Are advances in AI like IBM's Watson and Google's Deep Mind likely to filter through to IL2 and gaming in general?

[FlyingNutcase]

hi folks,

 

AI seems to be coming of age in the real world. Are advances in AI like IBM's Watson and Google's Deep Mind likely to filter through to IL2 and gaming in general? There seems to be a world of possibilities.

 

 

 

 

[II/JG17_HerrMurf]

http://www.theverge.com/2016/6/30/12067648/ai-pilot-defeats-human-combat-instructor-in-simulated-air-battles

 

My biggest takeaway was how little computing power was required for the program.

[69th_chuter]

It looks like DirectX 12 is our best hope.

 

http://www.extremetech.com/gaming/219358-why-ai-still-sucks-in-video-games-and-how-directx-12-could-change-it-for-the-better

[JG13_opcode]

http://www.theverge.com/2016/6/30/12067648/ai-pilot-defeats-human-combat-instructor-in-simulated-air-battles

 

My biggest takeaway was how little computing power was required for the program.

Bear in mind it didn't have to do any 3D rendering. It'll be years yet until we see AI of that calibre in a consumer flight sim. Partly because it's no doubt classified.

 

edit: hmm, there's a bug in the forum software. quoting a user with [] in their name kills the bbcode parser.

Edited July 21, 2016 by 13GIAP_opcode

[II/JG17_HerrMurf]

Why wouldn't it need to do any 3D rendering? It was a military simulator against a live instructor pilot. I'm not sure how it would be significantly different.

 

The first article I read said the AI used about the same computing power as an average smart phone.

Edited July 21, 2016 by [LBS]HerrMurf

[JG13_opcode]

HerrMurf" post="371990" timestamp="1469126577"]

 

Why wouldn't it need to do any 3D rendering? It was a military simulator.

 

The AI didn't, I'm saying. The AI just has to produce control outputs based on sensor inputs. It's not looking at a picture the way we look at our monitors, and the PC running the AI is likely not the same PC that the human is flying on.

Although apparently they can run the AI on a raspberry pi, so it's not as hefty as I initially guessed.

 

Still, we won't see this for years.

[216th_Lucas_From_Hell]

The specifics of it are foggy, though the developments sure are interesting

 

As opcode said one problem is the 3D rendering, and another one we have is that the AI needed in a game needs to be very dynamic (i.e. not perfect like one used to train pilots), and it also needs to interact with the environment and as part of it. On top of that the level of flight modelling involved is not known there, both for the real pilot and the AI. Many military-grade sims use relatively basic flight models and instead focus on systems modelling and specific handling of combat and non-combat situations, which is different from the experience expected from a customer-grade flight simulator.

 

It's one thing to simulate a strong AI in a 1 x 1 engagement, and even a lousy computer like mine can handle half a dozen 'ace' AI in a dogfight scenario. But for it to represent progress you need to have thirty bombers with individual AI for each crewmember doing a ground attack and accounting for the AAA and fighter opposition when doing their attack run, executing it realistically with a degree of human error, without a significant performance hit.

[seafireliv]

I am waiting for the day when AI can truly make sims immersive with pilots (friend and enemy) who actually act like the real ones would. This would be so good for so many sims and games.

 

The sad part is, if they can make such a smart AI, it will likley be used for bad things in the real world. Trying to buy something and you have to argue with an AI staff memeber? AI weapons left to decide whom to kill? So it`s a bit of a two edge sword.

[HippyDruid]

It's a very interesting field with many differents avenues of application.

 

Microsoft have a Minecraft AI learning mod called Project Malmo which is worth looking at if you're keen on coding your own AI. 

https://www.microsoft.com/en-us/research/project/project-malmo/

 

It's interesting that they are using a virtual world such as Minecraft to assist in developing the AI.

 

I also find it interesting that people assume freethinking AI would have all the malevolent traits we humans have.

Edited July 21, 2016 by HippyDruid

[JimTM]

...Trying to buy something and you have to argue with an AI staff memeber?...

 

Ah, but you won't be doing the buying, your AI assistant will. While you may not be able to drive a hard bargain, your AI sidekick will be self-driving. 

Edited July 22, 2016 by JimTM

[Rjel]

While I think having more human acting AI would be and someday probably will be great, I'm not sure having AI so good that it beats up on a professional fighter pilot. We might think we're all that and then some when it comes to our prowess flying sims. The reality would likely be very different with the AI smoking us at the end of every maneuver. On the other hand, if AI got to be so good, it might force everyone to fly online just to find some manageable opponents!

[ShamrockOneFive]

While I think having more human acting AI would be and someday probably will be great, I'm not sure having AI so good that it beats up on a professional fighter pilot. We might think we're all that and then some when it comes to our prowess flying sims. The reality would likely be very different with the AI smoking us at the end of every maneuver. On the other hand, if AI got to be so good, it might force everyone to fly online just to find some manageable opponents!

 

Here's the thing. We want a sophisticated AI but with a simulator we don't want an AI to be the best that an AI can be at flying a realistic simulator. What we want it to mimic the relative skill levels of humans. We want it to be programmed to panic, to make mistakes, to not have perfect aim, and be forced to scan the sky the way a human would.

 

These new kinds of AI technologies may let us do that although they haven't been explicitly designed to. Programming/generating human responses may be the most difficult thing.

[216th_Lucas_From_Hell]

I think Il-2 is on the right track in some areas. The AI definitely can panic and stall out when pressured, and they can jack up a landing or two if damaged and sometimes at random. Ground attack often has overly optimistic AI pilots going for a run then kissing a tree or slamming against the open fields, and bombers can freak out and break formation, dropping all bombs and doing a rough nosedive.

 

It's the fine-tuning that's hard - which aspects of the AI need to be tuned to get a certain response under specific situations while not affecting most situations negatively.

 

Not a flight sim at all but I was playing FIFA 12 on a PlayStation this week and at higher difficulties the AI got unplayable. However, through a set of sliders you could increase the probability of passing, crossing and shooting error, and also reduce the accuracy of man-marking. As a result with some slight alterations the game suddenly felt very real, and elaborate olay was possible while remaining hard and realistic, making you cherish that 1 x 0 with a goal on the 87th minute.

[Rjel]

Here's the thing. We want a sophisticated AI but with a simulator we don't want an AI to be the best that an AI can be at flying a realistic simulator. What we want it to mimic the relative skill levels of humans. We want it to be programmed to panic, to make mistakes, to not have perfect aim, and be forced to scan the sky the way a human would.

 

These new kinds of AI technologies may let us do that although they haven't been explicitly designed to. Programming/generating human responses may be the most difficult thing.

Yeah. I sorta got that.

 

I've mentioned in the past I'd like to see AI in sims evolve to the point that it be specific to a given A/C and use tactics that a Japanese pilot might use in Zero instead of it using the same tactics and maneuvers it would use regardless of whatever plane it got plunked into. One plane, one AI different than the other AI in a different type of plane. Someday maybe.

[FlyingNutcase]

Another immersive side of combat AI would be actually talking/asking/giving orders/engaging in banter with AI wingman or other friendly pilots, ground control, even fellow crew on the same plane, in a way that is natural.

 

Add that to realistic gameplay by all AI pilots & crew and VR and the single player or co-op environment becomes much more realistic than the typical unrealistic behavior of unmanaged multi-player. Food for thought! Perhaps another 10 years to get there? About the same maturity timeframe as sex robots. But I digress.

[Scarecrow]

 

 

 About the same maturity timeframe as sex robots. But I digress.

 

Built one already, FleshLight taped to the washing machine. But now I'm banned from Twitch

[FlyingNutcase]

Built one already, FleshLight taped to the washing machine. But now I'm banned from Twitch

 

ROFL; oh the imagery of that.

 

But pretty exciting times; so many developments in hardware and software going on in all theaters of life now. Big changes ahead.

[seafireliv]

I`ve been wanting better AI in games and sims for nigh 20 odd years.

 

I doubt we`ll see it in another 20 years.

 

While graphics and processing power has increased amazingly, AI is still laughably bad. It`s even more noticable with the almost life-like graphics we have now. I can`t figure out if it`s for lack of ability or due to consumerism - `it isn`t worth the effort to make a brilliant AI`, as I`ve heard some say. I`m leaning to lack of ability- It`s simply very hard to do. Who wouldn`t want to make an almost sentient AI if they could- Imagine the cash such an invention would make. Although there should be more to it than just money.

 

Well I hope someone makes the effort sooner than 20 years!

Edited July 23, 2016 by seafireliv

[FlyingNutcase]

I`ve been wanting better AI in games and sims for nigh 20 odd years.

 

I doubt we`ll see it in another 20 years.

 

While graphics and processing power has increased amazingly, AI is still laughably bad. It`s even more noticable with the almost life-like graphics we have now. I can`t figure out if it`s for lack of ability or due to consumerism - `it isn`t worth the effort to make a brilliant AI`, as I`ve heard some say. I`m leaning to lack of ability- It`s simply very hard to do. Who wouldn`t want to make an almost sentient AI if they could- Imagine the cash such an invention would make. Although there should be more to it than just money.

 

Well I hope someone makes the effort sooner than 20 years!

 

I'm seeing self-learning AI frameworks becoming available that will be to AI development what a website framework like Bootstrap is to web development, enabling some pretty wicked results without needing to develop the core engine; it just needing samples to learn from.

 

And big companies are taking on AI with a passion now - progress is likely to be similar to the progress in hardware and graphics over the previous 20 years, just twice as fast due to the world being so connected and suitable hardware being available due to, well the previous 20 years. :-D. 

 

My 2c.

[wtornado]

Just use the old ''ACE'' setting in 1946 you will have pilots swearing and getting shot down often.Hehehehe.