5 ms·

I am terrified at the concept of "self-driving" or anything marketed as "self-driving" even if it is really advanced driver assist. When I talk to my engineer f

I am terrified at the concept of "self-driving" or anything marketed as "self-driving" even if it is really advanced driver assist. When I talk to my engineer friends they are also deeply skeptical. Maybe its our age (mid 40's) or that after dealing with software development for 20+ years you tend to be jaded. Maybe it's that I know that ML based systems have limits based on the training data and it's not real deep AI. Maybe its that all the training is done in CA and someone will try to use this software where I live in New England and do it when it is snowing at night on a drive to Vermont and the system will fall apart. I don't think we understand the human factors involved and the the hubris that most people will ignore their role in oversight as the software gets more advanced. If the software is 90% effective for most driving people will tune out. Holy crap that is scary.

suifbwish 5y ago
I know that there are probably versions of AI that are hidden from the public both private and military but something tells me if we were close to the advent of anything self operational it would be evident. There would be transitional technologies emerging on the markets. More than likely there are pieces of very complex intelligence resembling mechanisms that are held together by teams of people. The idea of even the most modern intelligence system navigating through an environment as elaborate and dynamic as New York City or downtown Chicago should give anyone pause. I can see them being more useful for long distance journeys such as hauling a trailer across rural desert areas where all the vehicle needs to do is stay in its own lane, maintain a set speed and brake if something obvious gets in front of it.
- moffkalast 5y ago
  To be frank, the lane keeping ADAS has been out for years and most car manufacturers offer such a thing on their high end models. Usually radar supported and pretty reliable from what I hear. Perhaps it's not the best idea to use these more advanced Level 4 systems in the densest city yet, but most of driving does usually happen on highways and regional roads so it's not a major loss. I'm sure someone will eventually figure out how to detect "eyyy im walkin eer" and honk back automatically, and then it should be usable in NY as well. > versions of AI that are hidden from the public both private and military I do think you overestimate how much the military has its shit together. These systems need gigantic amounts of driving data to train and nobody else has that aside from car manufacturers with their fleets. And out of those Tesla is the one that's been doing it the longest.
  - YeGoblynQueenne 5y ago
    No, I think it's reasonable to wonder whether some powerful enough agents have some "super secret tech" and aren't telling. It wouldn't be the first time. For instance, the US didn't exactly advertise having the atomic bomb before they first used it (although it was a very short time after and they did specifically work on the atomic bomb to use it as soon as possible). That said, personally, I think it's unlikely. First, the vast majority of AI research is happening out in the open. It's often funded by militaries (e.g. DARPA) but it's not owned by them directly, rather the owners are universities and similar. Second, private corps promote their AI advances very aggressively and this marketing is a big part of the advantage they try to establish over their competitors, so it doesn't seem obvious why they would keep a new advance "under wraps". For example, what would Google do with a secret machine vision system capable of 20% better performance than that of its rivals? I can't tell. Of course we can assume that Google or someome similar have a much more advanced system than anyone else, like a real AGI. We can imagine such a system sitting at the heart of Google, dirctly advice Larry and Sergey how to run the company. But then, if we can imagine that, we can imagine anything. And I do mean anything: https://gatherer.wizards.com/Pages/Card/Details.aspx?multiverseid=73947 https://gatherer.wizards.com/Pages/Card/Details.aspx?multive...
    - moffkalast 5y ago
      Well I do think it's very possible if not certain that they have unusually advanced systems that have been kept secret, but I would expect them to be in the areas they actually have data for. As in security cameras, finances, cell tower data, etc. Facial and gait recognition, location tracking, etc. far beyond anything commercially available. But I'd doubt they have an AGI behind closed doors :P
  - suifbwish 5y ago
    Well actually if you look at the rest of what I said, it’s that if there were advanced systems hidden behind the curtain, transitional stages of said systems would have made their way into the public realm. You can’t just jump to something as massive as full blown AI and not have byproduct tech. Consider the space race for instance, and all the resulting byproduct techs it brought. So many inventions that were created along the way that made their way into the public realm. I find it difficult to believe that any other breakthrough of similar or greater magnitude would not do the same.
ajross 5y ago
This is broadly the reasoning behind the fear in the linked article. But it's... really it's just fear. All the stuff you posit is something that can be tested and analyzed. Thousands of people have been driving this stack around since March with no accidents. At some point data will win, right? Have you worked out in your head when that will be? > I don't think we understand the human factors involved Do we? Because human factors responsible for literally every traffic accident in history, and those don't seem as scary to you. Basically: it's fear. It's just fear. And it's reaching a peak right now (and being deliberately pushed, frankly) just as Tesla is reaching a public rollout. And why? Because there's another fear at work here too. If this rolls out and doesn't kill people. If it turns out that it actually is safe, then a lot of people across a lot of industries stand to lose a lot of money. Stories like this are, fundamentally, just marketing press hits. The public FSD beta rollout in some sense marks the last chance to kill Tesla's autonomy product before... it wins.
- moffkalast 5y ago
  Yeah, besides Tesla has been collecting data for like a decade now, I'm sure they have enough of it to make a robust enough system for the US at large, especially with their Dojo training chip. At the end of the day it just needs to be better than the average human, and that's frankly not the highest bar unfortunately.
  - darepublic 5y ago
    Afaik the full self driving is not better than the average human, not even close. The reason for the lack of accidents are the human drivers intervening when the system suddenly behaves irrationally, which is often. But if the FSD was left entirely to its own devices there would be a huge uptick in everything ranging from fender benders and fatal collisions.
    - ajross 5y ago
      That's a strawman though. The argument upthread and in the linked article is that the supervised FSD Beta is presumptively unsafe. Arguing that full autonomy isn't possible yet (duh, I mean, it says "beta" right there in the name) isn't an argument to disallow testing.
WouterZ 5y ago
I see it like this... Crashes will happen if we block self driving cars. Crashes costs lives... People will die. But if they die and the algorhythm learns from it and prevents the crash the next time... In a couple of years of further crashes maybe we can iron out most flaws and we save lives.
supperburg 5y ago
You think because the computer is in California that it only knows how to drive in California? Maybe you should learn the first thing about something before spreading FUD about it?
- harshaw 5y ago
  lol no, Its that we have different road conditions and unique circumstances in different parts of the country.
  - supperburg 5y ago
    It wouldn’t matter if the training computer were on the moon. Training data is collected by Tesla owners all across the country including the Eastern United States. Regardless of how well you think it can drive, it can drive equally well in New England as it can in California. And there may be an unequal number of Tesla’s on each coast but that’s irrelevant.

suifbwish 5y ago

I know that there are probably versions of AI that are hidden from the public both private and military but something tells me if we were close to the advent of anything self operational it would be evident. There would be transitional technologies emerging on the markets. More than likely there are pieces of very complex intelligence resembling mechanisms that are held together by teams of people. The idea of even the most modern intelligence system navigating through an environment as elaborate and dynamic as New York City or downtown Chicago should give anyone pause. I can see them being more useful for long distance journeys such as hauling a trailer across rural desert areas where all the vehicle needs to do is stay in its own lane, maintain a set speed and brake if something obvious gets in front of it.

moffkalast 5y ago
To be frank, the lane keeping ADAS has been out for years and most car manufacturers offer such a thing on their high end models. Usually radar supported and pretty reliable from what I hear. Perhaps it's not the best idea to use these more advanced Level 4 systems in the densest city yet, but most of driving does usually happen on highways and regional roads so it's not a major loss. I'm sure someone will eventually figure out how to detect "eyyy im walkin eer" and honk back automatically, and then it should be usable in NY as well. > versions of AI that are hidden from the public both private and military I do think you overestimate how much the military has its shit together. These systems need gigantic amounts of driving data to train and nobody else has that aside from car manufacturers with their fleets. And out of those Tesla is the one that's been doing it the longest.
- YeGoblynQueenne 5y ago
  No, I think it's reasonable to wonder whether some powerful enough agents have some "super secret tech" and aren't telling. It wouldn't be the first time. For instance, the US didn't exactly advertise having the atomic bomb before they first used it (although it was a very short time after and they did specifically work on the atomic bomb to use it as soon as possible). That said, personally, I think it's unlikely. First, the vast majority of AI research is happening out in the open. It's often funded by militaries (e.g. DARPA) but it's not owned by them directly, rather the owners are universities and similar. Second, private corps promote their AI advances very aggressively and this marketing is a big part of the advantage they try to establish over their competitors, so it doesn't seem obvious why they would keep a new advance "under wraps". For example, what would Google do with a secret machine vision system capable of 20% better performance than that of its rivals? I can't tell. Of course we can assume that Google or someome similar have a much more advanced system than anyone else, like a real AGI. We can imagine such a system sitting at the heart of Google, dirctly advice Larry and Sergey how to run the company. But then, if we can imagine that, we can imagine anything. And I do mean anything: https://gatherer.wizards.com/Pages/Card/Details.aspx?multiverseid=73947 https://gatherer.wizards.com/Pages/Card/Details.aspx?multive...
  - moffkalast 5y ago
    Well I do think it's very possible if not certain that they have unusually advanced systems that have been kept secret, but I would expect them to be in the areas they actually have data for. As in security cameras, finances, cell tower data, etc. Facial and gait recognition, location tracking, etc. far beyond anything commercially available. But I'd doubt they have an AGI behind closed doors :P
- suifbwish 5y ago
  Well actually if you look at the rest of what I said, it’s that if there were advanced systems hidden behind the curtain, transitional stages of said systems would have made their way into the public realm. You can’t just jump to something as massive as full blown AI and not have byproduct tech. Consider the space race for instance, and all the resulting byproduct techs it brought. So many inventions that were created along the way that made their way into the public realm. I find it difficult to believe that any other breakthrough of similar or greater magnitude would not do the same.

ajross 5y ago

This is broadly the reasoning behind the fear in the linked article. But it's... really it's just fear. All the stuff you posit is something that can be tested and analyzed. Thousands of people have been driving this stack around since March with no accidents. At some point data will win, right? Have you worked out in your head when that will be? > I don't think we understand the human factors involved Do we? Because human factors responsible for literally every traffic accident in history, and those don't seem as scary to you. Basically: it's fear. It's just fear. And it's reaching a peak right now (and being deliberately pushed, frankly) just as Tesla is reaching a public rollout. And why? Because there's another fear at work here too. If this rolls out and doesn't kill people. If it turns out that it actually is safe, then a lot of people across a lot of industries stand to lose a lot of money. Stories like this are, fundamentally, just marketing press hits. The public FSD beta rollout in some sense marks the last chance to kill Tesla's autonomy product before... it wins.

moffkalast 5y ago
Yeah, besides Tesla has been collecting data for like a decade now, I'm sure they have enough of it to make a robust enough system for the US at large, especially with their Dojo training chip. At the end of the day it just needs to be better than the average human, and that's frankly not the highest bar unfortunately.
- darepublic 5y ago
  Afaik the full self driving is not better than the average human, not even close. The reason for the lack of accidents are the human drivers intervening when the system suddenly behaves irrationally, which is often. But if the FSD was left entirely to its own devices there would be a huge uptick in everything ranging from fender benders and fatal collisions.
  - ajross 5y ago
    That's a strawman though. The argument upthread and in the linked article is that the supervised FSD Beta is presumptively unsafe. Arguing that full autonomy isn't possible yet (duh, I mean, it says "beta" right there in the name) isn't an argument to disallow testing.

WouterZ 5y ago

I see it like this... Crashes will happen if we block self driving cars. Crashes costs lives... People will die. But if they die and the algorhythm learns from it and prevents the crash the next time... In a couple of years of further crashes maybe we can iron out most flaws and we save lives.

supperburg 5y ago

You think because the computer is in California that it only knows how to drive in California? Maybe you should learn the first thing about something before spreading FUD about it?

harshaw 5y ago
lol no, Its that we have different road conditions and unique circumstances in different parts of the country.
- supperburg 5y ago
  It wouldn’t matter if the training computer were on the moon. Training data is collected by Tesla owners all across the country including the Eastern United States. Regardless of how well you think it can drive, it can drive equally well in New England as it can in California. And there may be an unequal number of Tesla’s on each coast but that’s irrelevant.