Artificial Intelligence and Spontaneous Collusion

Martino Banchio,Giacomo Mantegazza

2023-09-20

Abstract:We develop a tractable model for studying strategic interactions between learning algorithms. We uncover a mechanism responsible for the emergence of algorithmic collusion. We observe that algorithms periodically coordinate on actions that are more profitable than static Nash equilibria. This novel collusive channel relies on an endogenous statistical linkage in the algorithms' estimates which we call spontaneous coupling. The model's parameters predict whether the statistical linkage will appear, and what market structures facilitate algorithmic collusion. We show that spontaneous coupling can sustain collusion in prices and market shares, complementing experimental findings in the literature. Finally, we apply our results to design algorithmic markets.

Theoretical Economics,Artificial Intelligence,Computer Science and Game Theory

What problem does this paper attempt to address?

The paper attempts to address the question: In the market, does the interaction between learning algorithms lead to spontaneous collusion, even if these algorithms do not have explicit collusion intentions?

### Specific Questions:
1. **Mechanism of Algorithmic Collusion**: The paper explores how learning algorithms can form collusion through intrinsic statistical correlations (referred to as "spontaneous coupling") without explicit collusion intentions.
2. **Impact of Market Structure**: Investigates which market structures are more likely to facilitate algorithmic collusion.
3. **Impact of Algorithm Design**: Analyzes how the design of different learning algorithms affects the formation of collusion, particularly the balance between exploration and exploitation.
4. **Policy Design**: Proposes how to design market mechanisms to prevent algorithmic collusion.

### Research Background:
- **Application of Artificial Intelligence**: AI software is becoming increasingly common in various business scenarios, such as online auctions, pricing on shopping platforms, setting short-term and long-term rents, etc.
- **Regulatory Concerns**: Regulators and academic researchers are concerned that automated pricing and bidding software may facilitate collusion.

### Main Contributions:
- **Spontaneous Coupling Mechanism**: The paper reveals a new channel of collusion, "spontaneous coupling," which does not rely on monitoring technology or collusion intentions of market participants but is formed through intrinsic statistical correlations between algorithms.
- **Theoretical Model**: Developed a tractable model to study the strategic interaction between learning algorithms.
- **Application Examples**: Demonstrated how spontaneous coupling can sustain collusion in different economic environments through practical cases such as price fixing and market segmentation.
- **Market Design Suggestions**: Proposed methods for designing strategy-proof mechanisms to prevent algorithmic collusion.

### Methodology:
- **Continuous Time Approximation**: Approximated the discrete learning process as a continuous-time dynamical system to better analyze the behavior of algorithms.
- **Dynamical Systems Theory**: Utilized traditional dynamical systems techniques to analyze the outcomes of algorithms.

### Conclusion:
- **Pervasiveness of Spontaneous Coupling**: Spontaneous coupling can sustain collusion between algorithms in various economic settings.
- **Impact of Parameters**: The learning rate and other parameters of the algorithms affect the occurrence and duration of spontaneous coupling.
- **Importance of Market Design**: By providing sufficient feedback, market designers can create mechanisms that prevent algorithmic collusion.

### Significance:
- **Regulatory Policy**: Provides a theoretical basis for regulators to better understand and address the risks of algorithmic collusion.
- **Market Design**: Offers practical tools for market designers to create fairer and more transparent market mechanisms.