Calling SKYNet...

Forum adverts like this one are shown to any user who is not logged in. Join us by filling out a tiny 3 field form and you will get your own, free, dakka user account which gives a good range of benefits to you:

No adverts like this in the forums anymore.
Times and dates in your local timezone.
Full tracking of what you have read so you can skip to your first unread post, easily see what has changed since you last logged in, and easily see what is new at a glance.
Email notifications for threads you want to watch closely.
Being a part of the oldest wargaming community on the net.

If you are already a member then feel free to login now.

**Easy E**

http://www.popsci.com/ai-pilot-beats-air-combat-expert-in-dogfight

Pilot loses to AI in A2A combat sim.

n the military world, fighter pilots have long been described as the best of the best. As Tom Wolfe famously wrote, only those with the "right stuff" can handle the job. Now, it seems, the right stuff may no longer be the sole purview of human pilots.

A pilot A.I. developed by a doctoral graduate from the University of Cincinnati has shown that it can not only beat other A.I.s, but also a professional fighter pilot with decades of experience. In a series of flight combat simulations, the A.I. successfully evaded retired U.S. Air Force Colonel Gene "Geno" Lee, and shot him down every time. In a statement, Lee called it "the most aggressive, responsive, dynamic and credible A.I. I've seen to date."

And "Geno" is no slouch. He's a former Air Force Battle Manager and adversary tactics instructor. He's controlled or flown in thousands of air-to-air intercepts as mission commander or pilot. In short, the guy knows what he's doing. Plus he's been fighting A.I. opponents in flight simulators for decades.

But he says this one is different. "I was surprised at how aware and reactive it was. It seemed to be aware of my intentions and reacting instantly to my changes in flight and my missile deployment. It knew how to defeat the shot I was taking. It moved instantly between defensive and offensive actions as needed."

The A.I., dubbed ALPHA, was developed by Psibernetix, a company founded by University of Cincinnati doctoral graduate Nick Ernest, in collaboration with the Air Force Research Laboratory. According to the developers, ALPHA was specifically designed for research purposes in simulated air-combat missions.

The secret to ALPHA's superhuman flying skills is a decision-making system called a genetic fuzzy tree, a subtype of fuzzy logic algorithms. The system approaches complex problems much like a human would, says Ernest, breaking the larger task into smaller subtasks, which include high-level tactics, firing, evasion, and defensiveness. By considering only the most relevant variables, it can make complex decisions with extreme speed. As a result, the A.I. can calculate the best maneuvers in a complex, dynamic environment, over 250 times faster than its human opponent can blink.

After hour-long combat missions against ALPHA, Lee says,"I go home feeling washed out. I'm tired, drained and mentally exhausted. This may be artificial intelligence, but it represents a real challenge."
The results of the dogfight simulations are published in the Journal of Defense Management.

We are all doomed!

**TheCustomLime**

So a guy lost at a video game against an AI on hard mode? How is this a sign of judgement day?

**Jehan-reznor**

He should have done a Captain James T. Kirk on the AI

**SagesStone**

TheCustomLime wrote:
So a guy lost at a video game against an AI on hard mode? How is this a sign of judgement day?

Imagine if all those kids playing CoD suddenly got up and went outside to rage because the game stopped holding their hand. Like some sort of hilarious version of 28 days later most likely, but still.

**Alex Kolodotschko**

That would be a fun April Fool.

**Easy E**

TheCustomLime wrote:
So a guy lost at a video game against an AI on hard mode? How is this a sign of judgement day?

I think the big deal is that he consistently beat it many times before, and is having trouble with this new version..... which is also built on a cheap and easy to find processor.

**jmurph**

Of course porting this to a real world application is more than trivial. Real flying crafts have to deal with countless variables that a simulator (by definition) does not. That being said, large portions of commercial flight are already automated, so it is not unthinkable that there could be greater automation in combat operations. However, considering that even drones are largely manually controlled, I don't know that it will be anytime soon that we start putting up automated fighters.

**Iron_Captain**

jmurph wrote:
Of course porting this to a real world application is more than trivial. Real flying crafts have to deal with countless variables that a simulator (by definition) does not. That being said, large portions of commercial flight are already automated, so it is not unthinkable that there could be greater automation in combat operations. However, considering that even drones are largely manually controlled, I don't know that it will be anytime soon that we start putting up automated fighters.

And let's hope no one ever does that. Combat should only ever be done by humans. Only humans can make ethical decisions, and only humans can be held accountable when something goes wrong.

**Alex Kolodotschko**

Iron_Captain wrote:
Combat should only ever be done by humans..

Unless it's against other robots. I would love to tune in to Red Bull's "Rock 'em Sock 'em Robotapocolypse".

**OgreChubbs**

If I learnt anything about computers skynet will crash and be out of date in 3 years, wait it out.

**Monkey Tamer**

In a real world scenario the AI wouldn't have the benefit of instantly knowing the input the human pilot just put into his aircraft. This is like playing Street Fighter 4 on very hard. The machine has a built in edge to make it hard. Human opponents are harder due to their unpredictability.

**TheCustomLime**

Easy E wrote:
TheCustomLime wrote:
So a guy lost at a video game against an AI on hard mode? How is this a sign of judgement day?

I think the big deal is that he consistently beat it many times before, and is having trouble with this new version..... which is also built on a cheap and easy to find processor.

So? It's an AI in a video game. Are we going to report that RL soldiers can't beat COD on Veteran?

**redleger**

TheCustomLime wrote:
Easy E wrote:
TheCustomLime wrote:
So a guy lost at a video game against an AI on hard mode? How is this a sign of judgement day?

I think the big deal is that he consistently beat it many times before, and is having trouble with this new version..... which is also built on a cheap and easy to find processor.

So? It's an AI in a video game. Are we going to report that RL soldiers can't beat COD on Veteran?

I certainly can't. Im a much better marksman in real life. I suck at FPSs

**NuggzTheNinja**

Monkey Tamer wrote:
In a real world scenario the AI wouldn't have the benefit of instantly knowing the input the human pilot just put into his aircraft. This is like playing Street Fighter 4 on very hard. The machine has a built in edge to make it hard. Human opponents are harder due to their unpredictability.

That's a sensor problem, not an AI problem. Fuzzy logic makes probabilistic judgments about human "unpredictability" based upon human predictability.

**Peregrine**

Iron_Captain wrote:
And let's hope no one ever does that. Combat should only ever be done by humans. Only humans can make ethical decisions, and only humans can be held accountable when something goes wrong.

So we should only have hand to hand combat between unarmed soldiers? Like it or not we already have computers making decisions in combat. When a pilot fires an air-to-air missile at another plane they don't guide the missile by hand, they trust the missile to do its job and autonomously engage the target. Having a computer flying the plane (possibly with a human supervisor making attack/no-attack decisions before it engages) is just the inevitable next step in the process.

Automatically Appended Next Post:

NuggzTheNinja wrote:
That's a sensor problem, not an AI problem. Fuzzy logic makes probabilistic judgments about human "unpredictability" based upon human predictability.

Except it's a sensor problem that will realistically never be solved in the foreseeable future. An AI pilot isn't going to be linked in to the enemy pilot's control inputs and starting to react as soon as the enemy pilot makes a move, it's going to have to wait for the enemy plane to start its maneuver and predict what that maneuver is going to be. It's much harder to beat "human unpredictability" when you don't have the benefit of massively improved reaction times powered by unrealistic data access. Slow down the AI's data access and let's see how good it is.

**jmurph**

That's exactly the implementation barrier. In a sim, the AI has perfect access to information. In RL, information must be gained from a variety of sources (optics, radar, etc.), condensed into a interpretable form and then used for analytics and selection of action. Humans have highly complex biological systems to do this and use automation to facilitate parts of the process. So far, it has proven extremely difficult to give "senses" to a machine AI. Most successful attempts rely on a relatively stable scenario. Self driving cars is perhaps the most ambitious attempt to automate reactions, and it is struggling. Adding the complexity of flight AND combat is way beyond current technology and design.

As to the "morality" of automated warfare, we need only look at how the machine gun changed warfare. If the technology exists and provides advantage, it will be implemented. Human warfare is very much adapt or die with ethical concerns generally taking a backseat.

**NuggzTheNinja**

Peregrine wrote:

Except it's a sensor problem that will realistically never be solved in the foreseeable future. An AI pilot isn't going to be linked in to the enemy pilot's control inputs and starting to react as soon as the enemy pilot makes a move, it's going to have to wait for the enemy plane to start its maneuver and predict what that maneuver is going to be. It's much harder to beat "human unpredictability" when you don't have the benefit of massively improved reaction times powered by unrealistic data access. Slow down the AI's data access and let's see how good it is.

jmurph wrote:
That's exactly the implementation barrier. In a sim, the AI has perfect access to information.

A great deal of effort in autonomous systems systems is focused on decision making given uncertain information. Do you have reason to believe that the designers failed to account for information lag and uncertainty? Without any evidence that the designers have completely ignored a massive body of literature discussing the importance of addressing this issue, your criticism is based upon an assumption.

Their product description actually suggests quite the opposite - they mention that ALPHA is handicapped by not having access to AWACS information, indicating that it has incomplete access to information.

**jmurph**

I do not believe the designers failed, rather it is the inherent nature of such simulators to be very limited. In a sim, there are far fewer variables and areas of uncertainty. That's how sims work, By perfect access I mean their is no delay between what the human operator does and the AI getting that info. In a real world setting, their would be all kinds of potential uncertainty from sensors, which all, in tern, only provide a limited spectrum of information and have delays. Even vision is not instantaneous. And then it has to interface with mechanical systems that have their own issues. And then respond to that, etc.

This is definitely an improvement, but still well off of real world implementation. Again, I would point you towards automated driving to see how complex this is. Once we get that down, I think we will be considerably closer to automated fighter interactions.

**A Town Called Malus**

jmurph wrote:
I do not believe the designers failed, rather it is the inherent nature of such simulators to be very limited. In a sim, there are far fewer variables and areas of uncertainty. That's how sims work, By perfect access I mean their is no delay between what the human operator does and the AI getting that info. In a real world setting, their would be all kinds of potential uncertainty from sensors, which all, in tern, only provide a limited spectrum of information and have delays. Even vision is not instantaneous. And then it has to interface with mechanical systems that have their own issues. And then respond to that, etc.

This is definitely an improvement, but still well off of real world implementation. Again, I would point you towards automated driving to see how complex this is. Once we get that down, I think we will be considerably closer to automated fighter interactions.

Again, you are assuming that they haven't built in those kinds of uncertainties to provide a more realistic test of their AI.