A handful of developers are taking lessons from evolution and genetics to create living systems.
Last year was the year of the open world for games. Games such as Fallout 4, The Witcher 3 and Batman: Arkham Knight pushed the boundaries for creating explorable worlds. Quests here, there, and everywhere; collectibles hidden in caves, and characters strewn throughout to give the world substance. These games are made to feel, in some way, alive.
Small details come together to create the façade of a world and characters that are living. They might react to the player's actions, or be just so well animated that, at a glance, you think you're watching a live-action movie. But the worlds of these games aren't alive – they're carefully crafted sets for your adventures. Back in the 1990s, however, there were ambitions to go beyond this.
This was the field of artificial life, the pursuit of creating a program, robot, or something else entirely that could be considered alive. While there are many overlapping areas with artificial intelligence, the two approaches fundamentally differ. Artificial intelligence typically focuses on mimicking human intelligence, either through sheer volume of knowledge or the ability to comprehend information, while artificial life focuses on processes that mimic living systems. Artificial life can exist without artificial intelligence, or they can exist in a system together.
Artificial life may be intelligent, and it may understand surroundings, or it may simply act in a purely survival-orientated manner. The key is that it shows signs of what we consider life to be. When it's brought to games, we get programs that can “think”, react to a virtual environment, be hurt and helped, and don't just ride the rollercoaster tracks of preprogrammed events.
Little Computer People
The first game that used a form of artificial life was Little Computer People, released in 1985. By typing in certain phrases and requests, the player could interact with a person onscreen who had just moved into their new house. Each player had a unique character: when the game was first booted, certain traits would be applied to the little computer person. Some people were lucky and got a cheery little friend. Others found their new occupant getting irritable with them.
It was possible to annoy the person so much that they refused to communicate any more, simply doing their own thing independently of what the player would ask. You could be cruel or kind to them. It was even possible to kill them.
(Above: Little Computer People)
The unique character, the set of emotions they have available, and the requirements of food and water seem tame, and it would be hard to argue they constitute “life”. However, Little Computer People did inspire perhaps the biggest and most prominent example of artificial life in games: Creatures.
Designed by Steve Grand, a roboticist and programmer who has worked on a number of projects relating to artificial life, Creatures began life in 1992 when it was pitched as an idea titled “Little Computer Ewoks”. It was to be a game where the player would indirectly play with creatures and help them grow, learn and evolve. The history behind Creatures and its development was detailed by Grand on his old website.
There was never meant to be a goal and, as with Little Computer People, it didn't stick to the norms of what people expected from video games. The player could do whatever they wanted, and the alien creatures in the game, called Norns, would react appropriately. The technology jump from 1985 to 1996, when Creatures was released, meant that it was far more complex than Little Computer People. The Norns could learn and “think” using a complex neural network – a system that imitates the brain's ability to recognise different inputs and come to the correct output. For example, the Norn could experience something good, such as a food it likes, and learn to eat it. The neural network would then adapt so that this Norn knows to eat this food when it sees it and is hungry.
(Above: Creatures 2)
It was a huge step up from the basic person in Little Computer People, and this ability to 'think' meant the Norns did seem much more alive. Coupled with improved biological systems – such as individual genetics for each Norn and random mutations for their babies – Creatures made artificial life seem feasible.
A series grew out of Creatures, and each release added new features to make the Norns seem more and more alive. They would learn, evolve, form relationships, play, and more. There may not have been a beating heart within the Norns, but they certainly did seem alive.
Learning from neural networks
Games using artificial life didn't have to be quite as open-ended, though, as 1997 release Galapagos: Mendel's Escape showed. Rather than defying genre, it was a puzzle game, where the player would indirectly interact with a creature that could “think”. As the title implies, the game focuses on Mendel, an artificial creature that was made in a lab, trying to break free.
It utilised a technology called Non-stationary Entropic Reduction Mapping (NERM) to control Mendel, as the player could not directly interact with Mendel. This would act similarly to the neural network that made up the brains of the Norns in Creatures, although with a few key differences. Rather than working to include biological processes and emotions, it focused upon understanding the local environment and its details, constantly adapting to react to anything new. If a hazard appeared from around the corner, it would know to stay away. As the game progressed, Mendel would grow more intelligent – initially, it might not recognise certain areas as being hazards, but once it was hurt by them, it would remember that without the player doing anything directly.
(Above: Galapagos: Mendel's Escape)
The player would pull levers, flick switches and solve puzzles that would allow Mendel to pass through areas unscathed. It was, at its heart, a co-operative puzzle game, where your partner was a thinking program; a living artificial creature.
With the company behind Galapagos: Mendel's Escape moving away from game development, and the Creatures series effectively ending its ambitions with artificial life back in 2001, artificial life and games have drifted apart. Games today may have complex social interactions, stories presented in immersive ways, and expansive worlds, but they don't have the goal of artificial life – only the illusion of it.
The technology is still out there, though. A popular program among those interested in artificial life is DarwinBots. It allows people to write the DNA for “bots”, which will then try to live and evolve. They can be written to be successful, and live on for a long amount of time, or to burn themselves out quickly. DarwinBots allows for experimental simulation of genetic programming, a behind-the-scenes and more detailed look at the genetic systems seen in Creatures. For those adept at designing the DNA, it can be used as a challenge to create interesting predator or prey bots.
Where DarwinBots pushes boundaries with the biological side to artificial life, SethBling on YouTube found success in advancing the use of neural networks. He developed a neural network that, after a period of adaptation, managed to learn how to play Super Mario World. Called MarI/O, the system learned how the game worked based on principles similar to those in the human brain, coupled with ideas on evolution and genetics.
Robots and programs that can be considered “living” are a long way off, and people aren't making Skynet here. What I've referred to as creatures are in fact lines of code with the impression of personality. Nevertheless, they hint at a whole new world of gaming, where the computer learns with you, adapts and grows. The end point could be the creation of living worlds that the likes of The Witcher 3, Fallout 4 and Xenoblade Chronicles X can only allude to.