Video games engine provider Unity announced earlier today the introduction of two new machine-learning platforms, one of which in particular has developers and artists asking questions of the company that, at time of publishing, have yet to be answered.
From Unity’s blog:
Today we’re announcing two new AI products: Unity Muse, an expansive platform for AI-driven assistance during creation, and Unity Sentis, which allows you to embed neural networks in your builds to enable previously unimaginable real-time experiences.
Muse is essentially just ChatGPT but for Unity specifically, and purports to let users ask questions about coding and resources and get instant answers. Sentis, however, is more concerning, as it “enables you to embed an AI model in the Unity Runtime for your game or application, enhancing gameplay and other functionality directly on end-user platforms.”
Because “AI” is a technology that in many cases is utterly reliant on work stolen from artists without consent or compensation, Unity’s announcement led to a lot of questions about Sentis, with particular focus on the tech’s ability to create stuff like images, models and animation. Scroll down past the announcement tweet, for example, and you’ll see a ton of variations of the same query:
just to jump on the train, which dataset y’all pull the art from???
Unity needs to be fully transparent about what ML models will be implemented, including the data they have been trained on. I don’t see any possible way ML, in current iterations, can be effective without training on countless ill gotten data.
REALLY concerning image generator stuff. What datasets?
Hi, what dataset was this trained on? Is this using artwork from artists without their permission? Animations? Materials? How was this AI trained?
You do realize that AI-created assets can’t be used commercially, so what was the rationale for adding this feature?
Which datasets were used in development of this? Did you negotiate & acquire all relevant licenses directly from copyright holders?
It’s a very specific question, one that at time of publishing Unity has yet to answer, either on Twitter or on the company’s forums (I’ve emailed the company asking the question specifically, and will update if I hear back). Those familiar with “AI”’s legal and copyright struggles can find the outline of an answer in this post by Unity employee TreyK-47, though, when he says you can’t use the tech as it exists today “for a current commercial or external project”.
Note that while there are clear dangers to jobs and the quality of games inherent in this push, those dangers are for the future; for the now, this looks (and sounds) like dogshit.