Google PaLM 2 and AI will change gaming, for better or worse

Announced at Google I/O, PaLM 2 is one of the latest models joining the fast-growing artificial intelligence trend. However – there’s one crucial aspect that sets this one apart, and it’s going to change gaming forever.

The successor to the original PaLM large language model has four sub-models: Gecko, Otter, Bison, and Unicorn. The smallest model – Gecko – is targeting portable and handheld devices, which will work on-device and even offline. While OpenAI’s language model is being implemented on a mass industrial scale (take Microsoft and IBM, for example), the small-scale PaLM 2’s Gecko model is going to turn this dynamic on its head.

Already featuring across the Google Workspace in Docs and Gmail, the API is openly accessible to developers and tinkerers alike. However, we’re not sure whether to be excited or worried that the gaming industry could be the next group to adopt it.

The games industry is already adopting AI

In late March 2023, Ubisoft announced Ghostwriter, an AI writing tool designed to assist game-writers with barks (the filler phrases and murmurs NPCs espouse during cutscenes and alike.) This Ubisoft PR report reiterates that the AI service “isn’t replacing the video game writer” but will instead assist with some of the more menial tasks.

Public reactions to newly introduced AI automation software such as Ghostwriter have been mixed. The risks of jobs being replaced and creativity being distilled into a single, homogenous line of code are two of the biggest modes of opposition to artificial intelligence – though Ghostwriter itself doesn’t seem too harmful – in its current state, at least.

Role-playing games are limited in their scale by the amount of (quality) dialogue and script-writing the team of creatives can muster. This year we saw a Fallout 4 AI dialogue mod released that adds 300 new lines to the game. This mod undoubtedly took a long time to create and develop, and there was surely a sense of curation and editing that the modders oversaw to ensure we didn’t see some of AI’s notorious and nonsensical sentences end up in the wasteland. However, Google PaLM 2 could eliminate this need for human touch entirely.

Why PaLM 2 could be the AI that changes gaming forever, or ruins it

Google’s PaLM 2 tech report publishes some fairly promising results for the performance of the smallest sub-model, and while Gecko won’t really step on the feet of what ChatGPT is used for, it could look to offer new solutions instead.

Being able to run from a local device, even in an offline state, Gecko is going to be capable of near-instantaneously generating dialogue and in-game pathways for games ranging from mobile devices, gaming handhelds, and PCs, without having to refer to a distant server somewhere across the world. For reference, if a game wanted to make use of AI generated content, it would need to connect to a distant language model and wait for a response. However, seamless integration of unique content generated on-device could be the next biggest development in gaming since 3D graphics – and PaLM 2 could be the first readily available language model to bring this.

Of course, the quality of content generated by Gecko will be limited. Content generated offline on an Android device won’t compare to content parsed through thousands of graphics cards, but it will lay solid foundation for local AI generation.

For iterative Gacha games and dungeon crawlers – this could be fantastic. Imagine endlessly generated maps, characters, enemies, and weapons, alongside matching dialogue that accurately compliments it. A whole new genre of infinite gaming might awaken, where each world and narrative is woven according to what the player wants – but is that a cause for concern rather than optimism?

As seen by this attempt to enjoy a DnD game run by a ChatGPT Dungeon Master, the results are not particularly promising. The chatbot failed to grasp the creative spark that a DM needs, though it “felt like watching the same magic trick over and over again.”

If AI generated gaming really takes off – we’d imagine a solid time period of mediocre, un-convincing releases featuring soulless, derivative content. Google Bard, for example, had a critically poor launch, after consistently publishing plagiarised content while also contradicting itself as much as possible. AI generated games will see equally difficult launches. AI models need iterative training, and while PaLM 2 is promising, there’s still a sense that we haven’t reached the point where it’s ready to go. OpenAI’s Jason Wei, for example, thinks that AI generated game narratives alongside “[compelling virtual worlds] designed by language models themselves” are less than two years away, though with the current explosive rate of development there’s no saying how much shorter this time-frame will actually be.

While games might certainly benefit from the ease of generative item descriptions and NPC dialogue here and there, the eventual merging of all content creation hailing from a single, digital identity is inevitable – and the frighteningly small scale of Google’s PaLM 2 could push us closer to this.

AI games made-to-order for the user are now on their way.

Cover image generated with Dall-E 2, then edited-in-house.