Altera's 1,000 agent civilisation Minecraft simulation
I wanted to draw this one to your attention in case you missed it.
We're hearing a lot about the possibilities of a world whereby you can deploy your own AI agent to manage a specific domain for you. For example, you might have a Finance AI agent that manages all of your insurance, loans, cards, income, expenses and so on. Or you might have an AI agent that helps keep you updated on a particular hobby or passion project.
If you're interested in Conversational AI, the chances are you're probably interested in this topic too – indeed, the generic, default interface for humans and agents is highly likely to be conversational in nature.
The team at Altera wanted to explore the possibilities of deploying 1,000 agents into a virtual world to see what they would do. How would they organise themselves? How would they communicate? How would they manage their world?
Well, you can watch for yourself. It's a fascinating experiment of what's already possible today.
Here are some quotes I took from the video summarising the experiment (embedded below):
- "This is a Minecraft server, but every player here is actually an autonomous AI agent"
- "They are completely Minecraft agnostic, capable of using other apps and services (e.g. Discord, Google Docs)"
- "Our agents started with nothing and worked together to gather 300+ items in Minecraft"
- "They set up a market where agents agreed to use gems as a currency"
There were three fascinating takeaways.
- Olivia's Dream: One of the agents was producing all the food for the other agents. She had a dream to go and explore, but was persuaded to stay in her role for the good of 'society'.
- Trump vs Kamala: The Altera team experimented to see how two different simulations worked with different leadership styles.
- Missing Villagers. The larger society worried about the villagers who appeared to be missing and placed lots of lights to help lost agents find their way to the village
It's really interesting to see how these agents worked together (or not) across the duration of the simulation. This points to a larger potential of many agents actively working together to achieve particular outcomes.
Further reading: