Concerns about which datasets are being used have been largely brushed off thus far.
Unity has caused concern and exhaustion among some game developers with its new UnityMuse and UnitySentis tools, both of which it revealed yesterday, with many professional artists and developers lancing the tools for being either impractical, unethical, or legally dubious.
The law in different countries on the use of AI programs—which often scrape the internet for art, writing, and programming without their creator’s consent—remains unsettled, with lawsuits angled at both the tools themselves and the companies using them.
In the wake of this unstable legal status, one would expect Unity to have its AI-generated ducks in a row before sinking a bunch of money into tools like this. The first question on many developers’ lips has been “What dataset did you train these on?” But Unity’s announcement answered none of these concerns.
We appreciate all the questions and interest around our training data. To create Muse Chat, we licensed third party LLMs and integrated them with first-party Unity technical documentation like manuals, release notes etc. so you can get the most relevant and up-to-date info.June 27, 2023
Unity’s mentioned that it has “licensed third-party LLMs” is the kicker for many here. A LLM, or a large language model, is a machine-learning AI that—like all of these networks—needs to train itself on datasets. In the eyes of many developers, Unity’s current response isn’t good enough.
“Hi, please disclose the datasets used by these tools, otherwise commercial usage remains incredibly dangerous legally for your customers,” Mike Bithell, a developer known for Thomas Was Alone, Solitaire Conspiracy, and John Wick Hex, wrote in a quote tweet.
Leslie Van den Broeck, a 3D artist who has worked for Riot Games, Blizzard, and Larian studios, also wrote: “[This] looks like the cheapest non committal way to approach AI tools for gamedev. It’s clearly just riding the hypetrain instead of thoroughly considering what developers could want and use that is AI powered.”
The official blog post isn’t much better. There’s not a single word dedicated to reassuring its users about where Unity has sourced its information from, just the typical promises of power and productivity. These messages might ring as hollow in the wake of massive lay-offs Unity announced in May, and I’m clearly not alone in being severely sceptical of Unity’s actions.
“We licensed third party LLMs…” Without actually disclosing what those LLMs are, this is just the new version of being able to ‘blame it on the intern’ if things go awry.Why do NONE of these companies want to fully disclose their models or datasets? I think we all know. https://t.co/2Ptvt3X5MEJune 27, 2023
I think this marks a pivotal moment for me when my perception of unity has turned from “sometimes sketch but still valuable” to “actively hostile to workers” https://t.co/I72oZsbxtvJune 27, 2023
Ya I’m completely done with Unity. :’) https://t.co/FQGpon9jqIJune 27, 2023
This is a moment literally from your presentation. Just add physics and collision for a Match 3 game…Machine Learning in tools in some pipelines can be really useful as shown in other parts of the video. But the video claim as a whole is just… ugh. https://t.co/dZvorT8B2q pic.twitter.com/VwzLfxig3mJune 27, 2023
AI does have a future in game development in some form, and it can be used intelligently and ethically. Some studios—such as the minds behind Hidden Door—may train them on an in-house creator’s work as a direct collaboration with that person or group.
Others might use these deep learning models to quickly generate concept work which they can then use as inspiration or reference. It’s clear, however, that Unity’s plunge into AI technology has struck a nerve with many of the people who actually create games.
About Post Author
