Well said. I just read a report of how Tesla's purported autosteering reducing crashes by 40% was a case of bad statistics (albeit, the research was not done by Tesla, but by NHTSA. Though Tesla did trumpet the study a lot).
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple Reports Self-Driving Car Disengagements to DMV, Earns Worst Rank
- Thread starter MacRumors
- Start date
- Sort by reaction score
Well said. I just read a report of how Tesla's purported autosteering reducing crashes by 40% was a case of bad statistics (albeit, the research was not done by Tesla, but by NHTSA. Though Tesla did trumpet the study a lot).
Sigh... so much misinformation in this thread. 
As for BlackBerry, are they at https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/permit?
Their 2017 report accessible at https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/disengagement_report_2017 had some more background info.
[doublepost=1550132796][/doublepost]
https://waymo.com/journey/ says they been working on self-driving technology since 2009, so 10 years or so.
Arizona communities wouldn't be counted in the report. This is only for autonomous vehicle testing on California public roads.
As for BlackBerry, are they at https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/permit?
See https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/disengagement_report_2018
No. See above.
They do but if it's not a CA public road, it isn't counted in their stats. See their report at https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/disengagement_report_2018.
Their 2017 report accessible at https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/disengagement_report_2017 had some more background info.
[doublepost=1550132796][/doublepost]
No. Google was founded in Sept 1998, so that puts the company shy of 21 years.
https://waymo.com/journey/ says they been working on self-driving technology since 2009, so 10 years or so.
The bolded part is not correct.
See https://www.dmv.ca.gov/portal/wcm/c...essAV_Adopted_Regulatory_Text.pdf?MOD=AJPERES, the 1st PDF link from https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/testing.
Well in that case they're designing their own chips aren't they, they're not making them. They don't exist anywhere yet other than their own test labs - so back the original point, it's not unusual to see Nvidia on the chart considering every single Tesla out there today and for the foreseeable future contains the hardware.
I know - I was being sarcastic. The poster I replied to was suggesting Apple weren't very good because they are new to self-driving technology. The same as everyone else.
I am not surprised Apple is in dead last. This is not knocking apple but they just do not have the data everyone else has and they are starting way behind and do not have a great way to get updates.
For example comparing Apple Maps vs Google and Waze. Apple maps is still way behind and does not get updated as fast. It has fewer users to get real time data compared to google and waze and Google has a lot more years of experence in keeping up.
The Voice AI siri is very far behind Amazon and Google and lets be honest they will never catch up due to the shear lack of data and they can not pull it in as easily. Google was using Google Voice voice mails for years to help train AI in understanding human speech and in that case only voice mail volunteered for training where used. Amazon has alexia which is in everything. This is no different.
For example comparing Apple Maps vs Google and Waze. Apple maps is still way behind and does not get updated as fast. It has fewer users to get real time data compared to google and waze and Google has a lot more years of experence in keeping up.
The Voice AI siri is very far behind Amazon and Google and lets be honest they will never catch up due to the shear lack of data and they can not pull it in as easily. Google was using Google Voice voice mails for years to help train AI in understanding human speech and in that case only voice mail volunteered for training where used. Amazon has alexia which is in everything. This is no different.
At some level, yes it might be about information, but therein lies a basic problem - or at least naming issue - with AI (artificial intelligence). ML (machine learning) is probably a less misleading term to use. In other words, if you have this huge, massive set of data a computer can crunch through (something computers are good at), using human-derived intelligence, properly put into algorithms... you might get a pretty good analysis, and hence result.
Humans, on the other hand, drive around every day with far from a complete data set, and figure things out on the fly. If they have even slightly more than abysmal driver training, and aren't intoxicated or willfully-distracted, they do a pretty darn good job of it, too. I'd argue, this is because there is actual thinking going on.
AI isn't thinking, just following a program. It needs that data set and human-designed way of interpreting the data to even appear to the uninformed as though it is thinking (which it isn't doing). This presents two issues relevant to the discussion.
First, I think Apple's problem isn't on the data-set side of things, especially for many failures of Siri. In many cases, it has the data, it just doesn't know what to do with it. It seems (to me) that Siri suffers a similar problem to Apple's search technology (Podcasts, App Store, Spotlight, etc.) where it can't even do the basic things Google can do with ALL the data right in front of it. They can't even match a search engine like Alta Vista from the mid-90s!
More data isn't going to help if you don't even have the fundamentals of how to process and interpret it in place.
Second, in regards to autonomous vehicles, if they need a complete data-set, that's a problem. And, with some of these systems operating where every detail is mapped out, is kind of cheating (and problematic if they depend on that!). Things change all the time. Crazy events happen. Even if you have the core environment 100% mapped (which is impossible), there are animals, people, weather, etc. which never will be included.
If this stuff is ever to really work, it like us, needs to be real-time taking in and interpreting the data, and then decision making without a crutch of having to have perfect data (and previous results) to fall back on. Much of the argument I've seen from people who actually understand AI (so realize it isn't thinking) is that what this tech really relies on is building a more and more prefect data-set, such that the mess-ups keep dropping. Eventually, they think the mess-ups will drop below human-driving characteristics. (Currently at about 1 fatality per 100M miles driven, and that's with almost insane lack of training, and allowing the most stupid irresponsibility of human drivers... both which could easily be addressed. If they were, where would the stats quickly go? 500M 1T etc.?)
That might be, BUT we could EASILY improve human driving if safety were the actual motive. And, I'm not sure they ever will have that good-enough data-set and/or enough human-designed algorithms to correctly process and react to it.
(As I've heard Adam Curry say on the No Agenda show... that he'll start believing in this AI stuff when they can get email spam filters working. No doubt!)
Last edited:
Apple has one of the highest starting salaries for ML engineers, even compared to Google. Though it's obvious why people still pick other companies over Apple.
I wonder if the tech Apple is developing is entirely different from the "self-driving" tech everyone else is working on.
For self-driving tech to really work to reduce congestion and lower fuel consumption, the key is interconnected cars. Long lines of cars working together as a "train" and all accelerating, braking, and interacting together as one unit. Humans cannot do this. Machines can. And I wonder if the frequent "disengagements" of Apple's tech might suggest that cars running Apple tech are connecting to each other to form "trains" and disengaging when an individual vehicle leaves the formation.
For self-driving tech to really work to reduce congestion and lower fuel consumption, the key is interconnected cars. Long lines of cars working together as a "train" and all accelerating, braking, and interacting together as one unit. Humans cannot do this. Machines can. And I wonder if the frequent "disengagements" of Apple's tech might suggest that cars running Apple tech are connecting to each other to form "trains" and disengaging when an individual vehicle leaves the formation.
Amazon is on a roll. Now getting into the EV SUV space with major investment.
https://media.rivian.com/rivian-announces-700m-investment-round-led-by-amazon/
https://media.rivian.com/rivian-announces-700m-investment-round-led-by-amazon/
These stats are for tests on public roads, and no one has noticed multiple Apple cars together.
https://9to5mac.com/2019/02/20/apple-self-driving-safety-paper/ has a pointer to a brief whitepaper.
SteveW928 [Why... isn't this AI? Humans don't all have to communicate with each other in real-time in order to drive.]
Dude, we ALL have to communicate while driving in traffic. Subtle little looks at each other at intersections, in parking lots allowing one guy to go ahead of you, etc..... Many different subtle ways. Heck, I want to see AI give someone "The Bird"! ha ha
You're cracked if you don't think humans connect with each other while driving the crowded streets and highways of America. And AI is FAR away from having the mind of a human being. Probably like 300 years away!
Dude, we ALL have to communicate while driving in traffic. Subtle little looks at each other at intersections, in parking lots allowing one guy to go ahead of you, etc..... Many different subtle ways. Heck, I want to see AI give someone "The Bird"! ha ha
You're cracked if you don't think humans connect with each other while driving the crowded streets and highways of America. And AI is FAR away from having the mind of a human being. Probably like 300 years away!
Oh, I pretty much agree with you in that regard, I guess. I guess I meant that we're not all literally communicating, as in all talking to each other, coordinating our position, intentions, etc. We more 'read' that stuff, at least if we're actually paying attention (which too many aren't these days).
And, that highlights a huge problem with a mixture of AI cars and human-driven cars on the road at the same time. *Maybe* a few of these kind of things can be caught in terms of behavior and added to the ML algorithms, but most of it is going to be lost on the AI car. The problem is, that the human driver isn't going to realize that AI car is being driven by a virtual 'idiot' (though, as I said, there are exponentially more and more human idiots on the road these days, too).
As for 300 years, I'd change that to never. It isn't a problem of technological scale, but one of kind. As good as it will ever get is only going to be a feeble attempt to mimic. That's why I like ML (machine learning) better than AI, as I think people are being mislead about the 'A' aspect. They tend to think the computer is actually being taught to think, not just mimic a fake kind of intelligence.