Civilization failures occur when the system gets stuck in a state where obvious improvements exist but can't be implemented.

This chapter from the book Inadequate Equilibria categorize the causes of civilization failures into three buckets:

Coordination failures. We can't magically coordinate everyone to be carbon-neutral for example.
Decision-makers who are not beneficiaries, or lack of skin-in-the-game.
Asymmetric information. When decision-makers can't reliably obtain the necessary information they need to make decisions, from the people who have the information.

However, all of the above problems stem from a single cause: people don't share the same exact genes.

Clonal Ants, who do have the same genes, have no problems with coordination, skin-in-the-game or passing the relevant information to the decision-makers. Same goes for each of the 30 trillion cells we have in our bodies, which engage in massive collaboration to help us survive and replicate.

Evolution makes it so that our ultimate goal is to protect and replicate our genes. Cells share 100% of their genes, their goals are aligned and so cooperation is effortless. Humans shares less genes with each other, so we had to overcome trust issues by evolving complex social behaviours and technologies: status hierarchies, communication, laws and contracts.

I am doing Multi-Agent Reinforcement Learning (MARL) research where agents with different genes try to maximise their ultimate goal. In this sandbox environment, civilization failures occur. What's interesting is that we can make changes to the environment and to the agents themselves to learn what are the minimum changes required to prevent certain civilization failures.

Some examples of questions that can be explored in this setting (that I've called kinship-aligned MARL):

In a world where agents consume the same resources to survive and reproduce. If it's possible to obtain more resources by polluting everyone's air, can agents learn to coordinate and stop global intoxication?
What problems are solved when agents start to communicate? What problems arise if all communication is public? What if they have access to private encrypted communication?

Can you think of more interesting questions? I would love to hear them!

Right now I have developed an environment where agents with divergent interests either learn to cooperate or see their lineage go extinct. This envir...

Posts

Wikitag Contributions

Comments