Is there anything that comes close to the CM games?

[A Canadian Cat]

1 hour ago, Rattenkrieg said:

to construct autonomous tactical models. It's a forward looking hypothesis of a direction in which wargaming AI may evolve, and something game developers should be encouraged to explore,

OK exploring new AI techniques and how they apply to war gaming is likely a good idea. The level of exploring that BFC can do is probably a bit limited. My impression is they will need to jump in a little later than the research stage.

Your earlier posts left me with the impression you thought that, now, today, a small shop like BFC could just feed stuff to a deep learning engine and create a good AI for the game. If I am wrong about that then I trust you will enlighten us. I do not believe this is even remotely possible - at this time. My impression is that there are no tools in open source or at an affordable price that a small shop like BFC can use *now* to accomplish that. Again if I am wrong I will be interested to hear.

 

4 hours ago, Rattenkrieg said:

The completely unsubstantiated discrediting of a company perfectly fits your definition of a claim that misinforms the public. If you are, indeed, a scientist then it's even more shameful.

Can we let that go, please? @BletchleyGeek made a comment (over the top) that triggered a very dramatic response from you and you have not really addressed any of his other his valid areas of concern and instead created an appeal to authority argument to defend it and now you are gaining in ad homenm attacks - cut it out - and let us keep to the topic.

@BletchleyGeek please consider not taking the bait on this and we can see if @Rattenkrieg can fill in some of the gaps and answer our questions.

Please.

[Rattenkrieg]

2 hours ago, IanL said:

Your earlier posts left me with the impression you thought that, now, today, a small shop like BFC could just feed stuff to a deep learning engine and create a good AI for the game. If I am wrong about that then I trust you will enlighten us.

I struggle to see the connection between the suggestion they use data and the inference that there is a recycling bin approach to it. I suggest they start gathering gameplay data and analysing it, because they generate an enormous amount of it (every bullet, footstep and click) and have the competitive advantage of H2H and registering what players look at when they replay WEGO over and over. That's valuable.

BFC has been around for 24 years is it? I am sure they will want to be around for another 26 at the very least. They could reach out to universities and get interns who love military, simulation and AI, or a combination of those interns, or various military schools throughout the US (assuming they are a US Inc.)

They could apply for an SBIR grant, like this startup did for computational video editing and was awarded $224,734: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1842850&HistoricalAwards=false

Here is an example of a resource that my interns are using right now to restore old videos: https://xinntao.github.io/projects/EDVR

I chose not to get bogged down in a ML debate about the exact specifics of how a wargame could begin (note: begin, not deploy) exploring the possibilities of DL and AI in general. Firstly, BletchleyGreek mocked me by suggesting I work pro bono for BFC. Secondly, he made a spurious claim about a company where friends of mine work. Neither of those actions merit any response from me. He is comfortable in his knowledge of ML and his cognitive biases and he went on the attack from the first paragraph of his initial response, at which point I decided I will not respond. I'm sorry if that disappoints you.

2 hours ago, IanL said:

My impression is they will need to jump in a little later than the research stage.

But why wait? Research is the companion of development and DL is evolving so fast that it's better to start sooner rather than later.

On 8/20/2019 at 5:49 AM, BletchleyGeek said:

Luckily, other than perhaps Russia and China I think, nobody even considers to deploy deep learning systems for target identification and acquisition. If somebody does, they're criminally insane or selling snake oil, or both.

@IanL yet another sweeping claim. Again, it betrays an embryonic understanding of DeepLearning that drastically limits my willingness to engage. Perhaps he will say he meant "deploy them right now, today." But isn't that what I am being accused of insinuating with my encouragement of BFC to look into AI? That they are insane for not having already done it?

On the subject of DL systems for target identification and acquisition: https://digital-commons.usnwc.edu/cgi/viewcontent.cgi?referer=https://www.google.com/&httpsredir=1&article=5125&context=nwc-review

Although the entire paper is very interesting, Section 3 discusses autonomous technology in targeting in more detail.

Edited August 21, 2019 by Rattenkrieg

[A Canadian Cat]

I honestly have no idea what Stave and Charles are up to in the area of AI so perhaps they will take your thoughts to heart or are already. They have certainly done a great job with what they create so far.

 

42 minutes ago, Rattenkrieg said:

betrays an embryonic understanding of DeepLearning that drastically limits my willingness to engage.

OK well if you only want to discuss AI and or Deep Learning with people who already know a lot about it or look at it the same way as you then the rest of the humans in this world will be the poorer for it.

[Rattenkrieg]

1 minute ago, IanL said:

I honestly have no idea what Stave and Charles are up to in the area of AI so perhaps they will take your thoughts to heart or are already. They have certainly done a great job with what they create so far.

 

OK well if you only want to discuss AI and or Deep Learning with people who already know a lot about it or look at it the same way as you then the rest of the humans in this world will be the poorer for it.

Ian, but I am discussing AI and DL with you. The Nvidia video and paper are incredible advances that should get game developers very excited. Imagine being able to use a GAN-based SDK from Nvidia to create a photo-realistic WW2 game based on DL from 10,000+ hours (x25fps) of combat footage from WW2 that has been restored by AI to 4K? Imagine being able to train your "pixeltruppen" using Google's (DeepMind) autonomous locomotion framework for rich environments (when it is released)?

Just because two people who work in/with AI don't like how the other behaves on a forum doesn't mean there isn't already ample material here for enrichment.

[BletchleyGeek]

I have been ignoring Rattenkrieg for a while when became apparent he was not interested in having a technical discussion. There is, in my mind,  a very clear difference between promising and delivering, and between mocking and challenging.

I sketched a very reasonable benchmark to test Rattenkrieg's assertions, and asked for a budget. Research does not happen for free, and what Rattenkrieg proposes is a research project, not a software development project. The experience of the Leela project is well documented, and the challenges of integrating modern machine learning. Maybe he thinks I am mocking him because he has an idea of the costs involved.

There is active research on applying deep learning to professional wargaming. The problems that are being found are lack of availability of computational resources to match results within reasonable time frames, issues with the asymmetry of military wargaming scenarios that do not fit well with the well behaved and neat game theoretical structures of classic boardgames, issues with representation as CNNs cannot exploit high dimensional state representationsto come up with generalised features, and more, like the problem of composability I discussed earlier. People have been trying quite hard on this for a few years now.

Contesting wild claims by a company is not "over the top". We should all do it more often. But I have no "friends" in that company either. The USARL has been working for a long time on mapping out the challenges posed by the so called the Internet of Battle Things 

https://arxiv.org/abs/1712.08980

So let's say that, considering the claims made by Rattenkrieg, and looking at the challenges, I remain skeptical that either Palantir can solve that problem,  or that they need to solve that problem in order to provide the US Army with next gen information processing and communications systems.

Confronting the CV community with proof that well known approaches to object recognition suffer from massive overfitting was probably not going to go down well, but the facts are that they have been shown to break down when inputs are trivially perturbed with noise. Hence why they are not suitable technology, by themselves, to provide automatic target acquisition and I stand by my remarks. 

They can be easily spoofed and can pretty much track anything - see the examples on the papers I linked on the first post. So much for "sweeping statements" really. There is even a new field of research, very hyped up, called "Adversarial Machine Learning" just studying this, with varying levels of success.

The locomotion suite by Google is a very useful research platform. Still, there are fundamental limitations on what kind of tasks can be learnt by neural networks without relying on specialised architectures and carefuk selection of parameters, search and optimization algorithms to guide and implement the training process. The work on Universal Planning Networks illustrates this.

The locomotion results are relevant for the video games industry, and probably there is already work published on SIGGRAPH about it. Having an animated 3d character go from adjacent nodes in a navigation mesh is a remarkably easier problem than having humanoid robots going up the stairs of a randomly chosen house in America. The videogame developer has literally godly powers to shape physics in such a way that everything runs in real time and looks goid enough.

And with this I have pretty much said my piece. I hope some of the folks here found this readable and informative. If anybody wants to know more about any of the above they can PM me or ask for further discussion.

Edited August 21, 2019 by BletchleyGeek

[A Canadian Cat]

11 minutes ago, BletchleyGeek said:

Having an animated 3d character go from adjacent nodes in a navigation mesh is a remarkably easier problem than having humanoid robots going up the stairs of a randomly chosen house in America. The videogame developer has literally godly powers to shape physics in such a way that everything runs in real time and looks goid enough.

And the video game dev does not have to deal with spoofing the AI by an adversary because they control the environment.

I guess that means that applying these techniques to creating a war game AI would actually be easier than an AI that had to deal with the real world.

[BletchleyGeek]

23 minutes ago, IanL said:

And the video game dev does not have to deal with spoofing the AI by an adversary because they control the environment.

I guess that means that applying these techniques to creating a war game AI would actually be easier than an AI that had to deal with the real world.

For creating colourful, diverse animations, yeah, that is for sure.

One of the issues in applying AlphaZero to professional wargaming is dealing with the problem of time, space and coordination of assets to, say, evaluate possible COA implementing a Strike mission. Random exploration, which is what AlphaZero does, has exponentially diminishing probabilities of landing on a particular game state via a valid set of moves. In a game with two choices, the probability of a random exploration heuristuc generating a particular sequence is very easy to calculate and vanishes to zero as the length of the sequence grows.

It worked for Alpha because games have nice terminal states where you get a very simple payoff, and you can extract easily features that correlate well with possible game outcomes. David Silver had a lot of work with handcrafted features before turning to CNNs. And still, having to play games to completion is probably the most time consuming part of the execution of the Alpha Algorithms

In other settings you need a heuristic, or base policy, to show the way. That is, old good  "AI" techniques whose practitioners call model based search and optimization...

[Sgt.Squarehead]

5 hours ago, Rattenkrieg said:

Imagine being able to train your "pixeltruppen" using Google's (DeepMind)

That's where it all starts:

PS - @Battlefront.com  If a hobo ever tells you he's "Come through time to save you!".....Run the **** away! 

Edited August 21, 2019 by Sgt.Squarehead

[Rattenkrieg]

11 hours ago, Sgt.Squarehead said:

That's where it all starts:

PS - @Battlefront.com  If a hobo ever tells you he's "Come through time to save you!".....Run the **** away! 

 

I couldn't have come up with a better coda.

Here are 2 fun little nuggets for the curious.

1) Build your own Terminator Target Acquisition HUD in Microsoft Hololens.

2) Where did that "code" on the original Terminator HUD come from?

 

std::terminate

[Rattenkrieg]

For the curious on how AI is learning to play video games: 

 

[Erwin]

Absolutely fascinating as I used to be involved in creating AI for Counterterrorism strategies, Cyberwar and other DoD sims.  Nowadays, I barely understand a word lol.    But, thanks for posting...  

[benpark]

I'd like an AI based off of this opponent. Port it over to CM-

https://kotaku.com/report-jimi-hendrix-on-acid-was-unbeatable-at-risk-1839153940

 

[weta_nz]

I'm sure Jimi would have loved Combat Mission too!