In reply to tfb.

I agree it is a bad idea to build a system that puts people in a position to make flawed safety critical actions.  We know how that problem is mitigated, but this isn't the proper forum to discuss that.  The real problem with AV safety has little to do with tech and is mostly a matter of capitalism vs. a fair and safe public realm.

In reply to thielges.

That's missing the point.  Inherently large an (not inherently but probably) variable delays make a human in the Philippines driving the vehicle in the sense a human drives something with no automation implausible.  But the question is does the human in the Philippines ever have to make safety-and/or time-critical decisions?

Airliners mostly fly themselves because controlling an aircraft turns out to be a much more tractable problem than controlling a car, most of the time.  That doesn't mean an airliner with a variable few-tenths-of-a-second delay in its flight controls would be sensible.  Still less does it mean that replacing the qualified pilot by someone you hired off the street is sensible.  Or legal, in either case.  That is, at minimum, what these fucks are doing.

In reply to jwz.

Maybe it is better to express this in the negative: general remote driving would not work to fool people.  Not because signalling delays would turn remote drivers into drunk or senile drivers, but because human variability would shine through and lift the curtain on the mechanical Turk.

In reply to tfb.

This whole thing reminded me of a short story I read a very very long time ago: it turns out to be A meeting with Medusa by Arthur C Clarke.  Part of the plot is a crash of a vast airship caused by a camera drone's control switching to a circuit with a substantial speed-of-light delay causing its pilot to lose control of it.

If only the techbros could build something as cool as a giant airship to crash, instead of just killing pedestrians with their pretend-automated cars.

In reply to thielges.

You can imagine all sorts of things.

But what you are doing right now is giving the benefit of the doubt to the known-liar company who won't allow any third party to examine how their systems actually work. All you have are the carefully-tuned wordgames of whichever CxO drew the short straw this week and had to go get grilled by Congress (not under oath).

So any attempt to favorably interpret their words into "well maybe it's technically true" is just caping for them.

Always assume that what they are doing is maximally stupid, dangerous and venal, until explicitly verified otherwise.

It is possible the remote "drivers" aren't really driving but instead just nudging stuck cars.  You can imagine an AV coming onto a situation that doesn't match any scenario it knows.  For example a wreck with paramedics, tow trucks, and police mingling.  The AV might just stop and call for help.  The human at the other end might be able to figure out that cars are being directed to the left into an oncoming lane to navigate around the crash. "Veer slowly to the left and follow that red car with license plate ABC-123 for 80 meters and then back onto your lane.  Then resume autonomous mode."

I'm now waiting for some business to appear where cars are self-driving but a purported "passenger" can "provide guidance", and for it to become a huge hit with people too young to get licenses, and also with people who have been barred from operating such vehicles.

But I guess then we'll anyway just see the US capitol hill doing gymnastics to legalize it...

(That said, I have no ideia if such a thing already exists. It probably does...)

Can we please stop getting business ideas from Charlie Stross?

From Halting State, almost 20 years ago: "You try not to shudder: You hate the whole idea that some bored drone pusher in a remote driving centre has got your life—and half a dozen other lives—in his hands. At least on the motorways the cars steer themselves, that’s within the capabilities of today’s AI."

In reply to Kyzer.

That's not my intention.  The best possible case is increasing the human reaction time by about 120ms.  That's about the same increase that is the legal
limit for drink-driving (in the UK the limit in Scotland is lower, but elsewhere it corresponds to this).

So the best theoretically possible reaction time impairment is right on the limit for alcohol.  If you assume the realistic achievable round-trip time is more than double this, and that each human will be looking after more than one vehicle you're looking at the equivalent of a person who is too drunk to stand controlling these things.

This is criminal.

In reply to CSL3.

The 'human assistance' is likely 2-4 reaction times away.  Say 1s.  13 metres at 30mph.  The child is dead before the human even knows they're there.

In reply to tfb.

You're making this sound dangerously plausible. Perhaps Waymo should maintain collections of poor people in lots of countries, so they can remotely drive your taxi for lower wages than the poor people in your own country. Show customers a list of countries with ping times vs cost-of-living index, so they can pick who will remotely drive them today. Even though it costs more than just paying a local person, because Waymo needs a cut.

We could also get poor people in foreign countries to scan your groceries... oh, Amazon already tried that.

What about poor people in foreign countries delivering packages or cleaning the streets? Mucking out drains? Caring for the elderly? Arbitrage has never been so futuristic.

In reply to Stuart Longland (VK4MSL).

The great circle distance from Manilla to SF is 11,224km.  Speed of light in fibre optic cable is about 2/3 c.  So one-way transmission time is about 56ms.  Two-way (so the very shortest response time) is therefore 112ms.  That's a lower bound, assuming any processing time is zero, no packets are lost and so on.  It's not zero, packets are lost.

Human reaction time might be 250ms.  

In reply to CSL3.

The boot-stomping-a-face future is people paying money for the 'post dad game' where, instead of gardening or driving your kids to their activities, you sit on the sofa 'playing' in control of the robots doing those same tasks.

K3n.

In reply to Louis.

Maybe they have some sort of Alameda-Weehawken Burrito Tunnel thing going on, but for fiber.

So Waymo, a US company of some sort, employs people without California drivers' licenses as commercial drivers and puts them on public roads.

That seems like a bit of a slam-dunk for a prosecutor, but what do I know? I don't have millions to bribe city/state officials with.

In reply to Stuart Longland (VK4MSL).

26 ms is faster than the speed of light over a direct fiber optic cable from the Philippines to DC..

ah the classic ‘is it advanced tech or just mturks’

In reply to John Ripley.

I said then, and I'll say now, that it would probably be cheaper to just have ICE abduct some kids, stuff them in trunks, and pay them 3 cents an hour (40% of which goes to ICE). Let's get disruptive!

"I'm sick of AI=Actually Indians jokes. Get contractors in a different country this time."
--Some Waymo exec, probably

In reply to Zygo.

This Taxi Could Have Been a Zoom Meeting.

"Are you muted? I think you're muted."

In reply to Stuart Longland (VK4MSL).

Pinging a host in a data center won't tell very much about how long it takes a packet to traverse the wireless network surrounding a moving vehicle.  Roaming, dumb carrier tricks, unfavorable QoS...literally anything you could insert in that list adds latency, the only question is how much.

Assuming "remote driving" means "live digital video and/or audio", there will be also codec and framing delays before and after the data is on a network.

Because the robo-taxi needs human assistance before it runs over a kid. Hell, the human probably thinks of it as a real-life version of Crazy Taxi or somethin'.

time to post up some cardboard signs around town

In reply to Notavi.

https://www.c-span.org/program/senate-committee/tesla-and-waymo-executives-others-testify-about-self-driving-cars/672835

The "curious" quote seems from the 01:31:10 mark of the Senate Commerce Committee, 4 February 2026, and it seems the senator quoted was Ed Markey, United States Senator from Massachusetts. No idea why the author at the linked article chose to not provide details.

In reply to Doctor M. Popular.

💯 also JWZ can we get Dread Zeppelin to perform? Sorry for even asking.

One oddity about the first article - they quote a senator grilling them about their use of "advisors" in the Phillipines, but as far as I can tell they never identity the senator.

Not sure what to make of that, but it does seem like an odd omission.

(Quote post) 🗣 TOLDYA!

fwiw, am almost certain i read about the remote control Waymo con on EU or Brazilian media. it’s just that mainstream USA journalism has been taken hostage by the techoligarchy

jwz:

Waymo Exec Admits Remote Drivers in Philippines.

Mechanical Turk, meet Mechanical Filipino: Mauricio Peña, the company's Chief Safety Officer, confirmed under questioning that the Google subsidiary employs human operators abroad [...] "They provide...
https://jwz.org/b/yk3E

There was some prior, unrelated bombshell about this, I think which came out in discovery: each additional Waymo on the streets requires closer *two* people to manage it, than the one you'd need for a taxi/uber/lyft. So they're not even an efficient form of automation. What's compelling, though, is the operators don't need to be local, or even in the same country, and you don't even need to pay those pesky income taxes to the local government.

In reply to Stuart Longland (VK4MSL).

For reference (I'm assuming `whitehouse.gov` is hosted in the US and isn't using some anycast CDN. It was the first US-based site I could think of.)

```
stuartl@rikishi ~ $ ping4 -n whitehouse.gov
PING whitehouse.gov (192.0.66.51) 56(84) bytes of data.
64 bytes from 192.0.66.51: icmp_seq=1 ttl=56 time=25.9 ms
^C
--- whitehouse.gov ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms
rtt min/avg/max/mdev = 25.850/25.850/25.850/0.000 ms
stuartl@rikishi ~ $ ping6 -n whitehouse.gov
PING whitehouse.gov (2a04:fa87:fffd::c000:4233) 56 data bytes
64 bytes from 2a04:fa87:fffd::c000:4233: icmp_seq=1 ttl=47 time=221 ms
^C
--- whitehouse.gov ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms
rtt min/avg/max/mdev = 220.696/220.696/220.696/0.000 ms
```

25ms might be low enough to not care… but 221ms definitely is "drunk driver" territory.