A strange piece of software has recently landed on the PC gaming store Steam. And “software” feels like the cleanest way to describe it. Existing somewhere between a full-blown life sim, a science project and a kind of haunted fish tank, Anlife: Motion-learning Life Evolution probably would have disappeared without making much impact if it wasn’t for one unusual factor. Several years ago some of its creators were absolutely roasted on camera by one of the genuine legends of Japanese animation.
Back in 2016, Hayao Miyazaki, the director of movies such as Princess Mononoke and Spirited Away, was shown new technology that used AI in order to animate models. Faced with a zombie that utilised its head to move by knocking its skull against the ground and wriggling its body like a fish, Miyazaki declared what he had seen was “an insult to life itself”. It’s hard not to watch the clip without feeling slightly seared – but now, a decade later, the ashen-faced developers from that room have sufficiently recovered to make their work widely available.
Judging by the chatter surrounding the launch, at least some of the people downloading Anlife are doing so in the hope that it might provide some kind of indication of the current state video games’ relationship with AI. Putting aside the often broad use of that term, this is certainly a thing that’s worth trying to understand, whether it’s because of the job losses caused by AI or blamed on AI, or the sheer number of games made with the assistance of AI models now landing on storefronts such as Steam.
There’s a problem here, though. And it’s that Anlife itself is such a cheerfully inconsequential thing that it’s hard to read too much of anything into it.
Anlife promises players an evolution simulator where “AI-driven block creatures move in unexpected ways.” What this comes down to for the most part involves placing a range of different creatures into a small environment and then watching as they learn to get around.
Visually Anlife is pure Frutiger Aero, offering landscapes of green valleys and sparkling water that could be the kind of soothing images MRI technicians sometimes encourage you to look at during a lengthy scan. Sonically it’s equally inoffensive: with a range of bloops, bleeps and popping sounds, we’re pitched right into the soundtrack of a million 00s day spas.
This desire to soothe permeates to the level of mechanics, too. Over the course of an idle morning with Anlife you can place a range of simple creatures in the environment and then give them food that will encourage them to breed or mutate. You can expand your territory and then lure creatures towards water or up into the air in order to create more variations. There are plenty of things to unlock (including a shadow tech tree that has you covered if you want to annihilate your digital sea monkeys rather than watching them flourish) but the ecosystem is kept simple. It’s a game about watching how things decide to crawl towards food.
The thrill of all of this presumably comes down to the “how” in that last sentence. This is where the game’s slightly mysterious use of AI is probably involved. And it’s true that, after a few hours of play, you’ll have Anlife’s funny little blob creatures exploring new kinds of joints and body arrangements as they swim and fly and generally roll around eating.
(AI has a history of being used to make creatures walk around, incidentally, often using small neural networks and computer evolution. In 2009, the UK-based games technology company NaturalMotion developed a project in which a bipedal model learned to walk using evolved neural networks. The company was subsequently bought by Zynga in 2014.
There are two problems, however, the first being that the focus on unlocking the skill tree gives the opening hours of Anlife the feel of a rather mindless clicker game that it struggles to shake off. The second comes down to something that I gather people studying procedural generation sometimes refer to as “the oatmeal problem”.
The oatmeal problem, which was first formulated by writer, developer and academic Kate Compton, hinges on the fact that every single bowl of oatmeal in the universe is unique. Just not in a very interesting way. Similarly, when Anlife’s creatures discover a new way of rolling or bouncing or flapping their bodies towards food, they’re still just moving towards food. It makes for a game that’s either about really, really paying attention to tiny variations in detailing, or about completely zoning out and just enjoying the floaty aesthetic. Over the course of my time with Anlife, I’ve generally started with the first approach and then discovered, after 10 minutes, that I’ve slid into the second.
The more I played Anlife, the more I ended up thinking about something I heard from one of the first AI researchers I ever spoke to, all the way back in 2013. The real value of AI, they explained – and I am paraphrasing here – is that it could one day become a completely alien form of intelligence. Which means that us humans would have the lens of a different kind of mind through which to see ourselves and our quirks and cognitive fallacies – things that can be hard to spot when you’re talking to people who think in a very similar way to you.
Since then the apparent focus for a lot of companies such as OpenAI seems to be completely the opposite: creating mimetic plagiarism machines that sometimes seem to exist only to tell people what they already want to hear.
It feels as if Anlife is filling a very small and specific niche. It isn’t trying to disguise its AI use, as far as I am aware. It’s actually trying to draw attention to the way it uses AI to allow its little creatures to move around. It’s not just a game born of AI, in other words. It’s a game that in some small way is actually about AI. It wants to be slightly alien. It doesn’t want to hide itself or ingratiate; it wants to examine AI’s innate capacity for otherness.
• Anlife: Motion-learning Life Evolution is out now on Steam; £11.79/$13.99. Thanks to Dr Michael Cook for his help with this piece.
