The US Air Force successfully tested AI-controlled jet fighter

Ethics don't go out the window because of autonomy. It's dark, but it's not somehow easy to shrug off as "ok no human then no ethics". We already have semi-autonomous weapons such as CIWS (Goalkeeper/Phalanx) which are autonomous because it's very tedious for a human to sit idle and wait for threats and then it's really hard to hit them. Perfect job for a robot. But the ethical job of the human in that case is ensuring that the weapon isn't armed if there is too high risk of collateral damage (These sit on ships, typically, so a human decides whether it's safe to arm the weapon given the situation around the vessel).

The dumbest autonomous weapons are mines, they are hundreds of years old. The ethical decision is made when laying down the mine, not when the explosion is triggered. So yes it's a slippery slope, but the "human in the loop" isn't only ethical when it's the kill decision. There is a "human in the loop" when laying a mine too. The ethics involved consists of weighing the risk of killing the wrong target, properly documenting where they are planted and so on. There is no human "in the loop" at the time the mine kills someone. But that doesn't somehow throw ethics out the window. The person deploying the mine - or the slaughterbot - is the person in the loop.

> In 20 years time we'll probably have aircraft that do figure eights over enemy territory for hours before automatically engaging 'military aged males'.

I think that is unlikely, but possibly "armed" personnel. We already have craft that do figure eights for hours before automatically engaging specific signatures of vehicles (if we'd want to). And as I said, even artillery shells do this. But yes, finding a concentration of troops and deciding whether they are soldiers/civilian isn't even 20 years away.

Of course there is. "AI" controlled hardware doesn't care about getting out alive. A human is more likely to have an itchy trigger finger because a mistake means death.

The promise is already a farce because autonomous weapons that sense the environment and make unilateral decisions to kill have already been in use for decades now. From low-tech landmines to high-tech ship-classifying CAPTOR torpedoes which fire themselves, a weapon sensing its surroundings and deciding to kill is nothing new.

There are many factors on and off the battlefield, such as the domestic and geopolitical spheres, that go in to targeting. Only at the minimum level of an AGI that’s capable of understanding human intentions and situational nuances can potentially remove a human from the loop.

That does depend a lot on the detection vs kill range vs time to kill. In a close air dog fight sure but most modern fighter doctrine is focused around long range kills and stealth already. Cannons are nearly vestigial on the F35, it only has ~4-5 seconds of ammo worth to fire.

> The last war broke out. All the people are dead, but the machines continue dutifully to follow orders.

Other problem is that they won’t be cheap. High performance military aircraft are expensive. My guess is only get factor of two in cost. Also, there is size tradeoff with payload. Drone that can’t carry AMRAAM is useless.

The purpose of fighters have changed. Missiles have taken over many of the roles. The roles that remain require intelligence. Both Ukraine and Russian fighters are not being used much because SAMs make things too dangerous. When they do, it is low level attacks that would be hard for drones.

I think developing drone fighter AI would be hard. Recent AI advances have been more about generation than planning. We could do it but would be expensive; look at how much has been spent on self-driving. The basic things are simple but the important things are complicated. For example, drone could have intercepted and shot down balloon but couldn’t have inspected balloon or involved in planning.

Finally, the US has pivoted from developing drone fighters to bomb truck. The idea would that accompany fighters and carry more ammo. It would be less maneuverable and simpler and use extra space for fuel and bombs. I think it is telling that multiple countries have tried drone fighters and none have gone for them.

One of the reasons for it being pilotless is it can fly moves/turns that create G forces no human pilots could stand with their jump suit.

But you know, Taranis has all gone very quiet, slipping out of the public's memory.

This theory is in the latter part of Tyler Rogoway’s “The Alarming Case of the USAF’s Mysteriously Missing Unmanned Combat Air Vehicles”

https://www.thedrive.com/the-war-zone/3889/the-alarming-case...

Carrying enough fuel to go some far distances and back takes up space. Missiles and bombs aren't always small. If you're carrying a lot of payload/fuel weight, you're going to need to have enough surface to generate enough lift, etc. If we're planning on these birds to run similar-ish kinds of activities, I don't know they'll be that radically different. But maybe the missions will be radically different.

> A jet fighter can very much be considered disposable

A jet fighter still usually costs a few million to acquire, before the cost of the pilot even come into the picture. Once again, if they're gonna be built to handle the same mission profiles, I don't know the acquisition cost is going to be radically different. Probably a bit cheaper, but what costs more the seat or the engine? The stick or the hydraulics that actually actuate the wings? I wonder what the real savings are to take the seat and life support out of the plane. Like if you made something like the F-35, but no seat and no life support and no fancy targeting helmet, what's the per-unit cost? Something tells me still probably a few million. It'll be a lot cheaper to lose one in the end not having a human in the seat, don't get me wrong, and I imagine there would be less hesitation to deploy them in riskier deployments since there's no human in the seat, but I don't know I'd use the term disposable to talk about a few million dollar piece of hardware.

But I doubt pilots will be dog-fighting or doing their own bombing runs in the near future.

Is how we get SkyNet.

Hard pass.

https://youtu.be/d9RNmuBm27s

And if I were to write this 20 years you might have still dismissed me as a kook or doomsday prophet, but now one could argue that's already half-implemented anyway rather than a description of a future dystopia.

Even if you need a human crew to handle the stuff around the coasts it seems like a good place to start with automation.

Maybe something like a Jetson for the ANN processor? [2] I'm sure NVidia could provide a milspec part for their use.

[1] https://www.nasa.gov/centers/dryden/research/IFCS/index.html [2] https://developer.nvidia.com/blog/jetson-nano-ai-computing/