|
The one issue of cars communicating with each other is bad faith actors. What if I hack my car so everyone else defers to me?
|
#
?
Jan 5, 2016 03:43
|
|
|
|
| # ? Jun 1, 2024 23:49 |
|
|
There is a fascinating talk on self-driving cars from NVIDIA going on right now: http://livestream.com/accounts/11530022/events/4599293 Probably better to watch the replay right now since we're 45 minutes in, but it's pretty interesting on the state of the art. Paul MaudDib fucked around with this message at 03:46 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 03:44
|
|
|
McDowell posted:The one issue of cars communicating with each other is bad faith actors. What if I hack my car so everyone else defers to me? You cause a pileup and go to prison? I really can't think of a reason to do this. Hell, you can hack cars to kill their occupants now. https://www.youtube.com/watch?v=3jstaBeXgAs There are well-known ways to secure systems like that. There just needs to be a reason to bother.
|
|
#
?
Jan 5, 2016 03:51
|
|
|
KillHour posted:More durable, too, when you're talking about sensitive electronics. They have to be able to survive an assembly line and shipping, which is more than you can say for most test/research equipment that lives its live a spec of dust or errant vibration away from being a million-dollar paperweight. That's still probably way less durable than what's required for regular use.
|
|
#
?
Jan 5, 2016 03:53
|
|
|
No a good hack just gives your car a priority status reserved for emergency vehicles. In the self driving car world no one pulls over for the ambulance, it just 'swims' through the stream of traffic.
|
|
#
?
Jan 5, 2016 03:55
|
|
|
McDowell posted:No a good hack just gives your car a priority status reserved for emergency vehicles. In the self driving car world no one pulls over for the ambulance, it just 'swims' through the stream of traffic. This would be trivially easy to detect and would be treated the same way as someone driving through intersections with fake police lights nowadays.
|
|
#
?
Jan 5, 2016 04:07
|
|
|
Why are we trying to invent things that people already do with current vehicles?
|
|
#
?
Jan 5, 2016 04:12
|
|
|
Let's not trivialize the complexity and risks of large critical networks.
|
|
#
?
Jan 5, 2016 04:32
|
|
|
McDowell posted:No a good hack just gives your car a priority status reserved for emergency vehicles. In the self driving car world no one pulls over for the ambulance, it just 'swims' through the stream of traffic. You can do this now. https://en.wikipedia.org/wiki/Traffic_signal_preemption It's extremely illegal.
|
|
#
?
Jan 5, 2016 04:34
|
|
|
I'm pretty sure that if the system is wired well enough that there's a monitor on it that you'd have to initiate a false emergency report for it to coincide with with traffic change which has it's own bundle of "no you loving did not just do that". I know personally I'm expecting that some time within the next 10-15 years we're going to need municipality based air traffic control (for drones) and piggybacking general traffic automation on that at the same time seems logical.
|
|
#
?
Jan 5, 2016 04:43
|
|
|
McDowell posted:The one issue of cars communicating with each other is bad faith actors. What if I hack my car so everyone else defers to me? There are robust algorithms that resist this type of attack, and of course you can always double check your neighboring car with your own sensors and worse case scenario, just go lone-wolf. But yea the ambulance scenario will be an interesting one. Of course all this poo poo is at least 50 years away, even if the technology to do it exist today.
|
|
#
?
Jan 5, 2016 04:56
|
|
|
Powercrazy posted:There are robust algorithms that resist this type of attack, and of course you can always double check your neighboring car with your own sensors and worse case scenario, just go lone-wolf. But yea the ambulance scenario will be an interesting one. Of course all this poo poo is at least 50 years away, even if the technology to do it exist today. It'll happen a lot faster than that. There's too much potential money at stake, even if it's just to be gained by firing all the delivery and taxi and truck drivers.
|
|
#
?
Jan 5, 2016 05:03
|
|
|
I don't know if you would be able to go lonewolf since automated freeways could go in the 100s of MPH. There was an ARG for Halo 2 called 'I love bees' which gives us automated highways, smartphones, NSA spying, and 3d printer/fabricators A spy gets killed when his car gets remotely locked into manual on a freeway. There is a dystopian aspect here with the loss of autonomy. Why buy a car? I would only accept self driving vehicles as part of a socialist program of a public fleet giving personalizied travel. Lyft has some new scheme of renting cars to drivers. I say rubbish - you either bring your tools or they are provided - you'd be a fool to rent from the employer.
|
|
#
?
Jan 5, 2016 05:07
|
|
|
You can now watch the replay of NVIDIA's self-driving car presentation. Skip to 26:00 if you don't care about the hardware.some sort of fish posted:It should be noted that these sort of sensors had, afaik, almost no large scale commercial application until quite recently. The latest buzz in the valley is interconnectivity, "the internet of things", and drones, which has cratered the cost of cameras and other sensors as companies cram them into everything they can. There's going to be a Qualcomm demo tomorrow at 3pm ET, and to quote The Verge: quote:Qualcomm (12PM PT / 3PM ET) As LIDAR gets miniaturized that would be pretty awesome on drones. And it will probably be essential for the larger ones to navigate controlled airspace safely. Where must we go, we who wander these wastes, in search of our better selves? --the First History Man The chick bawling at 3:17 is fantastic. Paul MaudDib fucked around with this message at 05:29 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 05:22
|
|
|
Hieronymous Alloy posted:It'll happen a lot faster than that. There's too much potential money at stake, even if it's just to be gained by firing all the delivery and taxi and truck drivers. True. But there is a lot of institutional inertia behind cars, even outside of America. Though I could potentially see some enterprising European countries start doing it before my arbitrary 50 year timeline. McDowell posted:I don't know if you would be able to go lonewolf since automated freeways could go in the 100s of MPH. Doubtful. Your highest fuel efficiency is going to be below 100mph for pretty much all cars, not too mention the stability and maneuverability issues at high speeds. ate shit on live tv fucked around with this message at 06:02 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 05:58
|
|
|
Powercrazy posted:Doubtful. Your highest fuel efficiency is going to be below 100mph for pretty much all cars, not too mention the stability and maneuverability issues at high speeds. There's all kind of improvements you could make to stability and maneuverability, though. Sloped, wide-radius turns, grippier road surfacing (since you can guarantee semis won't be wrecking your high-speed lanes), etc. Arguably you could evolve tech such that the distinction between drafted road vehicles and detachable high-speed rail cars would start to fade.
|
|
#
?
Jan 5, 2016 06:10
|
|
|
Nevvy Z posted:Is there any reason why self driving cars wouldn't limitedly connect to one another anyway? Things would start getting crazy fast and efficient, especially for complicated high intersections in rush hour. Because it requires the entire industry to cooperate and come up with a comprehensive universal standard and protocol so that they can all handle behaviors that are an order of magnitude more complex than solo self-driving in the exact same way, while all using the same algorithms and behaviors and such? It's not like you can just slap a Wi-Fi router on the top of every smart car and watch cooperative behavior emerge out of thin air.
|
|
#
?
Jan 5, 2016 06:31
|
|
|
Main Paineframe posted:Because it requires the entire industry to cooperate and come up with a comprehensive universal standard and protocol so that they can all handle behaviors that are an order of magnitude more complex than solo self-driving in the exact same way, while all using the same algorithms and behaviors and such? It's not like you can just slap a Wi-Fi router on the top of every smart car and watch cooperative behavior emerge out of thin air.
|
|
#
?
Jan 5, 2016 06:42
|
|
|
I think the problems with this are more get government out of my medicare kind rather than technical feasibility.
|
|
#
?
Jan 5, 2016 07:01
|
|
|
Powercrazy posted:True. But there is a lot of institutional inertia behind cars, even outside of America. Though I could potentially see some enterprising European countries start doing it before my arbitrary 50 year timeline. It's probably not going to be Europe or the US. Japan wants autonomous taxis for the 2020 Olympics, and Nissan has said they want their fully autonomous vehicles on the road within the next couple of years. Both of those are probably optimistic, but unsurprisingly I think we're going to see major innovation from countries where preserving jobs is way less important than dealing with a demographic crisis in the workforce. I have no doubt that there'll be significant resistance here, with the long-term effect of further suppressing wages for workers in the transportation field by putting an absolute cap on their value. Anyway, Lyft and GM are partnering up to try to build on-demand autonomous cars for the future, and to allow Lyft drivers to use rental fleet cars in the present. It seems like every major automaker at this point is throwing at least some weight behind autonomous vehicle development.
|
|
#
?
Jan 5, 2016 07:22
|
|
|
Paul MaudDib posted:The chick bawling at 3:17 is fantastic. the giant pileup from 1:57 on is actually from ohio, i dunno why it's in that video
|
|
#
?
Jan 5, 2016 18:03
|
|
|
some sort of fish posted:Ya an entire industry forced to create communication standards across disparate devices with different capabilities we've sure never done that before lol Yeah, just look at how quickly and seamlessly industries have cooperatively come up with entire communication standards from scratch, with full and reliable interoperability and no problematic vendor-specific bugs or misbehaviors or extensions, and the stability necessary to ensure that hardware will not be rendered obsolete over a timeframe measured in decades. It's certainly possible, but it's not easy, especially when there's currently not much cooperation between prospective smart car vendors. Compatibility isn't something we can just assume will happen, and even if it does, it's not like proper behavior will just arise out of thin air. Spontaneously designing cooperative behavior through ad-hoc peer-to-peer communication requires the car to understand an order of magnitude more about its surroundings, and requires far more complex decision-making. Main Paineframe fucked around with this message at 23:03 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 23:00
|
|
|
No it doesn't. Case in point: swarming is an emergent form of cooperation that requires no coordination. Similarly, a bunch of cars all independently trying to maximize their gas economy will exert a smoothing effect on traffic. Yes, there will be standards that emerge over time. Your phone doesn't refuse to talk to an access point just because it's a different brand. There will be some consortium that comes together and defines a protocol. Also, do bear in mind that the FCC just came down on hotels spoofing the WiFi protocol to jam hotspots. There will be other vehicles and your own set of sensors, integrating all those signals and rejecting erroneous data is an inherent part of whatever system is doing the occupancy mapping. Paul MaudDib fucked around with this message at 23:30 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 23:26
|
|
|
You guys are way over focused on self driving cars working or not and not the bigger picture
|
|
#
?
Jan 5, 2016 23:49
|
|
|
But...but, snowstorms.
|
|
#
?
Jan 6, 2016 00:28
|
|
|
Main Paineframe posted:Yeah, just look at how quickly and seamlessly industries have cooperatively come up with entire communication standards from scratch, with full and reliable interoperability and no problematic vendor-specific bugs or misbehaviors or extensions, and the stability necessary to ensure that hardware will not be rendered obsolete over a timeframe measured in decades. CalmDownMate posted:You guys are way over focused on self driving cars working or not and not the bigger picture
|
|
#
?
Jan 6, 2016 04:12
|
|
|
some sort of fish posted:What bigger picture? The threat of automation is about as scary as the invention of the automobile. The automobile was terrifying initially. http://www.smithsonianmag.com/innovation/when-pedestrians-ruled-streets-180953396/#uEssMB3oSY5qyo1U.99 Smithsonian posted:"At the turn of the century, motor vehicles were handmade, expensive toys of the rich, and widely regarded as rare and dangerous."
|
|
#
?
Jan 6, 2016 05:16
|
|
|
some sort of fish posted:What bigger picture? The threat of automation is about as scary as the invention of the automobile. By and large the people running around in circles about automation ending huge swathes of labor in the near future usually are clueless about current ai and machine learning trends. The few who do tend to be extremely optimistic about the current trends of rapid advancement and also don't really have any experience working in potentially affected industries. For instance people keep saying truck drivers are going to disappear, but they do a bunch of other stuff on site during delivery that isn't worth automating. So maybe we need fewer truck 'drivers' but the profession won't change all that much other than being able to sleep while on the freeway. Personally I think current ml stuff is far too domain specific to be the existential threat people make it out to be. Wasn't the automobile invented a mere 20 years before the biggest economic collapse in the history of the modern world? Do you guys even comprehend the concept of diminishing returns? I mean walk around scratching your heads and wondering why the us has lost so many manufacturing jobs and then say automation poses no danger to the future of our economy or society.
|
|
#
?
Jan 6, 2016 07:07
|
|
|
The 19th amendment was passed under a decade before 1929.... coincidence???????????! BrandorKP posted:The automobile was terrifying initially. Like sure this tech will be disruptive, but that's every technology we ever invented. I don't see how this is uniquely disruptive, especially given the intensively domain specific nature of these techniques.
|
|
#
?
Jan 6, 2016 07:30
|
|
|
some sort of fish posted:Like sure this tech will be disruptive, but that's every technology we ever invented. I don't see how this is uniquely disruptive, especially given the intensively domain specific nature of these techniques. People are talking about self-driving cars a lot because they're a big, obvious case of automation that will be almost immediately disruptive when/if they enter use. That's kind of irrelevant, though, because the real issue is just going to come from a continuation of current trends. Like, yeah, fully automated cars would have a very obvious and very negative impact on transportation workers, but even partially automated vehicles will reduce the value and over the long term suppress the wages of those workers. This is basically what's been happening with automation for the past couple of decades, at least. Jobs aren't outright destroyed, they're just polarized in a way that's pushing along the hollowing out of the middle class.
|
|
#
?
Jan 6, 2016 09:21
|
|
|
some sort of fish posted:What bigger picture? The threat of automation is about as scary as the invention of the automobile. By and large the people running around in circles about automation ending huge swathes of labor in the near future usually are clueless about current ai and machine learning trends. The few who do tend to be extremely optimistic about the current trends of rapid advancement and also don't really have any experience working in potentially affected industries. For instance people keep saying truck drivers are going to disappear, but they do a bunch of other stuff on site during delivery that isn't worth automating. So maybe we need fewer truck 'drivers' but the profession won't change all that much other than being able to sleep while on the freeway. Personally I think current ml stuff is far too domain specific to be the existential threat people make it out to be. Trucking jobs won't completely disappear overnight (they probably won't ever completely disappear), but like the poster above me said there will be fewer and fewer transportation jobs and truckers will be forced to compete for them by accepting lower salaries. Many will try to get one of those non-automated delivery related jobs that you mention, which will drive down the price of that work too. I don't know where you're getting the idea of ML being too domain specific but let me respond from an affected industry. I have a language degree and I used to get small translating jobs. Google Translate killed most of those jobs. I also used to real time interpreting and software like Skype Translator will soon start to make a dent in those jobs too.
|
|
#
?
Jan 6, 2016 11:04
|
|
|
Yea it is going to make a set of specific tasks easier to do, making it like the car and the cotton gin and the steam engine and the million other things we invent to make our lives easier. The 'it' I am referring to is not just self driving cars, but the general field of software that has driven people into a panic over ai and robots taking all of our jobs. Doctor Malaver posted:I don't know where you're getting the idea of ML being too domain specific but let me respond from an affected industry. I have a language degree and I used to get small translating jobs. Google Translate killed most of those jobs. I also used to real time interpreting and software like Skype Translator will soon start to make a dent in those jobs too.
|
|
#
?
Jan 6, 2016 15:01
|
|
|
I think there needs to be a new AI test. It needs to be able to 'watch' a tv episode like a sitcom. If it can recognize faces and understand dialogue it should be able to produce a summary of the plot. Something like that could have the 'cognition' for complex language tasks like translating a whole book.
|
|
#
?
Jan 6, 2016 15:08
|
|
|
Well that sounds like the sort of thing a phd student or postdoc might be working on right now. Off the top of my head there are a few papers that use some form of lip reading to improve audio to text transcription, but nothing comes up for the specific case of watching a sitcom and matching voices to faces. Which is quite surprising the data is there and it's labeled. I might steal this actually. Anyways, this whole thing breaks down once we start talking about cognition. We already have things that can summarize articles, but I wouldn't say a markov bot has "cognition" in any sense. It's basically flipping a coin and pointing at a word, yet they still summarize text quite well. And they can classify dialogue into categories like statement, question, agreement, etc. A method using a word embedding could certainly generate a passable summary of dialogue, but that doesn't mean it is going to do translation at all. Most of the modern nlp advancements you hear about are doing something called word embeddings. I could go into a technical explanation but the general gist is you take words, which are discrete entities, and map them to a continuous data space. This is commonly done by training a neural network that takes in words and does some basic task with them, like bucketing things into positive or negative. You then chop off the bottom layers of the network that do the work of mapping and use it for some other task. The resulting transformation does interesting things. A net trained this was could solve analogies (king-queen = boy -girl) without specifically providing this information in the data. So the hope of machine translation is that you can take the word spaces of arbitrary languages and find a mapping between them. This works relatively well for english and french, because the resulting embeddings have similarish properties. Chinese not so much, especially when you go from word to phrase/sentence embeddings, and it shows in the translation results. The ideal (to me) solution is find a way to make the mappings more similar, which is a hard problem. We don't know enough about why our networks can do this so well, and have a poor theoretical grasp behind deep learning techniques in general. There is another camp saying that if you make your model deep enough you can overcome these issues, but I'm very skeptical of that sort of philosophy.
|
|
#
?
Jan 6, 2016 17:04
|
|
|
some sort of fish posted:What bigger picture? The threat of automation is about as scary as the invention of the automobile. By and large the people running around in circles about automation ending huge swathes of labor in the near future usually are clueless about current ai and machine learning trends. The few who do tend to be extremely optimistic about the current trends of rapid advancement and also don't really have any experience working in potentially affected industries. For instance people keep saying truck drivers are going to disappear, but they do a bunch of other stuff on site during delivery that isn't worth automating. So maybe we need fewer truck 'drivers' but the profession won't change all that much other than being able to sleep while on the freeway. Personally I think current ml stuff is far too domain specific to be the existential threat people make it out to be. This is different from a simple market disruption. Middle class jobs are being hollowed out by new tech. That's a very different trend from the previous, where only highly skilled jobs were being broken down. Also, I'm sure an AI and Machine Learning Trend Expert such as yourself has something insightful to say about administration jobs being automated? They're just as vulnerable as transportation.
|
|
#
?
Jan 7, 2016 01:56
|
|
|
Freakazoid_ posted:This is different from a simple market disruption. Middle class jobs are being hollowed out by new tech. That's a very different trend from the previous, where only highly skilled jobs were being broken down. Define a middle class job, both in contemporary terms and relative to, say, 100 or 150 years ago.
|
|
#
?
Jan 7, 2016 02:17
|
|
|
Popular Thug Drink posted:they certainly will, as it's way easier to handle some navigational tasks in a swarm. each individual vehicle would have less computational work to do if all the vehicles surrounding it are broadcasting their location, direction, and speed Anyone who is using waze is already reporting this data. Google maps navigation uses waze data, but I'm not sure if it reports data though it probably does. 38 2016 models have both android auto and apple car play. Both make it even more likely those cars will report data and could increase the quality of the data (can detect turn signals, steering wheel instead of just a compass).
|
|
#
?
Jan 7, 2016 02:52
|
|
|
computer parts posted:Define a middle class job, both in contemporary terms and relative to, say, 100 or 150 years ago. 100 years ago, textile and car manufacturing underwent a shift from a few highly skilled individuals into huge groups of low-skill workers through automation. Now we're looking at jobs like insurance underwriters, claims adjusters, electrical and electronic equipment assemblers, and computer-controlled machine tool operators. These are not jobs that will be broken down to allow more workers, but eliminate workers entirely.
|
|
#
?
Jan 8, 2016 00:29
|
|
|
Freakazoid_ posted:100 years ago, textile and car manufacturing underwent a shift from a few highly skilled individuals into huge groups of low-skill workers through automation. I don't disagree with your premise about the future of white collar jobs, but this isn't true. 100 years ago is 1916, an era when textile manufacturing was mostly child labor, and car manufacturing was dominated by Ford. Even if you intended to be more loose with the time frame, if you go much farther back in time textile manufacturing is literally a cottage industry and car manufacturing didn't exist. Cars only ever became economically feasible because of automation, rudimentary as it was initially. Welding robots aren't the big labor-market-changing innovation of making cars, the assembly line is.
|
|
#
?
Jan 8, 2016 00:40
|
|
|
|
| # ? Jun 1, 2024 23:49 |
|
|
Tiny Brontosaurus posted:I don't disagree with your premise about the future of white collar jobs, but this isn't true. 100 years ago is 1916, an era when textile manufacturing was mostly child labor, and car manufacturing was dominated by Ford. Even if you intended to be more loose with the time frame, if you go much farther back in time textile manufacturing is literally a cottage industry and car manufacturing didn't exist. Cars only ever became economically feasible because of automation, rudimentary as it was initially. Welding robots aren't the big labor-market-changing innovation of making cars, the assembly line is. Textile manufacturing left its cottage industry roots during the 19th century, so I don't think I'm being too loose with that one. While child labor is deplorable, child labor was estimated to be around 25% in southern US textile mills in 1900. I am being a little loose with car manufacturing, but the industry started with high-skilled workers and I'm pretty sure they were eclipsed by low-skilled workers by 1916. I believe assembly lines count as automation since Ford utilized drive belts to achieve it. Either way, I'm not sure what computer parts was getting at.
|
|
#
?
Jan 8, 2016 01:28
|
|