# Accounting for AI and Users Shaping One Another: The Role of Mathematical Models

**Authors:** Sarah Dean, Evan Dong, Meena Jagadeesan, Liu Leqi  
**arXiv:** [2404.12366](https://arxiv.org/abs/2404.12366)  
**Date:** 2024-04-18  
**Summary page:** https://www.summarizepaper.com/en/paper/2404.12366v1/

## Key points

- Authors Sarah Dean, Evan Dong, Meena Jagadeesan, and Liu Leqi explore the relationship between AI systems and users in societal domains
- AI systems influence and are influenced by user preferences, opinions, and behaviors
- Advocacy for the development of formal interaction models to mathematically define how AI and users shape each other
- Formal interaction models serve purposes such as specifying interactions for implementation, monitoring through empirical analysis, anticipating societal impacts through counterfactual analysis, and controlling impacts via interventions
- Consideration of factors like style, granularity, mathematical complexity, and measurability in designing formal interaction models
- Importance of incorporating formal interaction models in the design or auditing of any AI system that interacts with users
- Critical assessment of emerging literature on formal interaction models using content recommender systems as a case study
- Call for a nuanced understanding of how AI systems and users co-evolve in shaping each other's behaviors within societal contexts
- Leveraging formal interaction models can help navigate complex dynamics in human-AI interactions to create more ethical and effective AI technologies

## Summary

In their paper titled "Accounting for AI and Users Shaping One Another: The Role of Mathematical Models," authors Sarah Dean, Evan Dong, Meena Jagadeesan, and Liu Leqi explore the intricate relationship between AI systems and users in societal domains. They highlight how these systems not only influence but are also influenced by user preferences, opinions, and behaviors. Despite this mutual shaping process, the design of AI systems often overlooks this dynamic interaction. The authors advocate for the development of formal interaction models that can mathematically define how AI and users shape each other. These models serve multiple purposes such as specifying interactions for implementation, monitoring interactions through empirical analysis, anticipating societal impacts through counterfactual analysis, and controlling societal impacts via interventions. The complexity of designing formal interaction models necessitates careful consideration of factors like style, granularity, mathematical complexity, and measurability. Using content recommender systems as a case study  the authors critically assess the emerging literature on formal interaction models in relation to various use-cases and design axes. They emphasize the importance of incorporating formal interaction models in the design  or auditing of any AI system that interacts with users. Overall  this paper calls for a more nuanced understanding of how AI systems and users co-evolve in shaping each other's behaviors and preferences within societal contexts. By leveraging formal interaction models  researchers and practitioners can better navigate the complex dynamics inherent in human-AI interactions to create more ethical and effective AI technologies.