based autonomous vehicles have been tested, as you say, on restricted roads, Tesla aims at creating a service where consenting individual can be brought somewhere they asked to in a safe, cheap, reliable, sustainable, socially aware, efficient, modern, attractive, compelling and sometimes outright needed manner
this an absolutely AMAZING conversation about the brain and vision, anyone know some things worth reading if I wanted to hear more about this kinda stuff?
The Bible, no joke.
@@e.vangelios7285 "have you heard about our lord & savior"much 🤦♂️
@@ottebya Do you understand it?
@@e.vangelios7285 my brain would be very happy watching the bible burn. You might be right
Honest question; what did you think was amazing? He is pretty much summing up some facts I learned in biology classes I had in high school. The brain is complicated, does a lot of work, eyes have blind spots etc. He is not a really fluent talker, and to me it seems that gets intepreted like he thinks really deep. I got a totally different vibe from the convo.
A book about the brain that you might like is 'I am a strange loop'! Really loved that.
Elon looks stressed tf out. Stay strong Elon.
That new haircut looks terrible - makes him look like one of the 3 Stooges.
He looks more rested and relaxed here (and all through December) than he had for a long, long time.
@@mariomeza3514 Yeah he's just getting older I feel
not as much as before. like 2018. But man look at him, running two multi-billion companies concurrently and has like 6 boys.
@@MattGarcyaDC didn’t ask you shit two weeks ago.
See this is why an Elon Musk type can make make self driving. No car company is up to the task because it would cost too much with the reality that it is so hard they will probably give up before they achieve it, so it would be a failure or a wast of money. It takes being one person's goal with the resistance to quit while still being able to afford going forward towards the goal.
6:50 I think the brain breaks things down to concepts, and we have a large store of concepts in out unconsciousness memory, and we can tie them together to aid in making decisions.
Self-driving is not the key. The advancement of public transport could be the solution of so many problems like;
not enough space in cities
much less carbon emission
less individual costs (for commuting)
less need to build new streets (which leads to less costs for tax payers and less carbon emission)
less traffic jam (leads to less time spent on the roads =less maintainance costs...)
cities could be more pedestrian friendly
...
If public transport was for freee and the transportation network was actually reliable and quick people would actually use it. The money which is usually spent on roads could be used for that
I like your comment but outside the city self driving will be key, also ridesharing/carpooling would make your list compatible with AV aswell
@@francescogiacomelli403 I agree. Public transport has always been problematic in areas with less density and cars will never disappear but cars are so toxic for daily commuting
I prefer to sidestep the self driving problem altogether and invest in transit and walkable cities
I do not know nothing about vector space but what I do know is that every and each of us need its inviolable space
in life .
Words that show depth "Control problems are not the hard part ..... [long pause]".
🤣It’s way harder than I thought so now I’m increasing FSD beta upgrade to $12,000 😭
*$15k now
Everyone expects Elon to answer basically everything, fantastic Musk
What makes you think you know everyone’s opinion?
@@MrMaxKey they just ask away no matter how complex the questions might be, real trust in his judgement and so far he lives up to it. If Putin starts fiddling about to much and US needs a solution, Elon would be the man to ask. We all have the same data but Elon concludes how it is and others do not.
@@rawstarmusic how would elon be the solution if 90% of the garbage he is trying to make is physically impossible or deemed too hard to do by the others who actually invented the concepts? my prediction is most of the shite he is peddling wont lead anywhere. hyperloop, more like hyperpoop .
@@stephenhartley2853 I am no friend of hyperloop for human transport, only cargo. Can not escape down there. But the rockets really work. The battery cars works. He aims far into the future. Neuralink is uncomfortable but might be the way we can develop human cognition, else tech develops and brain does not. If brain can develop we can solve all problems
@@stephenhartley2853 I would say Elon is comparable to a Steve Jobs in that he is more of a visionary if he thought his ideas impossible and didn’t pursue them then u may not have been able to type ur reply
what if we create road for self driven cars only
the reality is..cars drive in contact with the GROUND, the ground should be used as a 'gps' guide lane for automobiles on fast moving lanes, others self driving tech is only sufficient if your stray out of these mass transport lanes
Amazing: a CEO that actually understands and is able to explain what they're doing at the company
Check out Christian Von Koenigsegg, he is also able to explain every single thing about his company's cars.
The key is to develop a vector space.
Reminds me of 'Nobody knew that health care could be so complicated'
exactly, it's the same conman grift and the same overpromises that are never held accountable by their stans
Like creating a model of all the objects, knowing how to differentiate them, and predicting their possible future paths.
The economic problem of maintaining optimal driving infrastructure and affording expensive self-driving cars is even harder than the programming of self-driving vehicles.
Cars only transports about 1.3 person, it’s not efficient if we care about Climate change.
Cause he’s not an engineer. This is why a salesman should never make product predictions without consulting actual stakeholders.
a level 5 autonomous vehicle if turned on in space, would need to evolve arms and legs, build a space craft, and fly back to earth, where it can then pick you up at starbucks. It's hard.
The question is if FSD can be solved with the current configuration and hardware, personally I don think so, and here is why:
Vision of cameras right next to the glass similarly to vision through prescribed eye glasses get far more affected even by few droplets in front of the camera than when you looking at windshield half meter away from driver's position and hardware is not capable enough to compensate for smugged image in a way satellite pictures can be reprocessed to predict detail when smudged shape get reproduced to most likely jet fighter limited visual data point towards through machine learning trained on low resolution images of known objects, that kind of processing power in real time at least 30 frames per second is simply not there in any Tesla current car only on supercomputers and due to placement of cameras next to glass while lacking thermal cmaera to complement lack of processing power and bad sensor placement in bad weather means that government unlikely ever approve Tesla cars in current hardware configuration for full self driving without driver constantly watching over car not matter how many new Tesla FSD software updates will come.
For cameras to have decent enough visibility even during rain they would need to have comparable visibility to a driver which would be easiest to do by placing cameras between front seats (perhaps on a retractable arm inside the car activating only when necessary) or into the co-driver seat (which would make it unusable during autopilot). Alternatively and likely the most realistic solution, just put a thermal camera which cannot be blinded by incoming light sources so easily unless having a high thermal signature apart from not being affected by rain or fog very much even when mounted on the windshield or directly above it like LIDAR on current Lucid cars and enforce by law new road signs and road surface markings that are visible on thermal cameras which can't see "colors" (different visible wavelengths reflection and emitting properties) but only see differences in materials if they have sufficiently different thermal properties in order for the thermal camera to by reasonable bad weather replacement for visible wavelengths cameras mounted on the windshield that become useless during heavy rain to a point where reading street signs become unreliable to bet human life on it...
Why is Fridman asking Musk, he doesn't know.
Self driving cars can handle 99.99% of everything we encounter on the road with ease…. Its the 0.01% of the unexpected that make the problem extremely difficult
Isn't the biggest problem full scale adoption? As in, most problems would be solved through full communication between every sensor in every point in traffic.
Does not sound that hard to predict the motion of things (occluded or not) once you have a pretty accurate labeled "vector space" (hate this naming) and a few samples across time.
Using what algorithm?
maybe for asteroids/comets, but for pedestrians/cyclists in an urban environment?
I think Daffy Duck can say a thing or two about car doors from experience
I hope he's taking into account the other relevant senses. It isn't just visual information that we process. The other senses are being used in more subtle ways, at least to our conscious perception, but it's all important. You can feel vibration of vehicles passing. You can hear crowds chattering or a mom screaming at her kid to get out of the road. Not to mention micro-perceptions like changes in air pressure on our eardrums when large trucks are passing by, or our balance shifting when wind pushes the vehicle.
Beyond just basic sensory input there's interpretation and context. We know the roads are more dangerous based on the weather but also based on how the people around us are currently responding to it. We know when it's a Saturday night in a party town just after graduation and how that affects pedestrian behavior.
The only sane comment in this whole comment section. Also, he just blabbered high level keywords to sound like he knows the thing but looks like he doesn't want to get into the details.
@@suyashneelambugg I genuinely think he doesen't know a lot of those details. He does seem to be decent at finding other people that can figure these hard problems out, but when it comes to the real meat of stuff he seems to be clueless. Heck, in a recent discussion with George Hotz, Elon couldn't even explain what a Tech Stack is, which is like one of the most basic concept of any software development companies(Twitter in this case).
Jordan Peterson pointed this fact out in 2016.
lol. jordan pererson.
This guy could be focusing his efforts in so many better places rather than clowning around on Twitter and appealing to edgelord trolls.
thought I was about to see Elon play peekaboo with Lex . think I would have wet myself if he did because the videos over and I'm still giggling
I was really hoping they'd go off on a tanget, imagining a world where no single human ever developed object permanence
Probably thought of before, but have you tried mapping individual nearby pixels or small squares to neurons near each other to create a map much like the human brain does in the visual part? Might save some processing power. It seems like neural nets have a lot of overlap between layers and weights, I mean it may be possible to use less of them to achieve the same result. The brain has sections responsible for different agenda perhaps ML can be designed similarly, instead of just using the same inputs and weights for different things.
No front, but you’re not smarter than the whole world of academia and software companies. Of course they thought of it.
@@mynameisjeff9124 Hehe you just made me realize. I may be the smartest guy in the world, but I will never be smarter than our collective consciousness.
6:25 “you’re brain is trying to forget as much as possible” - perhaps the reason why memory is so bad is to remind us that all that exists is the present, pain memories stick to avoid future danger, pleasure memories stick to remember what feels good, the rest fades
We build as individuals and a computer has to build itself from nothing but our individual understanding of other individuals.
The hard part is that you never know how close you are until you reach "good enough" or "there". The last 0,0001% can take a very long time and get you to give up.
Just have the cars communicate with one another.
All of the situations that require the car to stop NOW are underrepresented in all the training data, there has to be a huge effort to rebalance this by having a team dedicated to tricking the car into accidents. Like a GAN, but with humans on the other side.
If Musk now thinks that self-driving is way harder than he thought, then his original claim that Teslas had everything for L5 FSD was not true. The question is, why did the CEO of a company that was massively in debt claim they had solved L5 which later was admitted by Tesla as being L2? Either he was being untruthful or is not a genius. Or both.
Lol, half the time we look at our phone while we drive!
never has any made something so simple so complicated
Change the road. Sensor's.
i’d hate humanity if youtube convinces lex to put these behind paywall
That's all well and good, but they need to spend more time and resources nailing the basics. My car still desperately tries to hit curbs and accelerates way to hard out of turns. Making decisions about when to make a then also is something that seems like rocket science for the car. Quit worrying about the exotic stuff and nail the basics.
Autonomous driving seems unsafe because we will never perfect it... But then you realize cars will never drive drunk or angry or will be texting..
"Self driving is way harder than I thought" - and exactly as hard as experts always claimed.
That isn’t really fair though. Creating a self landing rocket has experts thinking it wasn’t possible for decades as well. Nothing wrong with being an optimist and pushing the frontier of technology.
@@ThomasFoolery8 I'm not sure the comparison holds though. I don't think that anyone ever thought landing the rockets was impossible, but that it looked like a lot of effort for something that might not be economically viable.
And so far, although the demos of the landing rockets are one of the coolest things I have ever seen, it appears that it still remains to be seen whether it'll be worth the effort. I obviously ultimately should, which is why it seems to make sense, but the ultimate judge will be the economic numbers. Nobody said it couldn't be done, but questions were raised about the economic viability. And that aspect is still firmly in the 'remains to be seen' category - just as experts predicted.
Exact same problem with autonomous driving, which I happen to have some tangential experience with. Based on the original (pre-Musk) Darpa challenges, where various companies had vehicles slowly crawling through a desert at turtle speed, most not able to finish the course in the late 2000s, most experts more or less predicted that cars would be at current level about a decade later, which is where we are now. The current state: vehicles can do somewhat reliable traffic jam driving, follow road lines and a number of other enhancements are ready for general adoption, but full automated driving is far away because the systems are incapable of truly dealing with both the real-world chaotic complexity AND are completely inequipped to deal with humans who will take advantage of the AI's need to always err on the side of caution (try drive a robotaxi through Amsterdam or New York - it would never get out of the car park) - these were all known, hard-to-solve problems that look like 'the last few details' to a layperson (and Elon Musk, apparently), but looked like potentially unsolvable problems with the current technology to actual experts.
I used to think Musk was a genius, and he's obviously not stupid. But he has started to equate the output of his engineers with his own achievements. He's not an engineer. He's an intelligent manager who grasps much of what his engineers talk about and he has convinced himself he's one of them.
The fact that he's been doing that 'I don't want to be a manager, I want to be an engineer'-schtick recently actually tells a lot about his state of mind I think. He's been called out quite painfully as a failing manager due to his Twitter debacle, so he retracts in his self-invented 'engineer silo', because it's easier - and more plausible - for him to pretend to be an engineer than a manager now.
At some point, I expect the stories to come out about how he would present ideas his engineers habitually as his own. I mean, every time we have actually heard him talk about technology, he has always sounded remarkably superficial on most subjects. He has quite juvenile ideas about AI, living in the matrix, how to build hyperloops (remember his: 'it's really easy, it's just like an airhockey table' nonsense?). His public appearances are weird to watch, and really always have been. The crowd goes wild, but on stage stands a man who mumbles and stumbles and seems to be making up superlatives out of thin air on the fly - not a great stage personality.
The thing that intrigues me about Musk is how - as time goes by - he starts to sound like Trump in ever more ways, and I'm not even talking about politics now.
Something about having so much personal power apparently breaks many people; Trump and Musk are two examples of people who seem to have broken under the strain. And I'm not denying Musk must be under tremendous strain, of a kind I can't imagine. But the strain broke him.
@@damionm121 glad you wrote this up so I don’t have to. Musk went to UPenn and got into an applied physics PhD at Stanford before leaving to start his online payment company with his brother where they worked together to code it up before it got acquired by PayPal. How one can say he isn’t an engineer or all that smart is absurd. I went to an Ivy League school and work as a quant at a hedge fund and I can tell that Musk is smarter than me based simply on the fact that he got into that Stanford PhD program.
At the end of the day, op is a progressive leftie and he’s just parroting the anti Musk talking points he’s been programmed to spew by his news sources. Hilarious considering Musk single handedly ushered in the electric car era.
@@damionm121 The main reason for my 'attacks' are probably that I get the impression that Musk stopped being a 'net force for good' and has tipped into the area where he's doing more harm than good to the world.
Why didn't those engineers create SpaceX or Tesla on their own?. Well, that was probably because these people were engineers and not salespeople. That is the whole point. Musk is a salesguy, and he used to be a really good one, one that could sell fridges to eskimo's, or, as was the case with him, business ideas to investors. His great skill - and it's not an inconsiderate one - is his ability to sweet-talk people with money into buying into his ideas. There is nothing - beyond his own rather suspicious constant insistence on the idea that he himself is also a great inventor in his own right - to suggest that he's not ,mostly a guy with a gift for obtaining investment money and the dogged determination to get things done 'his way'.
He's not unlike Steve Jobs in this regard. That man couldn't invent the technology himself to escape from a paper bag, but he was very adept at telling others he wanted a cheap way to get out of a paper bag, but with a device that had nice rounded edges and needed a new type of charger. And then his minions went out and made that for him. That is how Musks cars and rockets came together as well I'm sure, and I'm also more than willing to believe that he has a strong enough grasp on all of this to understand and even contribute to discussions his engineers have when they are solving problems.
Einstein is actually a pretty good example of someone who sounds a little hesitant in his speech - especially if he uses his non-native English - but when you parse the words it's obvious he says important things. It's not the hesitation and mumbling per se, but what the words actually convey.
The problem with Musk is that his words are really only hot air most of the time. It didn't always use to be like that. Musk of ten years ago was also not a brilliant speaker, but a man focussed on what he was doing. The public Musk of the last five, six years is nothing more than a creepy second hand car salesman, promising the moon to shift the crap on his backlot with warrantees until they're off his property.
Perhaps I should have clarified that there is a young Musk and a current Musk we should be discussing. Young Musk had some bright ideas and he gathered people around him he managed to inspire to do great things. He may indeed have contributed a thought or two on the products himself along the way.
But today's Musk is really more like Trump selling NFT's of himself dressed as a cowboy or an astronaut; desperately announcing ever more surreal new products at ever crazier timelines. His M.O. is now basically: shout random shit on Twitter about things he clearly has no understanding of (but apparently the big man has plenty of time in his busy schedule to instantly reply to sometimes twenty tweets a day - how's that for efficient use of your time?), wait for the fallout to impact the shareprice of his two working companies, announce some kind of sci-fi tech (like a household robot) and pretend you're 'just about to release it next year', followed by 'it's really not that hard, a bit like an air-hockey table', and then, when people are pointing out that his idea is not really new (because he's basically always riffing on existing ideas from pulp sci fi magazines) he'll counter that his contribution to reviving this old idea is 'more profound than people think'.
Musk would have been a fine tech boss, like Bezos or Zuckerberg, quietly syphoning more and more money into your personal space, becoming richer and richer, contributing to the evolving wealth disparity in the world, and nobody would have been any the wiser. But he had to be a media personality as well to satisfy his apparently fragile ego, and the result is that people can actually see what he's about and he's clearly not in a good mental place at the moment. If I were working for any of his companies I'd be worried.
@@damionm121 Thanks for the kind response. Obviously my words are somewhat coloured by the fact that any time you voice any kind of Musk critique (which, as you can probably guess I've done more than once on occasion) is that no matter how subtle you formulate it (and I admit I wasn't being all that subtle here, although I do stand by the words by and large) you immediately get attacked by howling fans who would really want to tear you to shreds for daring to critique their god and that kind of adulation always rubs me the wrong way.
I maintain that Musk has had great personal success and that is always something to admire, but there is nothing he did that wasn't already sort of 'in the air'. Just like the world would have ended up with touch screen smartphones without Jobs, or cheap lightbulbs without Edison, or a working telephone without Alexander Bell, we would have had commercial rockets and electric cars without Musk as well. Maybe a few years later, but what's the big rush?
In the case of electric vehicles, for example, his big contribution to this space in terms of saving the planet is more than offset by his clear hatred for public transport, which would contribute more to fighting global warming than electric cars ever could. His whole Hyperloop/Boring Company side-project is clearly a (successful) attempt at syphoning off government money that was earmarked for public transport development; lightrail and things like that - stuff that would have made an actual impact on reducing car mileage. As you can clearly see from the speed at which the competition is now overtaking Tesla in terms of volume of production and reliability of vehicles it's clear that all he really did was kick the rest into gear a little. In the big scheme of things Tesla is an accellerator, not an innovator that can't be copied or improved upon. Yet it is still impressive that he was able to carve out a space for a new car company in an already overcrowded field. It's a shame he's running it into the ground at the moment.
By being vocally anti-public transport he's making his claims about saving the planet suspicious. By claiming he will be building intercontinental ballistic passenger rockets he shows he's either full of shit for marketing purposes or he has no idea how the economics of rockets work. By announcing he'll be putting housrhold robots on the shelf within two years he either shows he doesn't understand the current state of technology in robotics or he's just playing another Ponzi scheme. By buying Twitter without thinking it through and his subsequent activities at that company he shows he had no idea how social media worked. By claiming his cyber truck outperforms old gas-powered trucks, while simultaneoulsy playing the actual range performance of them close to his vest he again raises suspicions about his honesty. His Boring company is a joke. His AI efforts are mostly irrelevant when placed in context of what the field as a whole is doing. His brain-interface machine hasn't done anything (beyond torturing hundreds of monkeys) that hasn't already been published about by other groups, often more than a decade ago. He is mostly replicating research in those areas, which may or may not eventually lead to unique developments down the road.
When you list all the areas he pretends to be active in his success rate drops dramatically, but I will grant that for any individual, being able to boast TWO working companies you didn't so much found but at least put on the map by giving them focus and leveraging finance is an achievement that outstrips anything I could ever do.
I'm still most impressed by his rocket business. Without SpaceX, we would all still be looking at rockets as a solution that will be locked into disposable hardware for the forseeable future. His dogged determination to create something that looks like it came from the Thunderbirds has at least achieved something there.
Although having said that, I will also say that I have some petty things to say about what his rockets look like. He seems to have this love for making them look like forties pulp-sci fi rockets (probably the source of many of his ideas), making them look weirdly antiquated to my eyes. I prefer the 2001 style of NASA, with lots of greeblies and visible technical details. This, I will fully admit, is a very lame argument: saying his rockets look ugly, but I can't help feel what I feel, just as I appreciate the quality of Apple's products, but I actually hate their designs and UX choices most of the time.
Musk is a divisive figure becaus he's a very flawed, but brilliant person. I read an article the other day where he was compared to Henri Ford and how remarkably similar Musks career trajectory seems to be to Ford's. Both were great innovators of production processes, both were seen as founding fathers of new ways of doing business and being fawned over by lots of people, especially the rich and powerful.
And then Ford started to lose it. Admired by Hitler, and appreciating this adulation and finding lots of common ground between them, he became a nasty, reactionary antisemite who changed from someone who brought free personal transportation to the masses to a weird, rightwing bigot who wanted to force his ideas unopposed onto the world because he became convinced he was bascially sent by God to rule the world.
Musk is on that very same path.
Happy holidays! 🙂
Musk forgets that human beings have more than 50 types of perception, not just colour vision. What about sound (horns, sirenes, thunder), what about smells, what about situational awareness far beyond the street you are driving on? There is much more to FSD than just some cameras.
And that's the depressing part of "autonomous" driving
I think he knows about that
@@jamesalexander8872 If he does, why doesn't he try to take them into account? At least sound shouldn't be too difficult. Situational awareness is just impossible for at least the next 20 years, probably way more.
I hear that next Version of Tesla auto drive will have Smell-O-Vision based on your post’s comments
Maybe they making it over complicated, seems like you could beam out invisible laser sensors or something and that would detect any obstacles and you could have the car respond accordingly. But then again I’m no expert lol.
Why do we even need self driving vehicles? That's the question I'd like answered.
A family can then only need a single car instead of 2 or more. The car will drop the dad off at work. Come home to pick up the kids and drop them off at school. Then take to the grocery store. Then while she shops, it can pick the dad up from work.
That's just one example of saving money with self driving
@@nazar7700 - Couldn't you use public transport for that? Not that we have a decent system in this country, but maybe that would be a better place to invest.
@@paulkern7229 self driving cars won't replace all cars in the same way public transport hasn't replaced privately owned cars.
The point of my comment is what use a self driving car will provide. That's it.
Elon looks like Marlon Brando in Apocalypse now.
It's interesting: First we have to learn object permanence. Then later we have to learn object impermanence.
Brad Baymon wins Powerball jackpot on January 1, 2022
06,12,39,48,50, 07
I've never understood why this is where we've decided to make a minimum viable product in AI. Controlling thousands of pounds of steel with kinetic energy potential similar to an explosion. Replacing the burger flipper and retail worker is a lot easier and safer.
Other companies are already doing that
The potential is autonomous cars become safer than humans. Humans cause an accident in a vehicle every 60 seconds in the US. So the bet is AI can improve to the point that it replaces human drivers and saves lives.
@@MrLjs20 imagine believing we should keep cars.
Burger flipping isn't killing hundreds of thousands a year
@@bnmbg731 First it was about the environment, then convenience, then safety. That reasoning sounds just about right for more regulation and tyranny.
shame you didn't delve deep into the very hardest part of SDC - solving the trolley problem. we are faced with dozens of variants of this problem all the time - we chose a action that is ultimately a function of ALL our experiences we are exposed to since birth. SDC's can't hope to come close at least until the system has evolved to the point that it can experience pain.
Maybe we should renovate some roads so they are self driving friendly, blocking them from bikes, pedestrians, animals etc. then we can limit self driving cars to a limited number of roads which nonetheless get us close to almost everyplace we have to go. We can take over driving for the last 150 feet of our 12 mile commute. That should save a lot of lives until Elon finishes his job.
Waymo doesn't have this issue and is already at level 4 driving. Tesla needs LIDAR, they can't even reach level 2 self-driving.
All the technical challenges discussed here of having an AI drive a vehicle pale in comparison to solving the moral problems of having an AI drive a vehicle.
Should a vehicle ever choose a course of action that ends the owners life?
What number of people external to the vehicle need to die/be saved for that decision to be made?
How do you assign value to different people in order for the AI to decide the best course of action to take in an unavoidable accident?
With all of our differing political, religious, social, cultural views no-one can agree on what is morally "correct" so how can we expect to create an AI that makes choices that will be accepted by all of society?
If we can't all agree on the topic of abortion then how can we ever agree what value to assign to a woman in her first trimester of pregnancy so the AI can decide whether to hit her vs another person or persons?
I really don't see AI being used in any sort of truly advanced way until we first solve a whole host of moral and philosophical problems first.
A little bit of a moot point because I would imagine 99.99% of all dilemmas like this... Usually the right answer is to brake. And to come up for an answer for that remaining percent are situations human beings are rarely put in anyway. So definitely not the biggest problem to tackle in self driving AI
The prime objective is don’t crash. Decisions like this do not have to be made.