• Corpus ID: 246822961

Choices, Risks, and Reward Reports: Charting Public Policy for Reinforcement Learning Systems

  title={Choices, Risks, and Reward Reports: Charting Public Policy for Reinforcement Learning Systems},
  author={Thomas Krendl Gilbert and Sarah Dean and Tom O. Zick and Nathan Lambert},
In the long term, reinforcement learning (RL) is considered by many AI theorists to be the most promising path to artificial general intelligence. This places RL practitioners in a position to design systems that have never existed before and lack prior documentation in law and policy. Public agencies could intervene on complex dynamics that were previously too opaque to deliberate about, and long-held policy ambitions would finally be made tractable. In this whitepaper we illustrate this… 
Reward Reports for Reinforcement Learning
Taking inspiration from various contributions to the technical literature on reinforcement learning, Reward Reports are outlined as living documents that track updates to design choices and assumptions behind what a particular automated system is optimizing for.
Actionable Guidance for High-Consequence AI Risk Management: Towards Standards Addressing AI Catastrophic Risks
This document provides detailed actionable-guidance recommendations focused on identifying and managing risks of events with very high or catastrophic consequences, intended as a risk management practices resource for NIST for AI RMF version 1.0 (scheduled for release in early 2023).
Sociotechnical Specification for the Broader Impacts of Autonomous Vehicles
—Autonomous Vehicles (AVs) will have a trans- formative impact on society. Beyond the local safety and efficiency of individual vehicles, these effects will also change how people interact with the


Liability for AI: public policy considerations
Liability for AI is the subject of a lively debate. Whether new liability rules should be introduced or not and how these rules should be designed hinges on the function of liability rules. Mainly,
Toward A Thousand Lights: Decentralized Deep Reinforcement Learning for Large-Scale Traffic Signal Control
This paper tackles the problem of multi-intersection traffic signal control, especially for large-scale networks, based on RL techniques and transportation theories and shows that implicit coordination could be achieved by individual control agents with well-crafted reward design thus reducing the dimensionality.
Driving Through Ghosts: Behavioral Cloning with False Positives
This work proposes a new probabilistic birds-eye-view semantic grid to encode the noisy output of object perception systems and shows that this approach can safely overcome critical false positives that would otherwise lead to catastrophic failures or conservative behavior.
Non-Autonomous Artificial Intelligence Programs and Products Liability: How New AI Products Challenge Existing Liability Models and Pose New Financial Burdens
INTRODUCTION Continuous improvements to core computing technologies over the previous decade have generated an explosive growth in artificial intelligence (AI) research and development, facilitated
Infrastructural Regulation and the New Utilities
From too-big-to-fail financial firms to net neutrality to internet platforms and the water crisis in Flint, Michigan, we now face a variety of legal and public policy problems which all share a
4. The Public Utility Idea and the Origins of Modern Business Regulation
Th is essay concerns one of the more remarkable innovations in the history of demo cratic attempts to control the American corporation. In the late nineteenth and early twentieth centuries— aft er
What's Inside the Black Box? AI Challenges for Lawyers and Researchers
The main goals of this paper, written by Ronald Yu and Gabriele Spina Alì, are to analyse some of the factors that can jeopardize the reliability of AI-assisted legal research and to review some ofThe solutions to mitigate this situation.
Regulating Informational Infrastructure: Internet Platforms as the New Public Utilities
The power and influence of dominant tech platforms — Google, Facebook, and Amazon, in particular — have become a central topic of debate. In this paper, I argue that these firms are best understood
Model Cards for Model Reporting
This work proposes model cards, a framework that can be used to document any trained machine learning model in the application fields of computer vision and natural language processing, and provides cards for two supervised models: One trained to detect smiling faces in images, and one training to detect toxic comments in text.
Amazon's Antitrust Paradox
Amazon is the titan of twenty-first century commerce. In addition to being a retailer, it is now a marketing platform, a delivery and logistics network, a payment service, a credit lender, an auction