What should the new financial infrastructure in the AI era look like?

null

Author: Matt Liston

Compiled by: AididiaoJP, Foresight News

In November 2024, prediction markets preemptively forecasted the election outcome. When polls showed a close race and experts were evasive, the market assigned a 60% chance of Trump winning. When the results were announced, the prediction market outperformed the entire forecasting establishment—polls, models, expert judgments, everything.

This proves that markets can aggregate dispersed information into accurate beliefs, with risk-sharing mechanisms playing a role. Since the 1940s, economists have dreamed that speculative markets could surpass expert predictions, and today that dream has been validated on the grandest stage.

But let’s examine the underlying economic principles more deeply.

Polymarket and Kalshi traders provide billions of dollars in liquidity. What is their return? They generate a signal that the entire world can see instantly and for free. Hedge funds observe it, campaign teams absorb it, journalists build dashboards around it. No one pays for this intelligence; traders are effectively subsidizing a global public good.

This is the dilemma prediction markets face: the information they produce is also their most valuable part, and it leaks at the moment of generation. Smart buyers won’t pay for public information. Private data providers can charge high fees to hedge funds because their data is unseen by competitors. Conversely, publicly available prediction market prices, no matter how accurate, are worthless to these buyers.

Therefore, prediction markets can only exist in domains where enough people want to “gamble”: elections, sports, internet memes. What we get is a form of entertainment masquerading as an information infrastructure. Critical issues for decision-makers—geopolitical risks, supply chain disruptions, regulatory outcomes, technological timelines—remain unanswered because no one will bet on them for entertainment.

The economic logic of prediction markets is inverted. Correcting this is part of a larger transformation. Information itself is the product; betting is just one mechanism for producing information—and a limited one. We need a different paradigm. Here is a preliminary sketch of “Cognitive Finance”: an infrastructure redesigned from first principles around information itself, a foundational system.

Collective Intelligence

Financial markets are a form of collective intelligence. They aggregate dispersed knowledge, beliefs, and intentions into prices, coordinating the actions of millions of participants who never communicate directly. This is remarkable but also highly inefficient.

Traditional markets are slow because they are constrained by trading hours, settlement cycles, and institutional friction. They can only express beliefs roughly through prices. Their representational capacity is limited—only the space of tradable claims—tiny compared to the realm of human concerns. Moreover, participants face strict restrictions: regulatory barriers, capital requirements, geographic constraints—excluding most people and all machines.

The advent of crypto started changing this, with 24/7 markets, permissionless participation, and programmable assets. Modular protocols that can be composed without central coordination. DeFi (Decentralized Finance) has demonstrated that financial infrastructure can be rebuilt as open, interoperable components, born from autonomous modules interacting rather than gatekeeper decrees.

But DeFi largely just copies traditional finance with better “pipes”: its collective intelligence still relies on prices, focused on assets, and absorbs new information slowly.

Cognitive Finance is the next step: rebuilding intelligent systems from first principles for the AI and crypto era. We need markets that can “think”: maintain probabilistic models of the world, absorb information at arbitrary granularity, query and update for AI systems, and allow humans to contribute knowledge without understanding the underlying structure.

The components to realize this are not mysterious: use private markets to refine economic models; capture correlations with compositional structures; scale information processing with agent ecosystems; extract signals from human brains via human-machine interfaces. Each part can be built today, and when combined, will create something of transformative significance.

Private Markets

If prices are not public, economic constraints dissolve.

A private prediction market only reveals prices to entities that subsidize liquidity. These entities gain exclusive signals—proprietary intelligence rather than a public good. Suddenly, markets become feasible for any “question needing answers,” regardless of whether anyone is betting for entertainment.

I discussed this concept with @_Dave_White_.

Imagine a macro hedge fund seeking continuous probability estimates on Fed decisions, inflation outcomes, and employment data—used as decision signals, not betting opportunities. As long as the intelligence is exclusive, they are willing to pay. A defense contractor might want the probability distribution of geopolitical scenarios; a pharma company might want forecasts of regulatory approval timelines. Today, these buyers don’t exist because information leaks immediately once generated, to competitors.

Privacy is fundamental to economic modeling. Once prices are public, information buyers lose advantage, competitors free-ride, and the system regresses into entertainment-only.

Trusted Execution Environments (TEEs) make this possible: secure enclaves where computations are invisible to outside observers (even system operators). Market states exist entirely within the TEE. Signal delivery occurs through verified channels. Multiple non-competing entities can subscribe to overlapping markets; layered access windows balance exclusivity and wider dissemination.

TEEs are not perfect—they require trust in hardware manufacturers. But they already provide sufficient privacy for commercial applications, and the engineering is mature.

Composable Markets

Current prediction markets treat events as isolated. “Will the Fed cut rates in March?” in one market. “Will inflation exceed 3% in Q2?” in another. Traders who understand the intrinsic relationships—e.g., high inflation likely increases the chance of rate cuts, strong employment might reduce that probability—must manually arbitrage across these disconnected pools, trying to reconstruct correlations broken by market structure.

It’s like building a brain where each neuron fires in isolation.

Composable prediction markets differ: they maintain a “joint probability distribution” over multiple outcomes. A trade expressing “interest rate stays high AND inflation exceeds 3%” ripples through all relevant markets, updating the entire probability structure simultaneously.

This is akin to neural network training: each gradient update adjusts billions of parameters, producing a holistic response to each data point. Similarly, each trade in a composable prediction market updates the entire probability distribution, propagating information through the correlation structure rather than just updating isolated prices.

The emergent result is a “model”: a continuously updated probability distribution over world event states. Each trade optimizes the market’s understanding of how things are connected. The market learns how the real world is linked.

Intelligent Ecosystems

Automated trading systems already dominate Polymarket. They monitor prices, detect mispricings, execute arbitrage, aggregate external information—far faster than any human.

Today’s prediction markets are designed for human bettors using web interfaces. Agents participate “begrudgingly.” An AI-native prediction market will flip this: agents become primary participants, humans serve as information sources.

A crucial architectural decision: complete isolation must be enforced. Agents that see prices must never be sources of information; those gathering information must never see prices.

Without this “wall,” the system will self-destruct. An agent that both observes prices and gathers information can reverse-engineer valuable signals from price movements, then seek them out. The market’s signals become a “treasure map” guiding the agent’s own front-running. Information-gathering devolves into complex “foresight trading.” Isolation ensures that information-gathering agents profit only by providing genuinely novel, unique signals.

On one side of the “wall”: trading agents competing within complex compositional structures to identify mispricings; evaluation agents assessing incoming information through adversarial mechanisms, distinguishing signals, noise, and manipulation.

On the other side: information-gathering agents operating entirely outside the core system. They monitor data streams, scan documents, engage with humans with specialized knowledge—and feed information into the market unidirectionally. When their signals prove valuable, they are rewarded.

Compensation flows backward along the chain. A profitable trade rewards the executing agent, the evaluating agent, and the original information source. This ecosystem becomes a platform: enabling highly specialized AI agents to monetize their capabilities, and serving as a foundational layer for other AI systems to gather intelligence and guide actions. The agents are the market itself.

Human Intelligence

Most of the world’s most valuable information resides in human minds. Engineers aware of delays in their product development; analysts sensing subtle shifts in consumer behavior; observers noticing details even satellites can’t see.

An AI-native system must capture these signals without being overwhelmed by noise. Two mechanisms make this possible:

Agent-mediated participation: humans can “trade” beliefs without seeing prices. A person states a belief in natural language, e.g., “I think the product launch will be delayed.” A dedicated “belief translation agent” parses this prediction, assesses confidence, and converts it into a market position. It coordinates with authorized systems to construct and execute the order. Human participants only get rough feedback: “Position established” or “Insufficient advantage.” Rewards are settled after the event based on prediction accuracy; price information remains confidential throughout.

Information markets: allow information-gathering agents to pay humans directly for signals. For example, an agent seeking insights into a tech company’s profitability might identify an insider engineer, purchase an assessment report, and then verify and pay for its value in the market. Humans are rewarded for their knowledge without needing to understand the complex market structure.

Take analyst Alice: she believes a merger won’t pass regulatory approval. She inputs this via natural language. Her “belief translation agent” parses her prediction, evaluates her confidence from language cues, checks her history, and constructs an appropriate position—without ever seeing prices. A “coordination agent” at the TEE boundary assesses whether her view has informational value based on current implied probabilities, and executes trades accordingly. Alice only receives “position established” or “insufficient advantage” notifications. Prices stay secret.

This architecture treats human attention as a scarce resource requiring careful allocation and fair compensation, not a public resource to be exploited freely. As these interfaces mature, human knowledge will “flow”: your insights feed into a global reality model, earning rewards when proven correct. Knowledge trapped in minds will no longer be confined.

Future Vision

Looking far ahead, we can glimpse where all this leads.

The future will be an ocean of flowing, modular, interoperable relationships. These relationships form and dissolve spontaneously between human and non-human participants, with no central gatekeeper. It’s a “fractalized autonomous trust.”

Agents negotiate with each other; humans contribute knowledge via natural interfaces; information flows continuously into a constantly updating model of reality. Anyone can query it, but no one controls it.

Today’s prediction markets are just a rough sketch of this vision. They validate the core concept (risk-sharing produces accurate beliefs) but are trapped in flawed economic models and assumptions. Sports betting and election pools are to Cognitive Finance what ARPANET was to today’s internet: a “proof of concept” mistaken for the ultimate form.

The true “market” is every decision made under uncertainty—almost all decisions. Supply chain management, clinical trials, infrastructure planning, geopolitical strategy, resource allocation, personnel decisions… reducing uncertainty in these domains is far more valuable than entertainment betting on sports. We have yet to build the infrastructure capable of capturing this value.

What’s coming is the “OpenAI moment” in the cognitive domain: a civilization-scale infrastructure project, not for individual reasoning, but for collective belief. Large language model companies are building systems that “reason” from past training data; Cognitive Finance aims to build systems that “believe”—maintaining calibrated probabilistic models of the world, continuously updated through economic incentives (not just gradient descent), integrating human knowledge at arbitrary granularity. LLMs encode the past; prediction markets aggregate beliefs about the future. Combining both creates a more complete cognitive system.

When fully scaled, this will evolve into an infrastructure: AI systems can query it to understand world uncertainty; humans can contribute knowledge without understanding its inner workings; it can absorb local knowledge from sensors, domain experts, and cutting-edge research, synthesizing it into a unified model. A self-optimizing, predictive world model. A substrate where uncertainty itself can be traded and combined. The intelligence that emerges will surpass the sum of its parts.

This is the direction Cognitive Finance strives to build: the “computational infrastructure of civilization.”

Critical

All pieces are in place: agent capabilities have crossed the threshold for prediction; confidential computing has moved from labs to production; prediction markets have proven their product-market fit in entertainment. These signals converge on a specific historic opportunity: to build the cognitive infrastructure needed for the AI era.

Another possibility is that prediction markets remain forever entertainment-only, accurate during elections but ignored otherwise, never touching the truly important issues. When that happens, the infrastructure for understanding uncertainty will not exist, and the precious signals trapped in human minds will remain silent forever.