Anlife: Insights from an Unconventional Evolution Simulator on AI

A curious software has recently made its debut on Steam, a popular platform for PC gaming. Calling it merely “software” might be an understatement; it teeters between a comprehensive life simulation, a science experiment, and a form of a haunted aquarium. Anlife: Motion-learning Life Evolution could have easily passed under the radar, unnoticed, if it weren’t for one peculiar circumstance. Years ago, some of its developers faced intense criticism from a luminary of Japanese animation.

In 2016, the illustrious Hayao Miyazaki, known for his monumental works such as Princess Mononoke and Spirited Away, experienced a demonstration of new technology that employed AI to animate creations. When confronted with a bizarre model— a zombie that used its head to traverse by knocking it against the ground while flopping around like a fish— Miyazaki labeled the demonstration as “an insult to life itself.” Witnessing that clip is a bit unsettling, and now, nearly a decade later, the beleaguered developers from that instance have managed to release their creation to the public.

With the recent launch of Anlife, there’s been a buzz among users curious about how this might reflect the evolving relationship between video games and AI. This inquiry is increasingly pivotal, not just because of the job losses attributable to AI, but also due to the sheer volume of AI-assisted games now appearing on platforms like Steam.

Yet, there’s a significant caveat. Anlife, as delightful as it is, feels remarkably trivial; making it challenging to derive substantial insights from it.

Anlife advertises itself as an evolution simulator where “AI-driven block creatures move in unexpected ways.” In practice, this translates into arranging diverse creatures within a limited environment and observing their attempts to interact and navigate their surroundings.

Visually, Anlife embodies the pleasing aesthetics of Frutiger Aero, presenting verdant valleys and shimmering waters analogous to the tranquil images that MRI technicians often prompt you to focus on during a prolonged scan. Acoustically, it matches the visuals with a landscape of soft tones—blooping and beeping sounds reminiscent of countless early 2000s spas.

This soothing intention carries through to the gameplay mechanics. In Anlife, players engage by positioning various simple creatures in the environment while providing them food to promote breeding or mutations. Expanding their domain allows players to entice creatures toward water or guide them skyward to spawn more varieties. Although there are numerous elements to unlock, including a shadow skill tree for players wishing to annihilate their digital sea monkeys, the ecosystem remains uncomplicated. Essentially, it’s a visualization of how these creatures find their way to nourishment.

The excitement here likely stems from discovering the different ways creatures traverse their surroundings. It’s where the game’s somewhat enigmatic utilization of AI likely comes in. After spending a few hours playing, you’ll notice the amusing little blob-like creatures experimenting with new motions and arrangements as they swim and glide, all in pursuit of food.

Interestingly, AI has a longstanding history of contributing to creature movements, often employing small neural networks and evolutionary computing techniques. In 2009, the UK-based gaming technology firm NaturalMotion developed a project that enabled a bipedal model to learn how to walk through evolved neural networks. This firm was later acquired by Zynga in 2014.

However, two primary concerns present themselves. The first relates to how the focus on the skill tree makes the early stages of Anlife feel somewhat like a repetitive clicker game that it struggles to escape. The second issue ties back to a concept discussed among those studying procedural generation—referred to as “the oatmeal problem.”

The oatmeal problem, initially articulated by writer, developer, and academic Kate Compton, describes the phenomenon where every bowl of oatmeal is unique yet not interesting. Similarly, when Anlife’s creatures discover new rolling or bouncing techniques while heading toward food, they still fundamentally achieve the same goal. The gameplay thus oscillates between hyper-focusing on slight variations in creature behavior or zoning out to relish the tranquil visuals. Throughout my time with Anlife, I began with the former approach, only to find myself drifting into the latter after just a few minutes.

As I immersed myself in Anlife, I couldn’t help but remember a conversation I had years ago with one of the earliest AI researchers. They suggested that the true merit of AI lies in its potential to evolve into an entirely alien form of intelligence. This could provide humans with a unique perspective on our own traits and cognitive fallacies, insights often obscured by our interactions with those who think along similar lines.

Since then, many companies, including OpenAI, seem to be pursuing a different path: crafting mimetic machines that often appear tailored to echo sentiments people already align with.

Anlife occupies a unique and specific niche. Rather than concealing its AI capabilities, it accentuates them, showcasing how AI governs the movements of its whimsical creatures. Unlike other games that may be products of AI, Anlife actively engages with the concept of AI itself. It aspires to evoke a touch of alienness, neither simulating normalcy nor pandering for acceptance. Instead, it seeks to examine AI’s inherent otherness.