I have a dream...

[ncc1701e]

I am mainly a solo player because I do not have time to do other things...

In my job, I am working in computer science and you have all seen the astonishing progress of Artificial Intelligence and Machine Learning over the last five years.
But, in computer wargames, I still often see complaints about the "weak" AI of some games.

However, there are good things available out there like the Keras library and the TensorFlow machine learning framework for Python but there are API for C/C++.

Why not using this to enhance our wargames? We could imagine:
1. Have several computer vs computer games been played to learn the best tactic/strategy (before publishing the game)
2. Ask the computer to learn from his human opponent to not repeat mistakes and counter him the next game (after the game is published)
3. Have all connected human players of one game populating data into the learning machine hosted by Slitherine (I already imagine the reactions about Big Brother)

That way no more human will win any wargame 2.0... [;)]
What do you think especially the guys who usually complains about AI? Are you ready for wargame 2.0? [:D]

[Kuokkanen]

ORIGINAL: ncc1701e

That way no more human will win any wargame 2.0... [;)]
What do you think especially the guys who usually complains about AI? Are you ready for wargame 2.0? [:D]

Do I want to play a game where AI is better than me...?

[berto]

ORIGINAL: ncc1701e

That way no more human will win any wargame 2.0... [;)]
What do you think especially the guys who usually complains about AI? Are you ready for wargame 2.0? [:D]

No, I am not. Why not? See here:

https://qz.com/639952/googles-ai-won-th ... -instinct/
https://singularityhub.com/2017/10/23/d ... we9b9jgw73
https://www.theverge.com/2017/10/18/164 ... elf-taught

For me, the measure of a computer wargame AI is not how "good" it is, in a competitive sense; rather whether or not it passes the Turing Test (or its wargame equivalent).

I don't want to play an AI that will give me the greatest challenge. I want to play an AI that is most plausibly human-like. Among all the generals across every battle/campaign/war in history, there were good generals, and bad generals, and everything in between. If I face off against a computer Napoleon or an Alexander in every game I play, how realistic is that? Worse, if an advanced computer AI has learned to "game the system" in bizarre ways alien to how real human beings have ever commanded, how realistic is that? Talk about destroying historical immersion!

I'll be happy to play against a better AI, sure, "better" in the sense I imagine myself playing vs. a real human being, faults and all. But play against an alien cosmic genius AI? No thanks.

[MrsWargamer]

Do I want an AI I can't beat?

Seriously, I don't want AI being made at all to be honest. But then I have read Dune which by the way was here loooong before the Matrix or Terminator.

I'd rather play the game as both sides if I have no human opponent, or not at all.

[rickier65]

ORIGINAL: ncc1701e

I am mainly a solo player because I do not have time to do other things...

In my job, I am working in computer science and you have all seen the astonishing progress of Artificial Intelligence and Machine Learning over the last five years.
But, in computer wargames, I still often see complaints about the "weak" AI of some games.

However, there are good things available out there like the Keras library and the TensorFlow machine learning framework for Python but there are API for C/C++.

Why not using this to enhance our wargames? We could imagine:
1. Have several computer vs computer games been played to learn the best tactic/strategy (before publishing the game)
2. Ask the computer to learn from his human opponent to not repeat mistakes and counter him the next game (after the game is published)
3. Have all connected human players of one game populating data into the learning machine hosted by Slitherine (I already imagine the reactions about Big Brother)

That way no more human will win any wargame 2.0... [;)]
What do you think especially the guys who usually complains about AI? Are you ready for wargame 2.0? [:D]

I've also pondered why there hasn't been some movement to incorporate some machine learning into wargame design, particularly item 2 in your list.

Though to be honest, I'm not one of those that routinely beats the AI now, so I'm not really sure I would enjoy it if it won even more.

But, there do seem to be a lot of folks complaining about 'weak' or 'poor' AI.

Interesting set of responses to your post.

Rick

[nicwb]

I am mainly a solo player because I do not have time to do other things...

In my job, I am working in computer science and you have all seen the astonishing progress of Artificial Intelligence and Machine Learning over the last five years.
But, in computer wargames, I still often see complaints about the "weak" AI of some games.

However, there are good things available out there like the Keras library and the TensorFlow machine learning framework for Python but there are API for C/C++.

Why not using this to enhance our wargames? We could imagine:
1. Have several computer vs computer games been played to learn the best tactic/strategy (before publishing the game)
2. Ask the computer to learn from his human opponent to not repeat mistakes and counter him the next game (after the game is published)
3. Have all connected human players of one game populating data into the learning machine hosted by Slitherine (I already imagine the reactions about Big Brother)

That way no more human will win any wargame 2.0...
What do you think especially the guys who usually complains about AI? Are you ready for wargame 2.0

This doesn't sound appealing at all, at least in terms of gaming. The idea in part of gaming is simply that, having something that can be won. there's no harm in having something challenging - that way there's a sense of achievement and a learning level. But having something that can't be beaten ? Where's the value to that. Given your avatar - I suspect you want all games to be a "Kobiyashi Maru" simulator - I think people would tire of that very quickly.

[demyansk]

All AI is challenging for me. I don't play the games too long to defeat AI all the time.

[JReb]

That all sounds good but;

How much would that increase the price of the game?

How much more development time will be required to develop new product, which is critical for smaller indie teams with smaller budgets?

How to prove to a certain segment of players the AI is not cheating but outsmarting them?

Would it affect turn resolution times, making them longer?

How would that impact the modding community?

A basic cost / benefit analysis would probably show that for turn based games on hex maps, there is not enough revenue generated by the genre to support that effort.

Pity though as there is a lot of room for basic improvements to AI that does not turn them into cyborg machines. I think it was HOI4 where I invaded AI-England and found the island practically undefended, everything was in N. Africa. In some of the earlier Total War games the AI would never garrison a city and so they made the garrison an organic part of the city so it wouldn't be so easy. Panzer Corps AI becomes pretty predictable and exploitable after some time even though the game is still a blast to play.

I guess my point is that maybe the AI can be improved to eliminate obvious blunders and bone headed decisions without it becoming so smart the player thinks it is cheating.

[ncc1701e]

ORIGINAL: nicwb

Given your avatar - I suspect you want all games to be a "Kobiyashi Maru" simulator - I think people would tire of that very quickly.

No, not really. And, the Kobayashi Maru test is programmed by a human to fail.

Artificial intelligence is far from perfect, but we could teach it to do what human will do. And most importantly, learned it not to fail in such a basic trap that sends all his troops into a great encirclement.

This is how I see a possible improvement.