Government by Algorithm

David Freeman Engstrom, Stanford Daniel E. Ho, Stanford Catherine Sharkey, NYU Mariano-Florentino Cuéllar, Stanford Governance Task Description Prototypical Examples asdfsdfa prioritizing help agency identify or prioritize . SEC, CMS, IRS predictive targeting tools enforcement enforcement targets . CBP and TSA facial recognition tools

. NOAA thunderstorm hazard prediction regulatory monitoring inform agency policy-making and . FDA analysis of adverse drug event data and analysis research . FCC analysis of rulemaking comments . CFPB analysis of consumer complaints . SSA Insight System for correcting aid agency adjudication of benefits, adjudicatory errors supporting adjudication licenses, regulatory violations . USPTO tools for adjudicating patent and trademark applications . USPS, HUD use of chatbots citizen services and support provision of services to . USPS autonomous vehicles project and engagement public or communication with public handwriting recognition tool aid agency resource allocation – internal agency and . HHS Accelerate tool to assist contracting managing employees, prioritizing personnel management officer procurement decisions work, procurement Case #1 – Enforcement (and Legal Puzzles of AI) UsEnforcement by Algorithmage Legal Puzzles

● Collision: legal reason-giving v. black boxes ● Heckler holds? ○ Reviewability concern = generalist judges, no “focal point” ○ Do algos make this worse or better? ● Transparency’s limits Algos encode law, priorities -- rebuts Heckler? -- rules subject to N&C? Case #2 – Formal Adjudication (and Capacity-Building Challenges)

The New Lawyering NLP-based error detection (“Insight”) The New Lawyering

“Please evaluate.” How Accountability?e

Benchmarking – “human alongside the loop”? Q&A

Technical Challenges New technical vistas

Input challenges and enforcement “whack-a-mole” Sky’s the Limit?e