Examples of single player games that use modern ML techniques in the AI?

Question

Are there any examples of single player games that use modern ML technique in its games? By this I mean AI that plays with or against the human player, and not just play the game by itself (like Atari).

"Modern ML techniques" is a vague term, but for example, Neural Networks, Reinforcement Learning, or probabilistic methods. Basically anything that goes above and beyond traditional search methods that most games use nowadays.

Ideally, the AI would be:

• widely available (i.e. not like the OpenAI Five, which was only available for a limited amount of time and requires a high amount of computational power)
• human level (not overpowered)

Ideally, the game would be:

• symmetrical (the AI has the same agent capabilities as the player, though answers similar to The Director would be very interesting as well)
• "complex environment" (more complex than, say, a board game, but a CIV5 game might work)

But any answer would be appreciated, as some of the criteria above are quite vauge.

Edit: the ideal cases listed above are not meant to discourage other answers, nor are they intended to be of strictly inclusionary (ie: any game would need to satisfy all of the above requirements)

Answer 1

There is Google Research Football, which is an open-source platform to develop reinforcement learning algorithms to play a game similar to FIFA or PES, although the football simulation is not as realistic as the current versions of FIFA or PES. You can play this game against different RL agents (e.g. DQN or IMPALA) and, of course, you can even develop your own RL agents and play against them. Here is a video that illustrates the environment. Here is the code and instructions to use it.

As far as I know, there isn't yet an AI that plays simulated football at a human-level (i.e. as good as the best human players). For example, I can regularly (although not always) beat the legendary level-AI at FIFA, but I also don't know the details about this AI (which could also be rule-based).

Answer 2

Beating the World’s Best at Super Smash Bros. Melee with Deep Reinforcement Learning

Firoiu, Whitney, Tenenbaum created a RL agent that plays and defeats professional players in Super Smash Bros Melee. The RL agent first played against the built-in AI, and then via self-play.

Only one character playing on a single stage was trained. The character picked (Captain Falcon) has no "projectile attacks" to simplify training.

Answer 3

I don't know about specific game titles but in terms of research University of Malta has a strong team working with application of machine learning to games. The key figure there used to be Georgios N. Yannakakis who published a lot of good papers and even wrote a book about content generation, smart game agents and, imho the most interesting, player modelling.

Answer 4

Not sure if this fits your requirements of the AI playing with the player, but I still wanted to mention it because to me it is the quintessential AI-based game:

AIDungeon, which is a text based story-telling game, where you can do literally anything. It's using GPT-2/GPT-3(paid) and has blown my mind several times. You've probably heard of it, but in case you haven't give it a try, only takes a couple of minutes to see what it can do.

Answer 5

There are many games where AI involved but less of them against the human player such as playing the video game. For example in this paper, they proposed a three-dimensional multiplayer first-person video game called Quake III Arena in Capture the Flag mode. Also, this paper and this paper show many games where AI involved and they show which kind of games that a human player can play against. I recommend also Awesome-Game-AI.