Gaming AI fighter pilot wins in combat simulation

[Notebook]

http://www.bbc.co.uk/news/technology-36650848

An artificially intelligent fighter pilot system has defeated two attacking jets in a combat simulation.

The AI, known as Alpha, used four virtual jets to successfully defend a coastline against two attacking aircraft - and did not suffer any losses.

Alpha, which was developed by a US team, also triumphed in simulation against a retired human fighter pilot.

Don't know why they are surprised I usually get hammered in all my combat sims. Usually looking for the switches...

N.

 

[Gordontrek]

Four against two? I'd like to see a more even matchup against perhaps a full squadron of real, trained Air Force pilots and see if this AI is as invincible as it seems. I have reservations about using this on real aircraft, but it could make real pilots exceptionally good if used for training.

 

[Urwumpe]

Four against two? I'd like to see a more even matchup against perhaps a full squadron of real, trained Air Force pilots and see if this AI is as invincible as it seems. I have reservations about using this on real aircraft, but it could make real pilots exceptionally good if used for training.

Same here. With equal weapons and jets, I would also have a hard time winning against four "faked intelligence" jets to the point of at least harming one of them. I won't even speak about AI there.

I still wait for an AI to even just replace the rear-seater of a classic jet properly. So far all attempts to do so resulted in the pilot being overwhelmed with workload - either for filtering the information or for configuring the computers.

 

[kamaz]

Four against two? I'd like to see a more even matchup against perhaps a full squadron of real, trained Air Force pilots and see if this AI is as invincible as it seems. I have reservations about using this on real aircraft, but it could make real pilots exceptionally good if used for training.

Oh, contrary.

The pilot is the limiting factor in dogfighting, because human body cannot withstand high g loads. An AI-controlled plane will be able to fly sharper turns, so it will outmaneuver any human pilot. So even if the AI is slightly dumber, it will compensate by aircraft performance.

Also --

Remember that theses are self-learning AIs, so they become better with time. So the next logical thing to do will be pitting this AI against a squadron of trained AF pilots, and it will ultimately learn how to defeat them.

And --

Remember that a pilotless plane will be cheaper than the piloted plane, so AI fighters can be produced in greater numbers. So if you are fighting a war, you can send up a bunch of AI planes. Some will get destroyed, but some will come back. So then you clone the returned AIs (what they have learned) into the next batch of hardware and send them out. Rinse and repeat, natural selection at work.

 

[Urwumpe]

Oh, contrary.

The pilot is the limiting factor in dogfighting, because human body cannot withstand high g loads. An AI-controlled plane will be able to fly sharper turns, so it will outmaneuver any human pilot. So even if the AI is slightly dumber, it will compensate by aircraft performance.

Still, it will be outperformed by a dumber missile....

 

[kamaz]

I still wait for an AI to even just replace the rear-seater of a classic jet properly. So far all attempts to do so resulted in the pilot being overwhelmed with workload - either for filtering the information or for configuring the computers.

That's the wrong approach.

How about building something which is cheap, small, fast, highly maneuverable and attacks the enemy using kamikaze tactics? Swarm the enemy with this at 50:1 ratio.

And if drones replace manned fighters — which they could, right now — all that wonderful cash will be out of reach, because as I keep saying here, drones have this one unforgivable flaw: They’re cheap. The MQ1 Predator, our main attack drone, costs $4 million per copy. That’s about two percent of what the F-35’s planned cost, and since you can safely double the planned cost, the MQ1 really costs about 1% of what the F-35 will cost.

One percent. For a lot of suffering USAF procurement officers, Texas congressmen, and DoD contractors out there, that’s the difference between a Maserati and a mere Porsche, a beach house in La Jolla vs. one in Oceanside, where beaches are infested with non-millionaires. It’s a real-live moral dilemma, a heart-rending tragedy, which is why they’ve triggered their tame pundits and the media outlets they place ads in to start moaning about how eeeeeevul drones are.

https://www.nsfwcorp.com/dispatch/cheap-drones/

There’s only one bit of good news here. The F-35 is a dog of a plane, yes. It’ll help to bankrupt this country’s hard-pressed taxpayers, absolutely. It’ll make a sleazy clique of contractors and Duke Cunninghams even richer, yup; but it may not make much difference in our military capabilities.

That’s because the whole notion of manned fighters, Top Gun crap, is over. If the US and China, or Russia, ever have that big war the DoD’s planners drool over, every manned aircraft will be blasted out of the sky in minutes. After that, it will be drone vs. drone, missile vs. missile. The Chinese manned fighters are way better than the F-35, as US simulations have shown conclusively, but they’ll vanish from the skies too. It will be a war fought by lumpy dweebs in recliners, flying drones and making embarrassing little video-game noises to themselves while feeling for the last Dorito in the bag, not fighter jocks in tailored uniforms.

I’ve said before that drones are already here, already capable of doing everything manned fighters do, only better.

The only reason they’re not the flagship of the US services already is the same reason the contemptible F-35 is still America’s main fighter: money. It’s a bitter thing for us guys who grew up on Jane’s and model glue and the entire Revell model catalogue, but this is all about money, and nothing to do with defense.

https://pando.com/2014/12/18/the-wa...try-has-nothing-to-do-with-defending-america/

 

[Urwumpe]

That's the wrong approach.

How about building something which is cheap, small, fast, highly maneuverable and attacks the enemy using kamikaze tactics? Swarm the enemy with this at 50:1 ratio. https://pando.com/2014/12/18/the-wa...try-has-nothing-to-do-with-defending-america/

That is something Germany and Israel develop, cheap air-to-ground drones, that simply fly into the hostile target, once identified.

Still, I don't think it is a wrong approach:

If the problem is just destroying n identified enemies with minimal effort, you are clearly right.

If the problem is finding n camouflaged enemies among m neutrals and l friendlies, you need at least one human in the loop, who is not viewing the world through a straw.

In such laboratory test scenarios as above, it is all easy, but the more you reduce the humans to biological agents, who only have time to execute orders without thinking too much, you really reduce the advantages of a human compared to the AI - you make humans bad enough to be beaten by AI.

Right now, humans are still way more flexible in their thinking and way better in fusing confusing information together than any AI can. Maybe it will change one day. But why wait for that day, when you could already help humans a lot now?

 

[MaverickSawyer]

Do you want Skynet? Because this is how you get Skynet...

 

[jedidia]

How about building something which is cheap, small, fast, highly maneuverable and attacks the enemy using kamikaze tactics?

like, you know, a missile? :lol:

 

[n122vu]

Do you want Skynet? Because this is how you get Skynet...

Exactly what I was thinking.

like, you know, a missile? :lol:

Again, exactly what I was thinking...

 

[n122vu]

Do you want Skynet? Because this is how you get Skynet...

Apparently, they do.

http://trailer-bodybuilders.com/new-market/wabco-zf-unveil-evasive-maneuver-assist-safety-technology