AI can be used to write dialogue, quests and item descriptions. In addition, it can generate art assets, game items and other objects. AI models will also greatly simplify the procedural generation of game worlds. The limits of the use of generative AI seem to depend solely on the imagination of the developers.
While large AAA studios are able to access their own extensive databases full of assets (text, images, 3D graphics, animations) and have the know-how and personnel to develop their own generative AI solutions and use them internally, smaller studios and indie developers will be dependent on the use of commercial solutions. Especially in these cases, the question arises as to whether the use of AI-generated content can constitute a violation of applicable law.
From a German perspective, there are two questions that arise regarding the copyright-compliant use of generative AI in game development:
- Is the content generated by AI protected by copyright?
- Can the use of generative AI infringe the copyrights of third parties?
1. Copyright Protection of AI-Generated Content
The question of copyright protection for content generated by AI under German law is clearly to be rejected. Section 2 (2) of the German Copyright Act (UrhG) requires an author’s own intellectual creation for the existence of a copyrighted work. This means that neither a machine nor a computer program can be the author within the meaning of Section 8 UrhG. It is presupposed that the intellectual creation is created by a human being.
Users of content that is generated by AI and then implemented in the game as is cannot claim any copyrights in relation to this content, even if the it was created by the AI solely on the basis of the user’s input. As a result, any third party is legally able to use and monetise this content, even without acquiring a licence or the developer’s consent. Since the games industry also markets characters, quotes, places and items beyond the game itself in order to create additional sources of revenue, the lack of copyright can lead to considerable damage to overall sales.
To mitigate these risks, it is advisable to manually modify the content generated by the AI. If users manage to revise the content created by the AI in such a way that it shows the user’s own creative features, the revised content can in turn obtain copyright protection. Using generative AI in this way would still effectively relieve the burden on developers. However, this relief would not be comparable to a full automation of the work processes.
The following should be noted for the practice:
- AI-generated content that has not been manually reworked should be used primarily for „unimportant“ NPCs or game passages. Users should be aware that this content is not protected by copyright.
- AI-generated content that has been manually revised by the user can be protected by copyright if the work shows the user’s own original intellectual creation after the revision.
2. Infringement of Third-Party Copyrights
Using the results of generative AI also raises the question of whether it infringes the copyrights of third parties. This essentially depends on whether the AI used was trained with copyright-protected data and how much the results deviate from this training data. Currently, several cases are pending against ChatGPT, Midjourney and others, dealing with copyright infringements resulting from the training of AI models.
In most cases, „data mining“ is used as the basis for collecting the dataset for creating one’s own generative AI. This involves automatically analysing datasets to extract information, particularly about patterns, trends and correlations. During the creation of the dataset, copyrighted works will also be copied and thus reproduced within the meaning of Section 16 UrhG. According to Section 44b (1) UrhG, text and data mining is permitted if the content used is lawfully accessible. It is important that no restrictions of use according to Section 44b (3) UrhG are evident for the type of use as AI training data. Such restrictions can take the form, for example, of watermarks on images, in the metadata of the files or in the general terms and conditions. Should the mining of data not have been lawful, the affected data must be deleted.
Developers should be careful when using existing generative AI models. It cannot be ruled out that these models have used copyrighted materials during training in violation of the law. If this is proven, developers using these models run the risk of being liable for copyright infringement through the use of the generated content. Namely, if the generated content itself constitutes copyright infringement. This would be the case, for example, if the generated content matches a copyrighted work in its entirety or in substantial parts. Although this scenario seems very unlikely in relation to an entire work, it cannot be completely ruled out that AI-generated content reproduces parts of copyright works that may constitute a reproduction within the meaning of Section 16 UrhG. This therefore poses a potential risk for the developers.
The importance of third-party copyrights is highlighted by Valve’s current positioning. The operator of the largest gaming platform Steam has decided not to publish games with AI-generated content. In an email posted on the reddit by the user potterharry97, Valve justifies the rejection of its game by stating that the game contains art assets generated by artificial intelligence that appears to be relying on copyrighted material owned by third parties. Since the legal ownership of such AI-generated works is unclear, Valve feels unable to ship such a game until the developers can can affirmatively confirm that they own the rights to all of the IP used in the data set that trained the AI to create the assets in their game.
The following should be noted for the practice:
- When developing and training their own generative AI, developers should take special care to ensure that the data used is freely accessible and does not contain any relevant limitations of use.
- Developers who use existing generative AI models should be careful, as they may be liable for copyright infringements by the generated content. In this case, the use of plagiarism detection software is recommended. If developers want to offer their game on Steam, they should currently refrain from using AI-generated art assets.
3. Conclusion
The use of AI-generated content in gaming offers immense potential, but can pose legal challenges such as loss of copyright and copyright infringement risks. Nevertheless, the benefits can outweigh the risks if developers take appropriate measures, such as manual revision of results and careful selection of training data.
You can read more about the copyright implications of AI in our article „AI and Copyright – How Does ChatGPT Deal With Copyright?”