News for hackers. Stuff that matters.

A comment on YOLOv4: Optimal Speed and Accuracy of Object Detection. Parent comment

Re:YOLOv4: Optimal Speed and Accuracy of Object Detection by 01100011 on Monday April 27, 2020 @04:33PM (#22999728)
The only thing stopping the robot apocalypse is the power problem. We have every technology necessary to enable mass casualties from commodity robots except long battery life. We don't need AI. We need decent enough pattern recognition that can pinpoint a weakness in the human body.

I often use my 'Stabby the Robot' diatribe to convey this to people. Imagine a robot that moves too fast to evade, that doesn't rely on expendable, limited ammunition(i.e. a blade or sharp surface), and can locate the jugular vein of a human. You could make one now out of a commodity drone, but it wouldn't last very long because of the power problem. AI would obviously make it more dangerous, but it isn't necessary.

We've already placed very dangerous tools into the hands of humans. Maybe it helps AlexeyAB sleep at night to stop working on YOLO, but his idle hands are not holding back the future.

[ Parent | on HN ]
Re:YOLOv4: Optimal Speed and Accuracy of Object Detection by genpfault on Monday April 27, 2020 @07:34PM (#23001484)
> I often use my 'Stabby the Robot' diatribe to convey this to people. Imagine a robot that moves too fast to evade, that doesn't rely on expendable, limited ammunition(i.e. a blade or sharp surface), and can locate the jugular vein of a human.

Oh hey I remember that write-up on qntm[1]!

[1]: https://qntm.org/robots

[ Parent | on HN ]
Re:YOLOv4: Optimal Speed and Accuracy of Object Detection by 01100011 on Tuesday April 28, 2020 @12:34AM (#23003403)
Never saw that but thanks for the link! It's a much better wording than I'd ever have come up with.
[ Parent | on HN ]
Re:YOLOv4: Optimal Speed and Accuracy of Object Detection by IdiocyInAction on Monday April 27, 2020 @04:59PM (#22999968)
While your robots do sound like they could kill a lot of people, I feel like this is a bit overblown. For a start, for "apocalypse", they would have to be produced in massive numbers, i.e. by a nation state-level actor. State-level actors can already make nukes or bioweapons.
[ Parent | on HN ]
Re:YOLOv4: Optimal Speed and Accuracy of Object Detection by 01100011 on Monday April 27, 2020 @05:46PM (#23000488)
What's so hard about manufacturing a few million drones? With a relatively modest amount of money(relative to, say, developing a nuclear program or chemical weapons program), you could hire a contract manufacturer to spin out a few million drones. They probably wouldn't even ask what the blade/needle was for, although it would be easy to craft a cover story(i.e. "They're for mass inoculation/slaughter of livestock"). Update the SW and congrats, you now have a weapon of mass destruction. You can target people by face, race, location, etc.

There's a reason the US military has been playing with drone clouds for years.

[ Parent | on HN ]
Re:YOLOv4: Optimal Speed and Accuracy of Object Detection by philipkglass on Monday April 27, 2020 @06:06PM (#23000664)
It's not hard for a nation to manufacture or order millions of drones. It's also not hard to obtain enough rounds of ammunition to shoot millions of people. It was decades if not a century ago when the primary obstacle to a nation killing millions of people ceased to be "inherent technical difficulty."
[ Parent | on HN ]
Re:YOLOv4: Optimal Speed and Accuracy of Object Detection by 01100011 on Monday April 27, 2020 @06:41PM (#23000972)
It's not hard to shoot millions of people, but it is hard to raise the army required to do so. It is hard to do so without raising suspicions. It is hard to do so without damaging infrastructure.

I rogue state, or even a non-state actor, can leverage cheap drones(assuming, of course, a solution to the power problem) to obtain a weapon of mass destruction. Drones are fundamentally different from bullets.

[ Parent | on HN ]
Re:YOLOv4: Optimal Speed and Accuracy of Object Detection by jefft255 on Monday April 27, 2020 @07:00PM (#23001180)
My roboticist opinion is that you need extremely developed AI to do what you’re saying. It requires navigational capabilities that don’t exist yet except in very open spaces like fields. Also, lightweight drones which could do these fast maneuvers aren’t very robust; you can literally slap them down with some pain. I get that even with a low success rate it could be a disaster if you have a huge amount of them deployed somewhat.
[ Parent | on HN ]
Re:YOLOv4: Optimal Speed and Accuracy of Object Detection by 01100011 on Monday April 27, 2020 @08:08PM (#23001788)
I don't think we've fully solved the navigation problem, but I think we're close and it will certainly be solved before the power problem. A quick google got me this:

https://www.youtube.com/watch?v=yOrtZ38olvc

and I suspect there are dozens of other similar projects out there.

[ Parent | on HN ]
Re:YOLOv4: Optimal Speed and Accuracy of Object Detection by jefft255 on Monday April 27, 2020 @09:13PM (#23002239)
This work is amazing and I’m a big fan of Tedrake; however the gap between what you’re showing here and a drone a chases humans and slice their jugular might be bigger than you think. It’s very hard to extrapolate stuff like that.
[ Parent | on HN ]
Re:YOLOv4: Optimal Speed and Accuracy of Object Detection by airstrike on Monday April 27, 2020 @05:18PM (#23000148)
Reminded me of https://www.youtube.com/watch?v=HipTO_7mUOw
[ Parent | on HN ]