62 comments

[ 3.4 ms ] story [ 125 ms ] thread
The Twitter post in question (https://twitter.com/wholemarsblog/status/1557112809132945408) says “(this is a serious request)”.

Fine, then let’s treat it like a serious request. Apparently someone volunteered a child (https://twitter.com/wholemarsblog/status/1557206098876194816, presumably their own child).

Has this (pseudonymous) person done a proper study workup? This is a proposal to put a child into a potentially life-threatening situation. I hope they’ve written up (and had reviewed) the rationale for why this can’t be simulated first. And I hope they’ve gotten the snapshot of the Tesla software used, for reference.

And for the person who volunteered their own child, I wonder if they should be talking to CPS right now.

The person who made the Twitter post mentioned the Bay Area. If they are serious, try going to a place like the Center for Automotive Research at Stanford, talking to the people there, and doing a study.

> And for the person who volunteered their own child, I wonder if they should be talking to CPS right now.

And the fan who made the request should get his driving license yanked indefinitely.

He just proved without a doubt that his character is unfit to let him drive a potentially deadly vehicle on public roads.

Seems easiest solution is for those wanting to test this just use themselves. After all, a squatting or sitting adult Is same height as a child, it would be the perfect substitute!
We can squat. Some of us just need a little help getting up.
> Don’t force a kid to walk in front of your 4,000-pound metal box traveling at god knows what speed

Some consolation, according to the tweet chain, is that it'll be at 5MPH and the child wouldn't be in front of the car (just on the sidewalk approaching the road). Still not sure the need to put an actual child at any risk when the car stopping for a dummy would prove the same point if not stronger.

There's a non-zero chance the Autopilot will think the kid is a freeway center divider and accelerate into it.
well at least we have 4 years to make good on this world if that's the case.
(comment deleted)
There is something seriously wrong with Elon fans.
nah its just Iterative testing :-)
There is something seriously wrong with people who have critical thinking skills who are unable to recognize the fabricated bullshit being written to discredit anything related to Elon.
Instead of writing on open letter why not reply to the Twitter post?
Had to make it sound official.
The open letter shows more people how crazy those Tesla fans are.

A reply won't broadcast.

Yeah, come on guy. Don't use a kid, use a little person. They can consent.
Agree with not using an U-18 human for the consent part
The person in question has already written a response to the open letter: https://twitter.com/WholeMarsBlog/status/1558122848601812993
> I’ll run over as many kids as I want and you can’t stop me!!

I'm actually not expecting the Tesla to run over the child (because either it will stop, or the driver will intervene and slam on the brakes or avoid the child). But it also gives me 100% confidence that this person is the exact wrong person to perform any scientific test of the system to even remotely validate its safety.

Anything can go wrong, will (eventually) go wrong.
Does a proximity to Elon Musk make people forget that optics are a thing? Yeah this is a dumb joke, making a dumb joke about running over children still makes you look like an asshole.
In the Muskiverse, that's called "being edgy", and it's considered a good thing.
Musk has that quality that Steve Jobs had - the so-called RDF or "Reality Distortion Field". Proximity to Musk seems to scramble a lot of people's minds and thought processes.
It kind of worked for those two because they made good stuff - iPhone, cars, rockets etc. You can get away with saying this will be great if you deliver occasionally.
It's a way of getting publicity. Seems to have worked for Musk and Trump.
Trolling on the internet for fun has been a thing long before Musk, and will be a thing long after.
Can someone explain to me what the point is in proving a Tesla won't run over children provided it detects they are real?
Nothing. This is idiotic attention seeking. Software engineers know that you don't set up a demo and proclaim that a previously demonstrated edge case isn't real because it "works for me". You just fix the issue. That's it.
Of course this half baked system that requires constant over the air bug fixes would plow through a kid in a crosswalk.

What puzzles me is that people actually bought into the marketing hype. Its a prototype quality car running alpha software…

So I've mentioned this before -- traditionally, journalism involves reaching out to the person you're writing about for comment before posting it. I wish they'd mark off as "opinion" pieces like this that clearly don't follow those methods.
It's an open letter. Did you expect an open letter not to be opinionated?
Why would you use a real child? That's so dumb
The car would know otherwise. It has to be real or the car won't take the exercise seriously.
This is sarcasm and not actually true right? Presumably a full self driving car wouldn't run into a stray shopping cart in the parking lot
Even if the AI could tell the difference between a cardboard cutout and a human, I still wouldn't want my car to run over a cardboard cutout, or a box, or anything else on the road. I don't trust it because I know first-hand that Tesla FSD consistently runs over both large pieces of debris and large potholes in the roadway that every other vehicle on the road avoids. So yeah, I don't trust the car is smart enough to avoid hitting anything other than another vehicle or bicyclist. I don't rely on it at all. It was a waste of $10,000. After two years' experience with FSD, I absolutely would not buy FSD again if I were buying another Tesla right now. But I still don't believe it's something for lawmakers to get involved in.
Personally I wouldn't be relying on a car not to hit anything at all, at least, not in this decade. To me, it seems like a last chance dice roll, to lean on the scales a bit and maybe help in a case where I mess up.
How many times would you need to test it to verify it worked? Different time of day, weather, lighting, etc could change the outcome.
Actually it doesn't matter how many times it has been verified working properly. One single failure would be more than enough...
Hmm, has there been one single failure of a human driver hitting a child?
That is the difference between an accident and murder.

If you try your best to adhere to all best practices, and still a bad thing happens, that is an entirely 100% different thing from the bad thing that happens when you voluntarily, deliberately, knowingly, do something wrong or unnecessary, and then a bad thing happens.

So the AI driver knowingly, deliberately decides to murder children?
get a new pharmacist
I’m not the one hallucinating that AI acts with deliberate malicious intent. Maybe you should get that checked out.
I didn't say or even imply any ai had any intent.

We're talking about whoever wants to stick a kid in front of a car, not the car. Holy cow.

That's not what I was talking about. I was responding to the person who implied that self-driving cars need to be completely error-free, which is an unattainable standard, and far beyond what human drivers are capable of. There will still be traffic accidents with self-driving cars, but hopefully far fewer. Holding them to an unrealistic standard of 0 errors would needlessly kill thousands of people.

It's pretty clear that nobody was ever going to put a kid in front of their Tesla, and The Verge and everyone here is being deliberately obtuse to indulge their Tesla/Musk hate.

I guess he should've run over a straw man instead, like the author of this letter
There are so many deer around california/oregon this time of year driving around outside of the city for a bit wouldn't take too long to find one. Not the same though of course
I mean, it's not even that hard to put together a child-size dummy of ballistic gel (a la Mythbusters). Doesn't have to be a work of art, and it should be indistinguishable from the real thing (in terms of radar signature, etc).

Why not just have someone put one on a low cart and drag it along the road in front of your speeding Tesla, and beta-test it that way? Isn't Tesla doing that?

You don't even need to match the radar signature of humans, it shouldn't be driving through any object of that size in the road.
I think he should let the kid drive the car, and then walk in front of it himself, on his knees so he's kid-high.