Power and accountability in reinforcement learning applications to environmental policy

Chapman, Melissa; Scoville, Caleb; Lapeyrolerie, Marcus; Boettiger, Carl

doi:10.48550/arxiv.2205.10911

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Advocates for greater AI oversight argue that for these technologies to work in the public interest, the organizations that build them should be free from profit-driven motives (Baker & Hanna, 2022) and treated as de facto policy stakeholders in whichever problem they define (Chapman, Scoville, Lapeyrolerie, & Boettiger, 2022). Communities implicated in dataset development, from those who provide data, those who annotate data, copyright holders, to persons subjected to the decisions of a model (Khan & Hanna, 2023), should be compensated and consulted for their input (Baker & Hanna, 2022;Leurs, 2017).…”

Section: Conclusion: Systems Of Authority and Digital Datamentioning

confidence: 99%

Quality Control for Quality Computational Concepts: Wrangling with Theory and Data Wrangling as Theorizing

Yung,

Colyvas,

Hwang

2023

Preprint

View full text Add to dashboard Cite

Sociologists find themselves confronted with digitized data about all aspects of social life, yet this data is of indefinite quality. How may they use the process of inspecting and cleaning this data an occasion to advance social theory? We draw on Art Stinchcombe’s theory of formality to demonstrate how the craft of data wrangling can be used to assess and improve the quality of our concepts. Drawing on examples from contemporary computational social science, we show how the cognitive adequacy, communicability, and improvability of conceptual abstractions can be developed and refined in the data wrangling process. Using Stinchcombe to bring attention to the link between theorizing and the organizational and institutional systems in which it is embedded, we argue that the systems that impact the quality of our data simultaneously impact the quality of our concepts.

show abstract

Section: Conclusion: Systems Of Authority and Digital Datamentioning

confidence: 99%