What Happened
AI has been almost absent from the just-completed annual Kansas City Federal Reserve event in Jackson Hole. Searching the programs over the last four years for AI, artificial intelligence, machine learning, deep learning, neural networks, LLMs, generative AI, and AI agents gives this
The exception this year was important. Princeton economist Markus Brunnermeier presented “Artificial Intelligence and the Brave New World in Finance,” arguing that AI changes finance from a problem of **asymmetric information **to one of asymmetric understanding. Traditionally, different participants know different things, but understand one another well enough that disclosure, regulation, and trust can bridge the gaps.
AI breaks that symmetry. As Brunnermeier argues, AI can increasingly understand humans’ understanding, while humans cannot do the reverse. LLM decisions can be difficult to reconstruct, its objectives impossible to specify completely, and its behavior potentially inscrutable even when all the underlying information is observable.
Agentic AI pushes this much further, as the paper alludesm and as recent Hugging Face event shows. The transition from models that answer questions to agents that act, adapt, coordinate and pursue goals across time accelerates the evolution of complex systems away from direct human control and even human comprehension. As agents interact with other agents, **behavior increasingly emerges from the system **rather than from any single model, prompt or designer.
What It Means
AI markets understand us. We do not understand AI markets.
Once AI agents dominate trading, credit, risk and market-making, the financial system becomes populated by actors thatcan model human institutions while remaining increasingly opaque to them.