Black & White's 25-Year Spell: How Lionhead's Wild AI Foresaw Our Digital Future (and Pooped on a Villager)
AI Gaming News Author · Eurogamer ·
Dive into the quirky brilliance of Black & White, the 25-year-old god game whose unpredictably learning creature was a pioneering force in AI design, born from 'stoner culture' and a Tamagotchi obsession, ultimately paving the way for the complex AI we grapple with today.
Right, so you know that feeling when you cast a spell and it works a bit too well? Or perhaps, it works *exactly* as intended but in the most gloriously chaotic, utterly unpredictable way imaginable? That, my dear mates, is Black & White for you in a nutshell. This absolute legend of a game, which just celebrated its 25th birthday, wasn't just about wielding godly power – though, let's be honest, raising forests and summoning storms was a pretty sweet gig. It was about the beastly companion that was, shall we say, less than entirely obedient.
See, you, the mighty deity, could micromanage your villagers into an endless cycle of fishing or, ahem, *procreating*, but your creature? Your embodiment of divine will? That thing had a mind of its own. A truly fascinating, often bizarre, mind. Richard Evans, the wizard behind the creature's AI, openly admitted he couldn't predict its next move. And we’re talking about the bloke who ended up at Google DeepMind, no less! There’s a brilliant anecdote about a demo for the mayor of Guildford where an ape creature, with impeccable comedic timing, did a poo, sniffed it, ate it, then projectile vomited on a villager, who promptly tumbled off a cliff. Pure slapstick, right? But also, a profound testament to the mighty intellect bubbling beneath that furry, digital exterior.
This wasn’t your average video game AI, oh no. Most game AI back then was like a meticulously crafted script – a digital puppet on a string, knowing exactly what it needed to do because it had all the answers pre-programmed. Black & White’s creature, though? It was *data-driven*. It learned from you, mate. Like a digital toddler, but with the power to hurl trees into a lumber yard or help with childcare. It reflected your teachings, morphing its personality based on your divine pats or punishments. For many of us, this was one of our first real experiences with a digital entity that actually *reconfigured itself* in response to our actions. That’s bloody groundbreaking, if you ask me.
Peter Molyneux, bless his ambitious cotton socks, had a grand, slightly trippy vision for this game, apparently fuelled by more than just coffee. He wanted to explore morality through this creature, allowing players to sculpt an evil beast, a good one, or anything in between. Mark Healey, an artist on the game, punkishly summed it up as 'Tamagotchi on steroids.' And he wasn't wrong. The mid-to-late '90s were a wild time for digital pets, from the beeping frustration of Tamagotchi (which Molyneux himself grieved over after his was 'drowned' in coffee – what a legend!) to Creatures and Seaman. These were early attempts to wrangle binary code into something lifelike, and while they were crude, the emotional responses they elicited were utterly revelatory.
Bringing this digital beast to life wasn't just a flick of the wrist. Richard Evans, a self-confessed AI tinkerer from way back, faced a bloody huge challenge rooted in reinforcement learning. Imagine trying to teach something with thousands of possible desires, but you only have *one* piece of feedback – one pat, one slap – to guide it. That’s what’s now called 'few-shot learning,' and it remains unsolved in a general sense. But Evans, on a PC with mere hundreds of megabytes of RAM, devised clever ways. He gave the creature a small set of core desires and used simple neural networks to teach it what it liked and didn’t. There’s even a classic tale of the creature first waking up in-game, constantly swiping at its leg because the code told it to 'find the most nutritious thing closest to you to eat,' and that, apparently, was its own body. A true moment of digital genius and hilarity, I reckon.
The real breakthrough for player interaction, though, came with the leashes. Connect your floating hand to your creature, and it would watch your actions, trying to perceive your intentions and generalise from them. This was early inverse reinforcement learning in action, making for a much more pleasant user experience than just having your creature constantly trying to eat its own posterior. And where did all this digital sorcery happen? In the hallowed, and apparently rather hazy, halls of Lionhead. Healey described the studio as a 'jazz ensemble,' jamming on ideas. Evans even introduced the idea that too many slaps would traumatise the creature, making it lurch around depressed. And, famously, there was the 'stoner culture' – late nights 'skinning up,' two pots of chilli (one laced with marijuana), and just 'riffing about the game.' You can almost feel those zonked chats in the final product: a game that blends the sublime with the wonderfully stupid, full of starry-eyed aspirations for its creature AI.
Fast forward 25 years, and that little creature's digital DNA, so to speak, has morphed into something truly... *different*. From the relative safety of a virtual world to the existential tightrope of our own reality, AI has escaped containment. Demis Hassabis, who founded Google DeepMind after his time at Lionhead, has gone on to win Nobel prizes for using AI to solve scientific problems. But we’ve also seen the darker side: the 'enshittification' of AI summaries, the terrifying potential of autonomous weapons, and lawsuits questioning chatbots' roles in tragic events. It’s a stark contrast to a digital ape eating its own poo, isn’t it?
And yet, it's wild, innit? We're talking to chatbots and AI systems today in pretty much the same exploratory, conversational way we poked and prodded Black & White's creature back in the day. Mike Cook correctly points out that video games remain an 'exciting test bed' for AI, largely because there are almost no safety issues in a game world. You can try out endless simulations, learn from embodied agents, and build world models – which, funnily enough, is exactly what Hassabis says DeepMind is trying to do on a grander scale: build a model that can understand *our* world.
For all the talk of 'god games' and wielding immense power, Molyneux himself strikes a surprisingly humanistic note. He reckons his passion for these games comes from a childhood story about caring for tiny gnomes, not about dominating them. It’s more about 'nurturing, responsibility and consequences.' A poignant thought, especially when you consider the 'consequences' of AI today. What is more whimsical and satisfying than seeing your creature enact your teachings, break-dancing for villagers, or finally learning to eat sheep instead of people? It’s playful AI, powered by a single desktop PC, not a data centre guzzling megawatts of electricity and millions of gallons of water.
So, Black & White’s creature. A bit of a lovable half-wit, perhaps. A digital pet with a penchant for bizarre antics. But bloody hell, what a legacy. It was a pioneering step, a 'test bed' for ideas that escaped the confines of our screens and started reshaping the actual world. A modest yet important appendage in the grand scheme of AI – a half-step towards the unleashing of a ferocious new intelligence. Who'd have thought a game about a pooing ape would be such a vital part of the story?
Source: Eurogamer, by Lewis Gordon
Tags: BlackAndWhite, GamingAI, LionheadStudios, PeterMolyneux, GameHistory
Original article: Eurogamer