From Simon’s Legacy to CI: The New Micro-Foundations of Finance

TL;DR

This paper serves as a seminal editorial guide that bridges the gap between Herbert Simon’s classical AI and modern Computational Intelligence (CI). It argues that the future of economics lies in Agent-Based Computational Economics (ACE), where autonomous software agents, equipped with learning mechanisms like Genetic Algorithms and Neural Networks, replace the idealized "rational man." The core insight is that CI allows us to model a "procedural rationality" that actually reflects human behavior in complex, adaptive markets.

Problem & Motivation: The Limits of Symbolic Logic

For decades, economics was built on the "Symbol-Processing" approach—a form of classical AI pioneered by Herbert Simon. While groundbreaking, this approach hit a wall:

• Where do symbols come from? Manual models couldn't explain how agents innovate or adapt to environmental feedback.
• Static Assumptions: Traditional econometrics assumes a "representative agent," ignoring the messy reality of heterogeneous, boundedly-rational actors.

The author suggests that the economic system is a Complex Adaptive System (CAS). To understand it, we must move beyond manually-driven devices to Autonomous Agents that can create and evolve their own strategies.

Methodology: The CI Toolkit for Economics

The paper details how CI provides the "muscles" for Simon’s "bounded rationality." The methodology focuses on three pillars:

1. Hybrid Systems: Instead of using one tool, the most effective models are synergetic. For instance, the Guarded Experts framework integrates:
   • Neural Networks for prediction.
   • Genetic Algorithms to evolve new strategies.
   • Classifier Systems to act as "guards" that switch between local models based on market regimes.
2. Field-Grounded Engineering: Unlike early ACE models, newer approaches (like Izumi et al.) ground agent rules in actual field studies and interviews with investors, ensuring the "Inherent Bias" of the agents reflects real human psychology.
3. Cross-Platform Integration: Systems like X-Economy allow human traders and software agents to compete in the same digital sandbox, studying the co-evolution of human-machine interaction.

Figure 1: The historical lineage and intersection of AI, CI, and Economics.

Experiments & Results: The Micro-Macro Paradox

The most striking experimental evidence discussed involves the Price-Volume Relation.

• Traditional View: Econometricians use Granger causality to see if trading volume predicts price.
• CI Finding: In the AIE-ASM (Artificial Intelligence Economics - Artificial Stock Market), macro-level tests showed that volume "caused" price changes. However, when looking at the micro-structure, zero agents were actually using volume data in their survival-set strategies.

Result: This proves a dangerous inconsistency—macro-econometric results can be "mirages" that lead to misleading conclusions about individual behavior.

Note: CI tools permit Survival Analysis of strategies, showing which economic variables are actually valued by agents over time.

Critical Analysis & Conclusion

Takeaway

CI has transitioned from a fringe computational tool to a "constructive foundation" for economics. By simulating the process of decision-making (Procedural Rationality) rather than just the outcome, we gain a deeper understanding of market stability and cooperation (as seen in Iterated Prisoner’s Dilemma simulations).

Limitations & Future Work

The author admits that adding Evolutionary Forces makes models incredibly difficult to trace. "Rich behavior" often looks like "noise." The next frontier is to balance this complexity—building robust markets with known variables first, then introducing evolutionary dynamics once the baseline micro-macro relations are understood.

Future finance will likely not be about finding a "global model" but about identifying "local models" in a piecewise stationary world.