Wow, so they let AI fly an F-16 fighter jet | Hacker News Reader

Wow, so they let AI fly an F-16 fighter jet

theregister.com

26 pointsrntn40 comments

War Zone has a less sensationalized, if prior stage, article with context: https://www.thedrive.com/the-war-zone/39899/darpa-now-has-ai...

I don't understand the use-case of adapting human piloted offensive military aircraft to autonomous operation. A pure UAV would be more optimized for this role.

I can't find the reference, but a business jet was experimentally converted to autonomous operation (under human supervision and remote control ability).

There might not be UAVs with the same performance as a real fighter jet.

An F-16 is certainly capable of doing more than a human pilot can withstand, so from that standpoint you could improve the capabilities of an F-16 with an AI pilot. But the aircraft is still designed around a human pilot with human variables in mind. A drone crafted for an AI could have substantially better acceleration and banking capabilities, but I believe most drones today are the very high altitude very long range variety and aren't focused on dog fighting capabilities. A purpose built drone dog fighter would be dramatically more capable than an F-16.

That might be where it ends up, but it would be silly to design a new platform before the AI is well understood. Better to retrofit something existing.

I'd agree. Proof of concept with existing technology to better understand the possibilities and problem areas. If you prove the AI can effectively fly an F-16, they are in a better place to build a drone suited to purpose.

We don't know what purpose drone fighter would be capable of. There are physical limits beyond limits of the pilot.

One indication that nobody is investigating high-maneuverability drones is that the demonstrators are mostly stealthy flying wings. Flying wings are less maneuverable and planes need stabilizers for that. It is the difference between B-2 and F-22. I think we will see drones in more in support roles; Navy made an aerial refueling drone.

Also, the tactics for air-to-air combat have changed. It doesn't matter if can maneuver more if can't outrun a missile. Russia still emphasizes dogfighting but the US is all about missiles and energy management.

It's called an SM-2, and it already exists.

It costs hundreds of millions to design a new jet, much cheaper to adapt an existing design.

Because this way you can reuse existing hardware, as well as gauge the AI's performance compared to human pilots on a similar platform. And, you're developing options for upgrading F-16s to autonomous, uncrewed jet fighters.

They've got F16s that are in use as drones right now. This is just adding the HiveMind AI to it.

I know they’ve used autopilot software for years (decades?) to fly old retired jets as drones, I guess for real world testing of air-to-air weapons. Anyone know how different this AI dogfighting system is?

Much, much more sophisticated. As far as I know target drones don’t even maneuver, let alone fight back.

I think they are transitioning from a simulator based test to real world testing. This system is adding BFM capabilities to the fighter (Dogfighting). Up to this point, drones are capable of BVR combat only. They were essentially automated missile trucks. This meant that once the jet got within a certain range, it was helpless and would have to retreat. It was kind of the missing link in combat air platforms.

This new AI would allow the jet to press on if it were to be surprised by an undetected enemy within 5-10nm. This is really important in mountain areas, and places where you don't have AWACS. Sending automated fighters ahead to send back link-16 data to manned fighters is a HUGE advantage when your enemy is shooting down your AWACS with long range air to air missiles. Also it allows you to leverage more airframes if you are short on pilots and preserve existing pilots.

Eh, I mean I'm not a pilot either and they let me fly one (was a guest in the back seat of a USAF Thunderbird and got to take control for a bit).

Wow, dream come true. How did you manage to get that ride?

related question to all in the know, why aren't UCAVs Unmanned combat air vehicles a bigger thing? I know there's some experimentation going on, but technologically it doesn't appear to be a big leap to expect that. Is it the scary-perception issue?

I guess because they are just being called missiles. Modern missiles are surprisingly capable devices, being able to autonomously maneuver, follow waypoints, and loiter before striking

Aren’t these called drones, and haven’t they been around a while?

Predator drone entered service back in 1995 per Wikipedia, and it’s use only skyrocketed since (Iraq, Afghanistan, etc) There are other models like Reaper

Turkey, Iran also seem to be big players here, among others

There are no drones in service (that I know of) that are designed for air to air combat - they are designed to attack ground targets.

Is air-to-air combat really a thing now? I was under the impression it was less important with advances in weapons guidance systems.

Yes, US aircraft have been participating in air-to-air combat in Syria in recent years against manned and unmanned aircraft.

In particular, the US military relies more on aircraft for air defense and air superiority while post-Soviet militaries rely more on surface to air missile systems.

I've been wondering the same thing! Teaching an AI to fly an F-16 is like putting a motor in a horse-drawn carriage. Once you get it working, it should be clear that your vehicle's design was dictated by a number of old constraints which no longer apply, and a total redesign is in order.

Putting a motor in a horse-drawn carriage would have been a pretty monumental leap in capability at the time horse-drawn carriages were commonly used.

And that's exactly how the automobile was invented. Look for exemple at Delamarre-Deboutteville's car[0], one of the first.

[0] https://all-andorra.com/fr/automobile-delamare-deboutteville... (In french but a lot of picture)

Obviously. But folks weren't driving around in motorized carriages, complete with empty driving harness! They adapted the rest of the vehicle immediately, for obvious reasons. It would have been strange if they hadn't!

Now, human fighter pilots are largely obsolete. We've known this was coming for decades, yet we have those AIs flying giant hulking planes that are designed to carry human beings around inside of them.

My point is that this is just as ridiculous as driving a carriage with an empty harness!

The short answer is expense; The US has two such programs in development (both called NGAD) and they're likely to be the most expensive aircraft ever developed by the Air Force and Navy.

This 70 minute talk covers all of the current 6th generation fighter programs in development and a major running theme is extreme cost. https://youtu.be/RPrWm6fWuaM

NGAD "might" have a loyal wingman component but you can bet your 401K it'll be crewed.

UCAV are a big thing but they don't look like sexy fighter jets. Most of them are smaller and slower like Predator or Bayraktar TB2 that Ukraine has been using. By not having pilot, they can spend a long time loitering and are cheaper. But they are super vulnerable to missiles and fighters.

They also don't make LTs into Generals. Politics is huge in the USAF.

Probably because the last time the USAF was dogfighting regularly was in Korea during the 1950s.

Air war has been ridiculously boring.

If only they'd train AI to hunt down, prosecute and imprison war criminals.

This would be a much better use for it.

This seems like a recipe for disaster.. Skynet vibes in all fairness.

Exactly just see how they were not able to get Sarah Connor and John Connor. I expect some criminals hiding from it as well which would be total PR fiasco.

People get randomly locked out of Google or Facebook accounts because an algorithm screws up.

There was a huge scandal with government benefits being denied to immigrants because of a bad AI program.

Now you want to give it missiles?

There will be no 100% ai, always a human-in-the-loop.

I don't trust most humans to be fully aware what they're doing themselves most of the time, never-mind trying to fully understand what the AI is doing/thinking.

You underestimate the likelihood for major confluences of incompetence in the AI babysitter role.

One of the things we are learning is that semi-automated driving, where the car handles spacing and lane alignment.. can actually lead to more instances of drivers not paying attention.

Partially automating systems can be very dangerous because the human in the loop becomes a passive observer who automatically presses the approve button.

This happened with drone strikes in Iraq. We used a program to calculate movements of targets and classify them. The humans in the loop stopped critically thinking about the programs output and just started blindly approving everything it put forward as a target.

Legal AI used for prosecutions is it's own special little nightmare.

Reply on news.ycombinator.com