prologue
Artificial intelligence and its many iterations, such as neural networks and machine learning, are nothing new to video games. In the 1950s, video games used procedurally generated responses to human player choices.
But in the last few months, AI has become widely available to the public, and the progress has been remarkable. From this moment on, AI will become even more advanced.
This article discusses two forms of AI:Generation AI”, and what we call ”Conventional AIIn general, “generative AI” refers to forms of AI that create new content such as images, text, audio, and video based on user prompts. “Conventional AI” goes by many names, but for this article, it’s about performing specific tasks and making predictive or automated decisions that don’t involve creating “artistic” or expressive work. We use this term to describe artificial intelligence. . For example, in the video game industry, traditional AI can be used in many different situations. This article focuses on the complexities of using traditional AI to restrict access to certain in-game content based on player age.
Across the gaming industry, generative and conventional AI pose new and complex legal issues, many of which have no clear solutions. That said, we’ll take a look at some legal issues gaming companies should consider before launching their favorite AI tools.
Create a video game with generative AI
Current state of play
When we think about AI today, content creation is usually the top priority. What if I told you that this entire article was generated by ChatGPT?
Generative AI raises various intellectual property issues, especially regarding copyright protection and infringement. The US Congressional Research Service (CRS) has published a legal sidebar (later updated May 11, 2023) that explores the role of copyright law in applying output from generative AI.
As the CRS points out, copyright protection rests in part on the concept of “author” and the artistic manipulation of the work. In general, authorship requires that a work be “created by a human being.” The Copyright Office further clarified this point in its recent guidance, denying copyright protection to works created “without the creative contribution of a human being.”recently Zarya of the Dawn With this decision, the Copyright Office granted protection to human-generated text in AI-generated comic books, but denied protection to images created by Midjourney.
Additionally, the potential for piracy caused by generative AI inputs and outputs is an open-ended question. At the time of this writing, the Copyright Office has not yet addressed whether the use of copyrighted works in generative AI constitutes infringement. But the answer may come in the form of a court decision.
In January 2023, a group of artists filed a class action lawsuit against Stability AI, Midjourney, and DeviantArt, alleging the companies used protected artwork to train generative AI models without the artists’ permission. Getty Images similarly accused Stability AI of copying “at least 12 million of his copyrighted images” from his Getty website to train Stability’s stable diffusion model. filed a lawsuit.
Notably, these lawsuits allege that both The act of using a copyrighted work as input, and any subsequent output generated accordingly, constitutes copyright infringement.
Nemesis system example
When thinking about copyright protection and infringement, video game companies must consider how they used AI-generated content in their games and the origin of the underlying content.
Regarding copyright protection, it is key for developers to document and consider how a particular generated AI output is incorporated into their game.In-game Nemesis system Middle-earth: Shadow of Mordor is a perfect example.
The Nemesis system is a form of so-called “artificial intelligence” that generates behaviors and attributes for non-player characters based on player hierarchies and individual reactions. Simply put, the Nemesis system learned and adapted to your in-game interactions, generating unique enemy orcs that knew you and could exploit your weaknesses.
Warner Bros. first patented the Nemesis system in 2021, but the underlying code may be subject to separate copyright protection. However, given the Copyright Office’s guidance on AI-generated works, certain “outputs” of the Nemesis System may not be copyrightable if they are purely procedurally generated. .
However, the Nemesis system is integrated into a larger game that contains elements that can be protected individually and as a whole. In addition, human artistic development and manipulation of these orcs and their responses may transform the output into copyrightable works.
Let’s extend this example to piracy. Hypothetically, what if the Nemesis system used some beautiful (and copyrighted) portraits of orcs to generate output without the artist’s permission?
If the lawsuit referenced above were successful, both the training input and the resulting output could be copyright infringement. Because of this legal ambiguity, game developers cannot train models or create outputs based on copyrighted works without obtaining the proper license from the original creators. should be aware of the risks involved.
Age guarantee using traditional AI
Current state of play
Like artificial intelligence, data privacy is evolving rapidly and becoming an area of interest to both the public and regulators. While the United States has yet to pass a unified federal data privacy law, 10 states (and will continue to do so) have enacted separate and comprehensive data privacy laws with varying requirements. Regulators also pay particular attention to children’s privacy and harmful online content.
In early June, the FTC announced a $20 million settlement with Microsoft for alleged violations of the Children’s Online Privacy Protection Act (COPPA) through Microsoft’s Xbox Live service. This is the third action the FTC has taken in less than a month to focus on children’s privacy, following the Edmodo-Amazon settlement. Specifically, the Edmodo settlement requires the FTC to remove “affected work products.” model again algorithm Developed using information illegally collected from children.
At the state level, California has passed a new data privacy law, the Age-Appropriate Design Act (AADC), which will take effect on July 1, 2024. In a blog post earlier this year, we explored how AADC could impact the video game industry. . In response to these growing concerns, some companies have started deploying artificial intelligence tools to improve their data practices.
One of the intersections of data privacy and traditional AI isautomated decision making(ADM) is the process of using algorithms and machines to make decisions with little or no direct human involvement. Many states have already enacted ADM laws, such as California’s Enhanced Online Transparency Act (BOT), which prohibits the use of automated bots when trying to incentivize or influence consumers. companies are required to disclose
There are various pitfalls gaming companies should consider before adopting artificial intelligence for their data operations. However, in this article, we will discuss age guarantee methods, a growing concern common to gaming companies.
Age Guarantee Considerations
Under the California AADC, game companies must begin estimating a player’s age within a reasonable level of certainty. Once age is determined, game companies use that information to either (1) prevent children from accessing potentially harmful in-game content (through so-called “age guarantees”) or (2) We need to design games for Friendly.
The California AADC was modeled closely following the British age-appropriate design code called the “Child Code”. In early 2023, the UK Information Commissioner’s Office issued guidance specifically for video game designers. This could reveal how California regulators will enforce his AADC. Out of this guidance, gaming companies implementing his AI-powered ADM should carefully consider how to estimate age with a sufficient level of certainty without violating data protection laws.
Gaming companies may choose to implement traditional AI to automate age-verification methods, such as determining a player’s age through verifiable government IDs or face scans. Part of the risk of age-gating with traditional AI is the proper handling of children’s data to mitigate the potential bias of such AI tools.
For example, the FTC and Microsoft orders require Microsoft to separately process children’s information collected prior to parental consent, subject to a separate two-week deletion period. But this FTC decision is about COPPA, no California AADC. AADC does not assume verifiable parental consent. Children’s data collected for age assurance purposes should therefore be treated with particular care, as it cannot be used for any other purpose without parental consent. Rather, gaming companies should set the age of their players while complying with data protection laws.
When game companies deploy traditional AI for age assurance, the child’s data cannot be used for anything other than verifying the child’s age. Developing, training, or improving AI can be a prohibited use beyond what is permitted. Any other information about child players collected by traditional AI should be kept separate and should not be used to improve AI tools.
Additionally, game companies may wish to estimate a player’s age through facial scanning. Without getting into the intricacies of the Biometric Health Data Act (Illinois and Washington State folks), age assurance that relies on AI-powered facial recognition poses a variety of risks, including inaccurate results.a A recent study found that AI technology suffers from exaggerated biases when trying to estimate a person’s age.
This goes back to the issue above regarding inappropriate use of children’s data.Game companies can’t scan children’s faces both It is intended for age-guaranteed purposes and to improve the accuracy of conventional AI. Rather, children’s data collected for inference purposes that’s all used for estimation. Gaming companies must be prepared to address and mitigate those risks if traditional AI fails to perform well with automatic age guarantees.
Gaming companies must consider the vast and complex risks before using traditional AI for age-guaranteed mechanisms. For future AI regulation, gaming companies should be prepared to provide regulators with documentation of how traditional AI was trained and how it made decisions. In particular, gaming companies must be able to prove that their AI-powered age-verification methods are not using children’s data inappropriately.
Conclusion
This article is by no means an exhaustive list of issues people in the gaming industry may face when using artificial intelligence. Also, this article is not intended to discourage or encourage the use of AI. Rather, AI is undoubtedly a powerful tool that will be leveraged for years to come, and the video game industry embraced AI long before it hit the public eye.
As gaming companies continue to implement generative and conventional AI, they must recognize the complex web of laws and regulations governing artificial intelligence.
