Please do! Because I totally didn't realize it was that Henry Petroski, I took the article at face value. I thought he wasn't a very good author, but it seems now almost certain that he had an inadequate editor. That doesn't diminish the terribleness of the article much, though - it mostly shifts the blame onto WSJ. What prompted me to write this in this specific format was to contrast the good and bad parts of the article. It's the Hubski version of a running commentary. So if anyone wants to roll with that idea, be my guest. I agree with you that the problems surrounding this topic have many aspects and multiple ways of advancing. The way I see it is that there is One Big Question that AV's attempt to answer: how do we convert human driving intelligence into machine driving intelligence? Every example or situation (or even trolley problem) is an extension of that question, and all the manifestations of automated technology are potential answers to it. My main point of criticism is that this article never makes that clear, because if that were clear, it means that you need to specify what part of the question and what sort of answer you're talking about. A Google car (answer) in a double-parked situation (question) is in an entirely different league than driving on the highway (question) in an Audi with ADAS (advanced driver assistance systems, an answer). This is why I hammered on the assumptions being made and the situations being discussed in the article, and how the article does a terrible job of it. I agree, but a more fair assesment would be that the attention on this topic is disproportionally small. The 1600-page tome has 38 pages on implications for the future, and I know a few articles that are thinking about this so I'm not a lone wolf here. It also needs to be noted that this research question is built upon the assumption that automated technology will move quite far beyond fancy cruise controls. (Just putting that on record here.) Maybe it's my Dutch urban planning background, but I think that people have forgotten the tremendous impact on cities and transportation of postwar mass car adoption. In the case of American cities, most of which are built around cars, I still think this can be an interesting and major shift. Interesting way of putting it. How do you see this playing out?And know that I will be grilling you on every one. ;-)
And your blind spot on this, I think, is that you see Petroski asking questions as if they're unanswered... rather than seeing it as Petroski asking questions that will be answered multiple ways.
That there's no easy thing to point to for parallel argument demonstrates that this is something nobody's thinking about much. And that's the gist of the article - "nobody is thinking about this much".
instead of recognizing that the probable eventuality is a massive publicly-funded FAA-grade "ground traffic control system"
Why are we trying to convert human driving intelligence into machine form? Why not allow these machines and algorithms to learn the best courses of actions to take to reduce accidents without trying to impose things like "you should hit the car instead of the bus of school-children"? You've made a system that learns how to best manage situations, you don't turn around and undercut the decisions this system makes in order to appease people proposing absurd hypothetical scenarios.how do we convert human driving intelligence into machine driving intelligence?
Two reasons - machine intelligence is still a long way from being capable of human-level reasoning. Most of what we have now (that works well) is very simple-minded, brute-force approaches to reasoning (i.e. expert systems). Second reason - we don't NEED human-level reasoning to make good progress in self-driving. As an example, a huge first step in this process happened long ago, and has hardly any innate intelligence at all - the automatic transmission.Why not allow these machines and algorithms to learn the best courses of actions
Most self driving systems today are using neural networks, I don't think expert systems would be able to manage the complex issues that pop up with driving. As a result, I don't think we could really understand all the nuances that these systems are capable of knowing once they have been trained for so long to drive. It's easy to create a network that learns to do a task relative to learning how the completed system is thinking. These aren't brute force approaches anymore, there is a system that is learning as time passes, and getting better at what it does as time continues forward. As such, these networks have one job, and do that one job very well. In this case that job is to prevent crashes. Fiddle with that network to impose artificial limitations and you impose on a system optimized to do something, and more crashes will result in the long run. Although I'm sure there are cases where things go wrong with the program, or things need tweaked, these aren't the same as directly interfering with the car when it decides to take a course of action that could lead to hitting a schoolbus vs a normal car. It may well be that hitting the schoolbus causes less total harm for some reason, and we be sure that we understand the reasoning of the machine before we decide to mess with it.
No offense intended, but colour me skeptical - can you support this? I believe there's a lot of research into neural networks for pedstrian identification etc, but I've never seen any indication that (for example) Google Chauffeur uses neural networks at all. Again, they appear to be throwing brute-force at the problem. [edit] Details about self-driving software are hard to find; however, I did find this little hint (from this article http://www.scientificamerican.com/article/autonomous-driverless-car-brain/ )Most self driving systems today are using neural networks
Yet as smart as today's cars may seem, they are cognitive toddlers. In a car brain, software, processors and an operating system need to run algorithms that determine what the car should do, and these decisions must be made quickly.
Nvidia thinks so: http://www.nvidia.com/object/drive-px.html Within a journalistic article like this, neural networks more than fit this definition. They are, after all, just a bunch of big arrays with a bunch of weights and activation functions.In a car brain, software, processors and an operating system need to run algorithms that determine what the car should do
Any system that assists or replaces driving is a system that converts a bit of human intelligence into machine driving intelligence. I mean, we can have a discussion about semantics and heuristics but that is what I meant with that statement. The interesting challenge to me is how we do this. Is that what you're alluding to? There is an interesting debate going on about whether automated systems should assist or replace human drivers. I recently read an interesting article that proposed a human-machine interaction with the 'horseman analogy', where your car has rules and can take some basic actions on its own (like a horse), and you as the horseman have the final say and decide on the general course.
I assumed you were referring to the idea of machines needing to know "human morality" in order to replace human actions, and was saying that machines shouldn't have our morality imposed on them, they should be allowed to come to decisions naturally based on the learning algorithms they are built with. I ultimately think human drivers should be entirely replaced by machine drivers. The horse-driver analogy is interesting, but ultimately it defeats the purpose of the self driving car. I think it makes a good stopgap, though.
I still maintain that you're experiencing your first instance of "professional disappointment" - when you read a popular discourse on a subject that you know enough to be an expert about and the focus of the author isn't the one you'd choose. Display technology articles in Scientific American used to give me headaches, and of the two people I know who have written in Sciam, both of them hated the fuck out of the end result (and generally hate what everyone else has to say in their field). That's the problem with populism - your publication automatically dumbs your words down to the lowest common denominator and whatever you, the expert, may think it requires too much basis of knowledge in order to refute. This is an article written for people who literally think As you yourself pointed out - there are good things in this article and there are bad. With nearly everything else it's all bad. You're disappointed that the article isn't better because it was good enough to get your hopes up. Get used to that - it's going to happen more and more often. I also think the "big question" is different from what you think - I don't think AVs are a question of converting human driving intelligence into machine driving intelligence, I think it's a question of intermutability between individual transit and mass transit. A car is an autonomous vehicle piloted by a lone human. It's part of a larger system but the links of that system are ad-hoc. A self-driving car is on a spectrum between autonomous and fully-dependent - on freeway or commuter flows it's a fully-integrated part of a mass transit system. On local flows it's an extension of a larger transit system. That this spectrum cannot be bridged by humans is the core of the problem - we have trains and we have cars and in between we have buses but from a transport perspective, those three things are discrete modes of transportation. Throw AVs into the mix and the distinction blends, particularly if it works out economically to get away from the 1car = 1family model. The "self-driving" part of it all is big news to the geeks, but it's a tiny aspect of the overall changes engendered: Apparently mk and I are the only ones who read that article, which is a shame, because some of the fallout of automated truck fleets includes: - 75% reduction in freight charges - 150-200% decrease in transit times - radical improvements to public safety - Loss of at least 1% of all jobs in the United States I mean, that's some straight-up disruption and nobody even gives a fuck about the autonomous technology because these are giant trucks and that shit is mostly handled. It's just a matter of time. And nobody read that. It's hidden on tech crunch. Nobody here read it. NOBODY CARES. So be thrilled that at least Petroski took a swing at it, and recognize that it's going to take a steady drumbeat of people like Petroski (hint hint) to get people to care. So grab some sticks. ________________________________ As far as "ground traffic control system" I think it's gonna be like this: Google, as you pointed out, is mapping shit down to the inch. That's a massive first-starter advantage that nobody else will have the ability to catch up with. Everyone else is going to have to combine Google's slavish mapping tendencies with shortcuts that allow them to compete. It's going to be chaos - you're going to see elected officials, who will read things like the Petroski article with slack jaws, gobsmacked by the implications that they've never considered. These will be people who wouldn't understand LIDAR with Bill Nye standing next to them with a PowerPoint and they're going to have to pass judgement on what constitutes "autonomy-ready" vehicles and what doesn't. And, like every jurisdiction everywhere, they're going to kick the can upstairs, where eventually it'll end up in the hands of a large bureaucracy like NHTSA, who will have to come up with standards for autonomous vehicles. And as every manufacturer will have been trying different things, and as they won't be able to come up with mutual standards without assistance and subsidy, there will be a big industry initiative which will be called something evocative like "DRIVE2100" (evocative is easier to raise taxes for) and it'll have recommendations for what needs to be built into lights, signs, addresses, crosswalks, intersections, roundabouts, onramps, offramps, parking lots, the whole nine yards, and it will be seen as a giant works project by every industrialized nation, and the ones that are ahead of things are going to be the ones that are going to benefit the most because these are going to be cost-plus contracts and it's going to be bigger than the Americans with Disabilities Act and the guys who get to play in that game are going to be the guys who made a credible effort up to this point and the guys that are going to be dictated to are going to be the guys that didn't and if you ever wondered why Apple and Google are playing games with self-driving cars it's because we're on the cusp of the single biggest worldwide infrastructure expenditure in the history of civilization and it's gonna make The New Deal look like Cash for Clunkers. And it starts with articles like this. ("This year alone more people will be killed in traffic accidents involving trucks than in all domestic airline crashes in the last 45 years combined. At the same time, more truck drivers were killed on the job, 835, than workers in any other occupation in the U.S.")
It's gonna take a while, I'm afraid. While one of the insights I gained last year from doing my thesis was exactly what you mention (AV's creating all in-between forms of transport), that still likely manifests itself after we know what's the best way to make vehicles drive themselves. There is a path dependency there that might have major implications further down the road. The first big success is likely going to dominate the field because it is so difficult to get up to that level of safety standards. Not saying that it isn't also a very interesting and Big Question - just that I think the machine intelligence sets the precedent. I DO! I just totally missed that article. My second choice of thesis topics is to investigate the benefits of quay-to-customer or quay-to-warehouse full automation for the harbor of Rotterdam, which already has a fully automated port.Get used to that - it's going to happen more and more often.
NOBODY CARES.
Well... I dunno. There's going to be patchwork acceptance, and there's going to be conflict, and then there's going to be a working group and standards and public input periods and we'll still end up with an "open" standard like docx except it'll cost $8k to implement per vehicle. I think our fundamental disagreement is that I don't think there is a "best" way for AVs to drive themselves. I think there are probably several approaches that can be made to work, depending on environment, and I think that the bigger you are now, the more likely you are to be able to force your approach. Even then, it's a gamble. Sony lost Betamax but won CD and BluRay. And this battle is going to dwarf those. There is a path dependency there that might have major implications further down the road. The first big success is likely going to dominate the field because it is so difficult to get up to that level of safety standards. Not saying that it isn't also a very interesting and Big Question - just that I think the machine intelligence sets the precedent.