The Emerging Machine
Ten years ago very few people were talking about ML. There was no Tensor Flow, no Deep Mind and no real awareness of what neural nets could achieve. At the time I was in love with Emergence and had been since I first read about Ken Arrow and Steven Wolfram in the mid 90's. I ached to run off and join the Santa Fe Institute and I was fascinated by self-organisation and complexity arising from very simple rule-sets and autonomous agents.
But I'd fallen out of love with how it was being used in games. I realised that you needed to pair it with ML to balance out instabilities, the tendency towards atrophy could ruin the most carefully crafted world. But most of all ML was needed to turn boring systems into something you could have fun with.
I had three applications in mind for ML as thought starters:
- Create better game cameras by having humans manually control cameras as exemplars
- To populate highly dense environments with objects so that they look realistic
- To responsively alter the game in realtime to give each player maximum enjoyment
I guess that the third is most interesting - and most game like. The first two are basically solves for production in a time when poly counts and model detail was going through the roof.
But realtime shifting a game world's layout and altering NPC behaviours in step with that seemed like magic to me.
How did I do?
Well three years later in 2008 Left4Dead arrived with its AI DIRECTOR, which I suppose was the start of crafting the game world around you for maximum fun rather than just ramping up the challenge over time. Essentially it was a smart monster spawner - the kind of thing that GTA was already doing, but with the express intention of making the game challenging, if you like your fun in that flavour.
In between there's been an explosion of RogueLikes with the emphasis on generating fun environments (again if fun means punishingly hard).
There's a massive amount of potential for ML in co-authored game experiences. It will explode over the next 2-3 years for certain.