I’ve developed an experimental artificial intelligence system for computer games, and now I’d like to develop a game to test and help market the AI. I’ve got two ideas for a game, and I’d really like to hear your opinions about which one you think is better. The ideas are:

1/ A world War Two real-time strategy game, set on an island fought over by the Soviets and Germans. The goal of the game is to completely conquer the island. Each soldier is modelled individually, and the war takes place over the entire island. The player can either control the army as a whole, or can automate as much as they want, allowing the player to focus on only the stuff that interests them (like controlling just the tank divisions or Special Forces). So the war will continue and evolve even if the player just chooses to control a handful of units.

2/ A real-time strategy game set in and around an encircled city. The aim of the game is to either defend the city from the army outside, or play as the army outside and conquer the city. Each soldier is modelled individually, and the defenders can make all sorts of traps and devices to place in the streets between buildings (as well as having a limited number of troops). There will be a part of the game where players can make their own traps and devices by combining base components, which can then be placed at locations where they think the enemy will attack. The setting will either be present day, or perhaps just after the Second World War and using that level of technology.

I really appreciate your feedback, and thanks for taking the time to read my post.

Hello, and thank you for your replies.

I'll try and answer your questions:

Norman: My AI is designed to replicate some aspects of human behaviour, so theoretically it should be able to play any type of game (I'm a psychologist and I've applied a number of cognitive psychology theories). It just so happens that I'm a fan of RTS games, and they are the ones that I have most experience with. Also, part of the AI system is the ability for NPCs to be modelled as individual agents, which can share the decision making parts of the overall AI system. And an RTS game is a good way of using lots of individual agents.

Originally I wanted to make a kind of society simulation, where lots of agents interacted, and even built social networks. But that would be quite complicated, and I needed to make a game smaller in scope. And with military games like the ones I mentioned, the agents would have far fewer types of goals and behaviours, making it easier to make as an initial game.

Tom: I'd like to make a game for my AI, as I used C# to make the AI, and with the Unity game engine I could make a game that seamlessly interfaces with it (and I'm not sure how I could hack into other games to try and interface my AI). Also, by creating a game myself I could potentially make some money from it, which means I could further fund the development of my AI.

One important thing you might be missing here is this - If you want to test out an AI, and you test it out on a brand new game, how do you know how good that AI actually is?
This is one reason why experimental AIs tend to play known and popular games. There is some idea of what 'good' and 'bad' play is and a set of players who are considered 'good' players of the game. There is then something to rank your AI against.
If you're game you code for this is very unpopular (likely), your AI is might only play new players. Let's say if you're AI reliably defeats these players. Is it then good? Impossible to say.