Creatures was a video game released in 1996 (or 1997, depending on the region) that contained a real, working neural network. It used the neural network to influence game play, by affecting the way the 'creatures' in the game learned to respond to various stimuli.

enter image description here

Which other video game(s) (if any) use a real neural network to influence game play?


  • This is about games that use neural networks, not the opposite (neural networks that play games, like MarI/O).

  • Neural networks that don't affect actual game play should be ignored. E.g. those used to distinguish between human players and bots in online chess, or algorithms used to detect fraudulent transactions of in-game purchases.

  • 3
    I don't see how this is exclusively recommendation, nor do I think the ID close reason is applicable (IMHO identification is about verifying (a) pre-established identity(ies)). I think this is more about the history of gaming. VTLO.
    – Joachim
    Oct 28, 2022 at 10:40
  • 1
    @Joachim: "Which game(s) satisfy <criteria>" is game identification. It's also not good for the Q&A format because there is no way to give a single correct answer. Oct 28, 2022 at 23:59
  • 2
    Agree with @BlueRaja-DannyPflughoeft. This is basically asking for a rolling list of games that meet a criteria.
    – Timmy Jim
    Oct 30, 2022 at 3:14
  • 2

2 Answers 2


Black & White

Black & White is a 2001 simulation and real-time strategy game developed by Lionhead Studios and designed by Peter Molyneux.

According to the research paper, Artificial Intelligence in Games: A look at the smarts behind Lionhead Studio’s “Black and White” and where it can and will go in the future, Black & White uses neural networks to influence gameplay (emphasis mine):

The artificial agent that makes “Black and White” such an incredible achievement in AI is the creature. As previously explained, the creature is used by the game player to do its bidding. It can also be seen as a child for the game player to both nurture and train. The chief AI developer for the game, Richard Evans, provided the gaming website www.gameai.com with some simple documentation of the game design. The desire that the developers had for the creatures was that they would be both very human-like and useful. To be human-like, the creature had to be “plausible, malleable, and lovable”. To be useful it had to be able to learn how to satisfy its master and know how to correctly act based upon its beliefs and percepts. Many recent games, such as “The Sims”, have made very human-like toy agents and many other recent games, such as “Daikatana”, have made incredibly useful agents; but no game before “Black and White” was able to combine the two elements into one seamlessly intelligent and empathetic agent.

The main aspect of the game’s AI that makes it so powerful is its mixture of different approaches of representing intelligence. Given a variety of techniques, the one that is most suitable for any certain task can be used for that task on the fly. [...]

Decision trees represent agents’ beliefs about general types of objects. Finally, neural networks of perceptrons represent desires.

There are numerous skills involved with learning and a large variety of ways in which a creature can learn. Creatures learn facts about its surroundings, how to do certain tasks, how sensitive to be to its desires, how to behave to or around certain objects, and which methods to apply in certain situations. [...]

- Wexler, J. (2002, May 7). Artificial Intelligence in Games: A look at the smarts behind Lionhead Studio’s “Black and White” and where it can and will go in the future. University of Rochester. https://www.cs.rochester.edu/~brown/242/assts/termprojs/games.pdf

Note: The author of the paper referenced an interview with Richard Evans (the developer of Black & White's AI) for the part about neural networks, but the provided link no longer points to the target website.

A YouTube video where Peter Molyneux talks about the development of Black & White's AI (relevant part starts at 07:22):


This Ars Technica article says the 2017 game Echo used machine learning to enable the game's AI to learn how to mimic the player. However, it initially made game play too difficult:

"People originally thought that we were doing a game that used some form of long-term machine learning1 in which, if you played the game for hours on end, it would eventually become you and copy exactly how you play.

We had a few early prototypes in which the game became a representation of how you interacted with it on a macro scale, but we found that became extremely punishing."

So the AI was programmed to unlearn some things to tone down the competence of the AI:

Emborg and his team opted for a system in which the AI learns and unlearns from you.

1 It's not clear whether this is a neural network or a different ML model.


Not the answer you're looking for? Browse other questions tagged .