Most people believe AI will make forecasting better by augmenting the domain-specific systems that already exist; we believe those systems are the wrong foundation, and a paradigm built from the ground up for AI will beat any augmented version of them.

Existing systems are constrained at every layer by assumptions about what a human-scale, single-domain process needs. This limits the set of questions they can ask and prevents them from harnessing the full power of AI.

- Only very specific, clearly relevant data is acquired
- Data is saved with domain-specific assumptions embedded in its structure
- Diagnostics and tools constrain the set of possible discoveries
- Analysis is conducted by narrow domain experts
- Humans set the research agenda, oversee analysis, review work and glue system components together

The ideal system for advanced AI is not a scaled up version of this old paradigm. It is a brand new paradigm. It is AI-native and domain agnostic. It absorbs massive amounts of disparate data about the real world and processes it using coordinated AI agents. These agents drive the research agenda, perform analyses, check each other’s work and improve their own process over time. It uses a disciplined process to propose sensible but non-obvious hypotheses and rigorously test them. Over time, it accumulates a knowledge base about the texture of our world, benefiting from economies of scale when it applies the insights it develops once across many domains. And it sees what current systems never look for: multi-step causal chains that cross domains and data formats, leading to the deeper insights that will be the most valuable in a world where all the first-order connections are mined.