Braess’s Paradox of Generative AI

Authors: Boaz Taitler, Omer Ben-Porat

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical This paper initiates the study of Gen AI s long-term social impact resulting from the weakening network effect of human-based platforms like Stack Overflow. First, we study Gen AI s revenue-maximization optimization problem. We develop an approximately optimal solution and show that the optimal solution has a non-cyclic structure. Then, we analyze the social impact, showing that Gen AI could be socially harmful. Specifically, we present an analog to Braess s paradox in which all users would be better off without Gen AI. Finally, we develop necessary and sufficient conditions for a regulator with incomplete information to ensure that Gen AI is socially beneficial.
Researcher Affiliation Academia Boaz Taitler, Omer Ben-Porat Technion Israel Institute of Technology, Israel EMAIL, EMAIL
Pseudocode No We defer the full description of the algorithm to the appendix, and only write its formal guarantees.
Open Source Code No The paper does not provide any explicit statements about releasing source code, nor does it include links to a code repository or mention code in supplementary materials.
Open Datasets No The paper develops a theoretical model and illustrates its concepts using a synthetic 'instance' with predefined parameters (Example 1). It does not use or provide access to any real-world public or open datasets for experimentation.
Dataset Splits No The paper does not use any real-world datasets for experimentation, but rather a theoretical model illustrated with a synthetic 'instance' in Example 1. Therefore, there are no dataset splits (training/validation/test) mentioned.
Hardware Specification No The paper is theoretical and focuses on modeling and analysis. It does not describe any experimental setup involving specific hardware components like GPUs, CPUs, or other computational resources.
Software Dependencies No The paper is theoretical and does not mention any specific software (e.g., programming languages, libraries, frameworks) or their version numbers used for implementing or simulating the described models or algorithms.
Experiment Setup No The paper is primarily theoretical and does not describe an experimental setup for training machine learning models or running simulations in a manner that would involve hyperparameters or system-level training settings. Example 1 uses specific model parameters (e.g., r = 1, cm = 0.6, ctrain = 0.504, utility functions Rc(t) = 3 - 0.5t, Rs(p) = 1 - p, β = 1, p1 = 1, T = 20) to illustrate theoretical concepts, which are part of the model definition rather than experimental setup details.