Sponsored By

Read More from GDC 2024 | Keep up with the latest game industry event coverage from GDC 2024, including news, talks, interviews, and more from the Game Developer team.

Unity says it shared AI training data with devs in bid for transparency

Unity wants to prove its commitment to transparency by sharing AI training data with developers concerned about copyright violation.

Bryant Francis, Senior Editor

April 2, 2024

5 Min Read
A headshot of Unity exec Marc Whitten.
Image via Unity.

At a Glance

  • After two years of bad press and withering critiques, Unity needs to prove it's rebuilding trust with developers.
  • Unity exec Marc Whitten says the company shared AI training data with devs to rebuild that trust.
  • Easing fears over generative AI's questionable copyright foundations could warm developers up on Unity's AI tools.

Unity's adoption of generative AI tools couldn't have come at a less opportune time for the company. In the midst of mass layoffs, a controversy over its rolled-back Runtime Fee fracas that threatened to nickle-and-dime developers over game installations, championing the controversial technology may have wooed investors but gave developers another reason to feel the company was prioritizing its stock price over customers.

Why? In case you're new to this party, generative AI technology has barreled onto the game development scene with the same fervent hype, proclamations of job elimination, and copyright-skirting that it has everywhere else.

A core issue with generative AI tools is that many of them are trained on datasets containing text, photos, videos, and other assets that the toolmakers may not have the copyright to. American courts are fielding many lawsuits on this topic.

Unity immediately faceplanted into those copyright-violating concerns when it opened its platform for AI toolmakers. 24 hours after launching, it yanked AI tool Atlas off its store after users noticed the developer may have stolen 3D assets.

In November, the company posted a blog explaining some of its practices around AI training data, breaking down some of its best practices to ensure a clear chain of ownership for the data it's drawing on.

According to Unity chief product and technology officer Marc Whitten, the path to rebuilding trust in Unity may begin with its AI tools. Or more specifically—breaking down the walls of obscurity and letting a limited pool of developers pour over the training data.

Unity is finally "having conversations" with developers

If developers are still on guard about Whitten's promises, that's for good reason. In 2022 he said Unity hadn't been "talking enough" to developers, and in 2023 with the Runtime Fee uproar, he admitted the company wasn't "listening enough" to developers.

In our interview with Whitten and Unity interim CEO Jim Whitehurst at GDC 2024, we asked if the company had finally figured out a good communication strategy with its customers. Whitten explained that the goal now was to have better "conversations" with its audience.

"The more we get into a conversation with them that's consistent, the better off we are," he said. "Frankly, if you were to [ask] what's the best communication we do, it's honestly engineer-to-engineer."

"When we do an AMA ("Ask Me Anything" session) with the dev team, and they're talking to people that are using those features, they're great."

He referenced an upcoming roundtable he and Whitehurst were about to have with "a few dozen" chief technology officers of companies relying on Unity. But CTOs and everyday developers may have very different concerns—did Whitten have anything more concrete to tell our readers about?

He said he did. "Right before Unite 2023, we were launching our AI tools Unity Muse and Unity Sentris," he began, acknowledging that though developer interest in generative AI is speeding up, there are "a lot of issues" with the technology.

"People are worried about where the data came from, what does it mean for their jobs—there are real issues associated it along with the productivity increases."

This was a candor about AI that we did not hear from Whitten during the launch of Muse and Sentris, where he was more keen to tout the benefits. To win trust from core "influencers" the company communicates with, he said he went to the AI teams and pressured them to share training data with that group.

"Originally the team was like 'this is the secret sauce...because we're training these models in a different way to make sure we can really stand by the fact there's not copyrighted work in them and that you can use it in games,'" he said. But Whitten said every time he spoke with developers, they expressed a burning need to understand that data themselves.

"We came back and said 'we've just got to be fundamentally transparent about it,'" he said, adding that the launch of the tools "fundamentally changed" after that process.

According to him, those conversations made the AI products "better" and helped Unity prioritize its next products. "They made our go-to-market much, much better."

Unity can explain where its data came from, OpenAI can't

As journalist Ed Zitron noted in a piece published right at the start of GDC 2024, executives at companies like OpenAI are either reluctant or unable to explain the origins of their training data fueling their technology. It's to Unity's credit then that it is able to dump that information out to developers who want to view it.

Whitehurst chimed in on this part of the conversation, saying that at Unite 2023 he and other company higher-ups sat in rooms with 25 developers at a time who would "pepper them" with questions hour after hour. (He added that he's not sure they will do this "every year" at Unite, but praised the process for 2023).

These examples were heartening to hear (prying clear examples from any executive is sometimes like pulling teeth), but just as with Unity's roadmap toward profitability, it carried baggage borne from the mistakes that preceded them. In theory, Unity had this process in place already, with developer advocates going back and forth and shuttling this kind of feedback up the chain at the company to improve the tools and make a better product.

Developers did have the chance to weigh in before decisions like implementing the Runtime Fee and updates to the Unity Pro license. But the company plowed ahead anyway.

Whitehurst and Whitten's hands-on approach to gathering feedback at a key moment for Unity is admirable, but developers who can't sit down for those chats are likely looking for proof that raising their complaints through normal chains of communication will produce results.

Releasing that training data to influencers is indeed a strong step for transparency. But restoring agency downstream to advocates and engineers who can spot these problems from a mile away feels like a key move Unity will need to take through the rest of 2024.

Correction: A prior version of this story referred to Unity's "community" without specifying it was the "influencer community." It has been updated for clarity.

Game Developer and Game Developers Conference are sibling organizations under Informa Tech.

Read more about:

GDC 2024

About the Author

Bryant Francis

Senior Editor, GameDeveloper.com

Bryant Francis is a writer, journalist, and narrative designer based in Boston, MA. He currently writes for Game Developer, a leading B2B publication for the video game industry. His credits include Proxy Studios' upcoming 4X strategy game Zephon and Amplitude Studio's 2017 game Endless Space 2.

Daily news, dev blogs, and stories from Game Developer straight to your inbox

You May Also Like