Why AP 2.0 Won't Be Here Soon, and It Won't Be What You Think It Is

[Topher]

I don't see how you can support the theory that humans, 100% of the time, without a single fail or falter, in the entirety of mankind, will always make the same decision given the "exact same inputs". If that doens't happen just once, your theory fails.

I am not supporting that theory at all. I am challenging YOUR theory that humans can make different decisions based on exactly the same inputs. You are the one making a positive claim, so the burden of proof lies with you.

But sure, go ahead and provide a counter example. You have already admitted that no such example is possible.

...since it's impossible to give humans the "exact same inputs".

Thank you kindly.

 

[Dynastar]

I don't see how you can support the theory that humans, 100% of the time, without a single fail or falter, in the entirety of mankind, will always make the same decision given the "exact same inputs". If that doens't happen just once, your theory fails.

'on a hot streak' versus 'not on a hot streak' is exactly what I would consider NOT the same inputs. Having won the last 10 rolls, is decidedly not the same input as having lost the last 10 rolls. Why would you think it was? Gambling programs take that into consideration why wouldn't humans?

Thank you kindly.

Really guys, we're going to argue determinism vs free will in an AP 2 thread?

 

[Topher]

I would hope gambling programs (and other predictive models based on independent event probabilities) wouldn't take that into consideration, because it's the gambler's fallacy. Gambler's fallacy - Wikipedia, the free encyclopedia

Nope. Winning 10 rolls in a row doesn't change the odds of winning the next roll, but it does radically change the correct way to bet since the available funds have drastically changed.

Thank you kindly.

 

[Bet TSLA]

I would hope gambling programs (and other predictive models based on independent event probabilities) wouldn't take that into consideration, because it's the gambler's fallacy. Gambler's fallacy - Wikipedia, the free encyclopedia

Of course models of abstractions wouldn't take streaks into consideration, but models of the real world have to consider that the events might not be independent. Just like people. How many coin flips in a row yielding heads do you see before you are pretty sure that the coin or the flip are not fair?

Auto-pilot is living in the real world. Things are messy.

 

[HankLloydRight]

I would hope gambling programs (and other predictive models based on independent event probabilities) wouldn't take that into consideration, because it's the gambler's fallacy. Gambler's fallacy - Wikipedia, the free encyclopedia

Exactly. There is no such thing as a 'hot streak' so it can't be an input. It's an entirely human perception.

 

[HankLloydRight]

Nope. Winning 10 rolls in a row doesn't change the odds of winning the next roll, but it does radically change the correct way to bet since the available funds have drastically changed.

What if you're not playing for money? The odds are the odds, regardless of previous outcomes. The level of monetary wagering shouldn't be considered for any subsequent gamble. But humans, being fallible, do. A lot. So given the same "inputs", people make different decisions. It's a really pretty basic notion, I really don't have time to dispute it further.

 

[proven]

I agree and disagree with the OP. Agree that L3 or L4 autonomy is quite a few years away. There are so many scenarios that are tricky to navigate that it would take some seriously good sensors/cameras and software to be successful. However, I disagree that the AP 2.0 hardware is that far off. I think you'll see the hardware sometime next year, but the software will be almost identical to the current AP at first and then slowly add features to improve.

 

[HankLloydRight]

I am not supporting that theory at all. I am challenging YOUR theory that humans can make different decisions based on exactly the same inputs. You are the one making a positive claim, so the burden of proof lies with you.

But sure, go ahead and provide a counter example. You have already admitted that no such example is possible.

I did prove it, IF you limit the inputs to a few discrete inputs, like gambling. Dealer shows an 8, you hold a 5 and a 6. Sometimes people will hit, sometimes those exact same people, with the exact same cards, will double down. And even sometimes, they'll hold.

If this weren't true, places like Las Vegas wouldn't exist. If people always made the same decision based on the same inputs, they would NEVER EVEN gamble.

 

[Topher]

Really guys, we're going to argue determinism vs free will in an AP 2 thread?

Are you kidding? HankLloydRight has TRIVIAL PROOF of free will. Do you know how much free advertising Tesla will get when it becomes known that the proof to one of humanities most challenging philosophical questions was finally cleared up after all this time on a Tesla discussion forum?

Thank you kindly.https://teslamotorsclub.com/tmc/members/hanklloydright.20679/

 

[HankLloydRight]

Really, proving humans are imperfect and fallible is one of "humanities most challenging philosophical questions"?

Wow.

 

[mhan00]

To get back to the actual topic, looks like the first delivered P100D does not have an additional camera and probably no additional autopilot hardware at all. So it doesn't look like Tesla was waiting for the P100D to release new autopilot hardware like some thought.

 

[HankLloydRight]

I think the fact that there is going to be an AP 2.0 using Radar instead of the camera is proof that the humans that developed AP 1.0 are clearly imperfect and fallible, and given the same set of inputs (the available sensors on the car) have now made a completely different decision.

By the way, I never said anything about "free will."

 

[ecarfan]

Still completely disagree. There is almost infinite human judgement and nuances that go into "driving." that computers will never be able to emulate, for instance my example above of a road hazard in one or two lanes over that the sensors can't detect where a human could easily detect and predict what other human drivers will do to avoid it. Even if the sensors could detect it, there are 1000 different possible outcomes and the computer has to use logic to pick one, and most likely that would not be the maneuver that a human would choose.

And it is entirely possible that the computer will consistently pick better outcomes than most humans.

Autonomous driving is a challenge that Google has already solved with their test cars in the SF Bay Area for years now. I see the Google cars on my local freeways and surface streets. Their safety record is far better than human drivers.

Tesla's systems have not yet reached what Google has already achieved but I bet Tesla will catch up very soon with fleet learning and a few more radars and cameras.

 

[bob_p]

Another take on this...

First, AP 1.0 is "beta" hardware - it will never have enough capability to achieve the goal of fully autonomous driving.

AP 2.0 is likely also going to be "beta" hardware - it will get closer to self driving, but likely will still fall short.

There are 3 major components to AP - sensors, processing and software.

AP 2.0 will have more and improved sensors. AP 1.0 is based on technology that is relatively old now, and it seems unlikely Tesla would put the Model 3 into production with the older AP 1.0 hardware. AP 2.0 is more likely to be a snapshot of the best available sensors at the time they decide to freeze the design - and they'll go with that for the Model 3 and updated S/X. And that technology will also be "old" in several years.

A big improvement will be the processor - adding something like the NVidia Drive PX 2 - essentially a supercomputer in every car, which is design to do many operations - in parallel. Image recognition is a major element of the AP software and applying more processing power will allow better real-time recognition of objects from each of the sensors and then comparing the images between overlapping sensors to better detect each object. AP 1.0 is using the same processor as other onboard software and will always be limited in how much sophisticated processing can be done on the images.

And the last piece is the software. As Tesla did for AP 1.0 and others have noted they'll likely do for AP 2.0, it's very likely AP 2.0 hardware will go into production - and it will take a year or longer before the software catches up and can take advantage of the new hardware.

We're all guessing here... My guess is that we'll see AP 2.0 late this year or sometime next year anytime up until they announce the Model 3 specifics and begin taking actual orders.

 

[Topher]

I did prove it, IF you limit the inputs to a few discrete inputs, like gambling. Dealer shows an 8, you hold a 5 and a 6. Sometimes people will hit, sometimes those exact same people, with the exact same cards, will double down. And even sometimes, they'll hold.

Not only isn't that a proof, it isn't even an example, nor would your caveats even apply to your original claim. I won't bother explaining card counting to someone who thinks holding on an 11 is something people might do.

Really, proving humans are imperfect and fallible is one of "humanities most challenging philosophical questions"?

That isn't what you claimed to have trivial proof for. You claimed you could prove that humans could decide differently given identical inputs.

By the way, I never said anything about "free will."

Free will was precisely what you were claiming, you just didn't know you were doing it.

Oh well. No fame and fortune for us today. Done here.

Thank you kindly.

 

[Topher]

Tesla's systems have not yet reached what Google has already achieved but I bet Tesla will catch up very soon with fleet learning and a few more radars and cameras.

Google has staked out the slow speed regime, Tesla, the high speed controlled environment regime. Hopefully they will meet in the middle and trade algorithms.

Thank you kindly.

 

[Alketi]

Any sufficiently advanced technology is indistinguishable from magic

I think this is one reason that some people can't comprehend the reality of a self-driving vehicle and thus are imagining a far more distant timeframe.

Yet, despite that, we have virtually every car company, who is intimately working with the technology, saying it'll be a reality sometime within the next 5 years --

Ford (L4 - 2021) Ford Targets Fully Autonomous Vehicle for Ride Sharing in 2021; Invests in New Tech Companies, Doubles Silicon Valley Team | Ford Media Center

GM ("sooner than you may expect") GM’s first fully autonomous car will be electric and launch on Lyft

Volvo (L4 - 2020) Volvo's first self-driving car has a big edge over the competition — and it's coming sooner than you think

BMW (L4 - 2021) BMW’s Bold Plan to Make a Fully Self-Driving Car by 2021

Audi (L3 - 2018) Audi kicks the race for self-driving cars into a higher gear

Tesla (L3 - 2017+, L4 - ???) Elon -- "A hell of a lot sooner than most people think. It blows my mind."

That was just a 5-minute search.

To not believe these dates is to believe that everyone in the industry, who are both testing and developing the technology, is either lying or wrong, solely based on our ability to comprehend the technology from the outside looking in. How long regulators take is another question, but the way things generally work is that someplace will allow these cars first (like Michigan), and then as the safety records become more and more apparent others will follow, state-by-state, and then federal and country standards catch up. And, then it's everywhere.

 

[RogerHScott]

Tesla's systems have not yet reached what Google has already achieved

Good thing there's no real difference between a laboratory (even if it is as big as the Bay Area) and a product. Man, it's annoying how
stupid and lazy all those commercial enterprises are!

 

[Stoneymonster]

Any sufficiently advanced technology is indistinguishable from magic

I think this is one reason that some people can't comprehend the reality of a self-driving vehicle and thus are imagining a far more distant timeframe.

Yet, despite that, we have virtually every car company, who is intimately working with the technology, saying it'll be a reality sometime within the next 5 years --

Ford (L4 - 2021) Ford Targets Fully Autonomous Vehicle for Ride Sharing in 2021; Invests in New Tech Companies, Doubles Silicon Valley Team | Ford Media Center

GM ("sooner than you may expect") GM’s first fully autonomous car will be electric and launch on Lyft

Volvo (L4 - 2020) Volvo's first self-driving car has a big edge over the competition — and it's coming sooner than you think

BMW (L4 - 2021) BMW’s Bold Plan to Make a Fully Self-Driving Car by 2021

Audi (L3 - 2018) Audi kicks the race for self-driving cars into a higher gear

Tesla (L3 - 2017+, L4 - ???) "A hell of a lot sooner than most people think. It blows my (Elon's) mind."

That was just a 5-minute search.

To not believe these dates is to believe that everyone in the industry, who are both testing and developing the technology, is either lying or wrong, solely based on our ability to comprehend the technology from the outside looking in. How long regulators take is another question, but the way things generally work is that someplace will allow these cars first (like Michigan), and then as the safety records become more and more apparent others will follow, state-by-state, and then federal and country standards catch up. And, then it's everywhere.

Sorry that I can't like a post more than once.

 

[RogerHScott]

To not believe these dates is to believe that everyone in the industry, who are both testing and developing the technology, is either lying or wrong

Do you believe these statements are primarily coming from chief technologists or marketing departments? To the extent you accept
that the latter may be involved is it that shocking that there may be a little exaggeration going on here?
"Soon", "sooner than you think", "real soon now", "right around the corner", etc., are not dates. The dates mentioned are all in the
neighborhood of five years out. The technology industry -- much like the rest of the world -- has a very poor track record of predicting
what they can/will deliver that far out in the future. Not only do they over-extrapolate current trends, as has been pointed out above they
also totally miss new things that may come along in the interim.