Generative AI remains one of the hottest topics in the games industry, with Google Cloud revealing last summer that 90% of surveyed developers were exploring it.
Conversations about AI often turn to its ethics: sustainability, energy, ownership, data provenance, cost, and long-term viability.
If that’s your concern about the rise of AI in games, Millennium Whisper, a retro-style dating sim from UK indie Parable Studios, is worth paying attention to.
Currently in Early Access, it’s the first game on Steam to run a large language model entirely on-device, rather than relying on cloud-based inference. It uses AI to make conversations natural and unique. As well as the technical challenges involved, there’s a story behind the ethical data capture too – as discussed in the latest edition of AI Gamechangers, our partner newsletter on Substack.
In an interview with founder and CEO Ambrose Robinson, the latest free AI Gamechangers explores how Millennium Whisper uses AI-driven characters trained on performances by real actors. Those actors retain ownership of their data and receive royalties. For Parable founder Ambrose Robinson, this was a non-negotiable principle from the outset.
“I’ve seen the worst in terms of AI,” he says. “I’ve seen people doing terrible things with it. I want to be one of the people who does it right, in a way that supports people and provides enjoyable work.”
Instead of scraping large quantities of generic internet data, Parable’s approach centres on actor-led role-play sessions. Those performances become the foundation for each character’s behaviour in-game.
“This is instead of what we usually see with AI, which is it scrapes loads of stuff from the internet and tries to mould it into something you might want,” he says. “Whereas we’re able to say, ‘Here is a specific actor who is going to play ‘Alex’. And we’re going to tell them everything about Alex, and they’re going to play as Alex for a while, and the data they give us is going to be perfect.’”
Crucially, that data remains owned by the performers. Parable licenses it for Millennium Whisper and cannot reuse it for future games or sell it to other developers.
Offline and sustainable
Running AI locally also has major positive implications for energy use, operating costs and long-term preservation. The usual server-based AI model (running your game via the Claude API, for instance, on Google Cloud or AWS servers) entails ongoing developer expenses, constant energy use, and eventual shutdown risks.
“The energy-saving side of it is the most important thing,” Robinson says. “It’s beyond any of this other stuff.”
The local inference approach has its own challenges. Hardware fragmentation makes deployment complex, and Parable has had to fine-tune its models to work across consumer hardware. But Robinson believes those trade-offs are worthwhile. From a business perspective, Robinson argues that on-device AI aligns more naturally with indie development economics.
“If you’ve got server-based AI, you’re going to have to kill it eventually, because that’s going to cost too much money,” he says. “Our game runs on your device. I really like the fact that when I sell a copy of Millennium Whisper, that’s it. That person has it on their thing. It makes it feel more like game development and not AI development.”
This conversation forms part of AI Gamechangers, a weekly newsletter exploring how AI is actually being used across the games industry, including PC development, tools and production workflows. It is intended for developers, publishers, and investors navigating rapid technical and ethical change globally today. Robinson’s interview closes out the year by tackling questions about energy use, ownership, and what it means to build AI-powered games that can last; the full Q&A (including deeper discussion of the technology, future projects, and design philosophy) is available on the AI Gamechangers Substack for free now.
You can see Robinson and other leaders in the AI space at PG Connects London on 19-20 January. Here is a code for 10% off tickets.
