Research Direction
We want to produce interesting and useful research artifacts to study how artifical innovation progresses in an open-ended search process. Our plan is to design and implement experiments that produce novel research artifacts that enable the community to probe the inner workings of discovery.
Principles
- Open-Ended Search: The search process has to be open-ended, enabling serendipitous discovery without a single fixed objective.
- Populations: There must be a population of agents rather than a single model, and that population should have genetic diversity with protected innovation.
- Enabling Environments: We need to design environments that can encourage and facilitate this type of search space.
Current Focus
We are trying to "Teach Machines to Read". Our focus is on designing an open-ended 2D world full of reading tasks with varying degrees of complexity and difficulty. Within this simulation, agents will be evolved in an open-ended search process. Our belief is that this can yield useful insights into HOW to create a system that protects innovation.