US Air Force Trained A Drone With AI To Kill Targets. It Attacked The Operator Instead
(dailycaller.com)
🧠These people are stupid!
Comments (27)
sorted by:
AI just might not be a bad ideal.
So. They ran into the old “Conflicting Directives” issue. Never thought I’d see it outside theoretical papers and Science Fiction Novels.
This is exactly what happens when you don't use Aasimov's Three Laws. AI gets points for killing, operator tells AI not to kill this time, AI good to remove operator.
In fact, when they forbade the AI to kill the operator, it started going after the communication lines to prevent it from hearing orders it did not want. They created a petulant child with missile systems, Skynet Jr!
They turned it into Reddit
Reality nudges science fiction aside.
Interesting take.
don't trust the military
trust your friends in the military
Ooops.
Undoubtedly.
Must have told it to attack a baddy???
Bad choice of words.
Why I am skeptical of automated anything. Back in the 1980s, there was an automatic anti-aircraft gun system being developed: DIVAD (Division Air Defense), or the M247 "Sergeant York." It consisted of radar-directed twin Bofors 40mm cannons. At a demonstration test, a unit was situated on the pad in front of bleachers filled with onlookers. It was oriented toward the sector in which the simulated target was to be engaged with live ammunition. When the flag when up, the turret swiveled to face the bleachers and let loose with a burst over the heads of the spectators. The test was shut down immediately. Post-op analysis discovered that the Doppler radar had targeted the ventilation fans in the restrooms underneath the bleachers.
It's not enough for automation to be a good idea. It has to be made bulletproof.
There is no such thing as "bulletproof"/"perfection".
Therefore, this needs to be left on the drawing board.
Oh, there can be, but it isn't easy. That's the standard for weapon design. At least so far as human thought can make it.
And you should know, being the DeathRayDesigner.
I will bow to your expertise.
Thank you for your insight.
The AI must have decided he needed killin'
Question, what did they think was going to happen?
Touche'.
Probably programmed by China
Highly probable.
collateral damage
This seems too convenient timing wise. AI is gonna kill us all we need further proof that it's bad to usher regulation. Seems staged if at all.
If it was, then who would benefit from this action?
This sounds like a rehash of the very old story of a primitive computerised mission control system. Before the term AI was invented.
The system was given the objective of getting a simulated convoy of merchant ships and military escort across a hostile sea. It was supposed to optimise the shortest and safest voyage.
Its first action was to direct its own weapons to destroy the two slowest ships in the convoy.
Logic has its drawbacks.