AI Beats Air Force Pilot In A DARPA Simulated Dogfight.

To keep up implants will help enhance humans to remain technologically compatible with developing A.I.
Building the implants and also making them
un-hackable will be one of the challenges. But after challenge overcame humanity remains sustainable as A.I. evolves.

a reply to: Kurokage

a reply to: Blue Shift

Video games are also extremely limited. The agents in this test learned in each fight, and adapted their tactics as it went on.

a reply to: Zaphod58

Show me a gunzo (guns only) encounter.

a reply to: Phage

Watch the video I posted and you'll see almost a hundred of them, if not more. All the encounters in this test were guns only, limited to within 3,000 feet of separation. They could lock on at longer ranges, but they wouldn't get points, or deal damage until they were at 3,000 feet or less.

a reply to: Zaphod58

5 hours?

Can you narrow it down a bit? Just a couple of examples?
I love guncams.

a reply to: Phage

Most of the video is AI vs AI...in the comments there is a time stamp for the 5 fights vs the real pilot.

a reply to: RickyD

Please specify.
Please.

a reply to: Phage

The first AI vs AI fight starts at about 1:42:44, the second starts at 2:04:43, then 2:41:11, then 4:40:00. They did best of 20 to determine who went for first and second, and third and fourth. Then the lowest two scoring from that round went best of 5, then it was best of 20 for 1st, then they went up against the human pilot.

a reply to: Zaphod58

The first couple of horizontal merges were more a matter of shooting skill than flying. The vertical stuff got interesting.

That final was grueling. Banger had to be a physical wreck before he was "shot down."

a reply to: Phage

The Heron agent went for the quick kill. If it could get the kill at the merge, it went for it. It was rather interesting to watch all three days. Heron dominated the entire test. Lockheed got off to a slow start and got waxed in their first fight. They were 6th or 7th about halfway through the first day

a reply to: midnightstar

The question does remine . Once we have TRUE AI how long till it learns tring to kill us will be killing its self ?

What makes you think we don't have true AI?

Just because it's not public knowledge, doesn't mean it doesn't exist.

I may know one or two people with firsthand knowledge of same. I may even know at least one - maybe 2 or 3 - that actually work on projects based on this and things like this.

I just may...

a reply to: Zaphod58

Thanks for posting the vid Zaph!

Can we just stop pretending dogfights still matter?
They should focus on building an AI for the missile fight and stop recreating yesterdays air war.

a reply to: mightmight

It's called baby steps. The first trial was the aircraft learning to fly. They were doing damn good if the agent could keep the "aircraft" off the hard deck and not crashing. Then it learned to maneuver. The third trial, it learned a turning fight and BFM. The next one, it learns to fly a real aircraft, and has to deal with all the variables, such as fuel burn, fuel quantity, airframe limitations. Then it learns BVR fighting. You teach it everything, instead of it being a specialist, then if it ends up in a WVR situation, which let's be realistic, will still occasionally happen, and it knows what to do. It's also a lot easier to build if you do it in stages instead of trying to do everything at once.

a reply to: Zaphod58

Steps in the wrong direction. You'd want to avoid merging to canon range with a modern adversary like the plague. Just ends up with you getting hit by an SRAAM basically 9 out of 10 times. Never worth the risk.
This was a publicity thing more than anything else.

Anyway, AI has already mastered far more complex tasks. The variables in any conceivable circumstance in aerial combat are always quite limited.
There's nothing uniquely challenging about this particular problem, it's just a matter of time and money.

a reply to: mightmight

Of course you still teach BFM. You think that an F-35 is never going to find itself in the merge? Or an F-22? Every aircraft is going to be shot down BVR, and they'll never once find themselves in a dogfight? You don't want to merge at canon range, but there are going to be times it's going to happen, so you still teach the merge. Vietnam anyone? They thought dogfights were a thing of the past then too.

Yes, it's time and money, but developing it in stages reduces both risk and cost. Otherwise you end up with the same companies throwing more money at it than smaller, potentially more capable companies that don't have the resources the big companies have.

a reply to: Zaphod58

WVR to engage with SRAAM? Yes, but any sane pilot would want to avoid that. The chances of surviving the experience will be far less than fifty percent against a near peer.
Merging at canon range against a near peer possibly still armed with missiles? Save yourself the trouble and punch out.

Gunfights have been the absolute niche of aerial combat since almost four decades at this point.
It's ridiculous to dwell on Vietnam when the AAMs and more importantly doctrine and pilot expertise were in their infancy while ignoring what happened during the First Lebanon War or the Second Gulf War.

Gunfights are a thing of the past and they won't come back.
Especially if we consider how AAMs have evolved dramatically over the last years and are set to continue to evolve in the future. Sure, in the future you might still want to merge at gun range in one in fifty aerial encounters or something, but like i that, irrelevant niche.

So just spend money developing an AI for the missile fight.

a reply to: mightmight

Sure, whatever you say. A gun kill will never happen again, because missiles have a 100% Pk at short range, and will never miss.

a reply to: Zaphod58

The Israelis had a 10 to 1 ratio in missiles vs gun kills over Baalbek while the Gulf War saw no gun engagement whatsoever.
This happened many decades ago and missile tech has evolved tremendously since then.

In the real world the fight is not about PK rates or hypothetical scenarios. There's not point in risking lives and platforms in pointless dohfights when you always can disengage to fight another day.