April 1, 2024

This Week in Responsible AI, Apr 1, 2024

This Week in Responsible AI, Apr 1, 2024

"Safety"

• Why Are Large AI Models Being Red Teamed?

• Using ASCII art to jailbreak chatbots

• AI safety is not a model property

• $250,000 in prizes for ML Safety benchmarks

• Detectors for Safe and Reliable LLMs: Implementations, Uses, and Limitations

Fairness

• SCoFT: Self-Contrastive Fine-Tuning for Equitable Image Generation

• Maximizing Equity in Acute Coronary Syndrome Screening across Sociodemographic Characteristics of Patients

• Sensitivity, Performance, Robustness: Deconstructing the Effect of Sociodemographic Prompting

• Digital Fairness for Consumers

• Fairness Feedback Loops: Training on Synthetic Data Amplifies Bias

Participation

• "How can NLP/AI practitioners engage with oral societies and develop locally appropriate language technologies?"

• Recourse for reclamation: Chatting with generative language modelsx

• From Fitting Participation to Forging Relationships: The Art of Participatory ML

Transparency

• Technical Readout - Columbia Convening on Openness and AI

• Models All The Way Down

• Revealed: the secret algorithm that controls the lives of Serco’s immigration detainees

• New YouTube policy: creators must disclose 'realistic' AI-generated content

• The Impact of Explanations on Fairness in Human-AI Decision-Making

• Anyone Can Audit! Users Can Lead Their Own Algorithmic Audits with IndieLabel

Law/Policy

• "This is not a data problem": Algorithms and Power in Public Higher Education in Canada

• Africa’s push to regulate AI starts now

• NTIA Artificial Intelligence Accountability Policy Report

• The EU AI Act passes. Here’s what will (and won’t) change. Some are unhappy with it.

• VP Harris announces new requirements for how federal agencies use AI technology

• FTC and DOJ File Statement of Interest in Hotel Room Algorithmic Price-Fixing Case

• Constructing AI Speech

• AI Countergovernance

• We’re headed for big problems if gardaí get facial recognition technology

• Google fined $272M by French government over AI use of news content

• NYC’s AI Chatbot Tells Businesses to Break the Law

• Why we’re fighting to make sure labor unions have a voice in how AI is implemented

• AI Nationalism(s)

• U.S. Department of the Treasury Releases Report on Managing Artificial Intelligence-Specific Cybersecurity Risks in the Financial Sector

• Envisioning a Global Regime Complex to Govern Artificial Intelligence

• Revealed: a California city is training AI to spot homeless encampments

Data

• Data Collection in Music Generation Training Sets: A Critical Analysis

• A Data-Centered Approach to Education AI

Elections

• Google restricts AI chatbot Gemini from answering questions on 2024 elections

• AI image-generator Midjourney blocks images of Biden and Trump as election looms

Privacy

• Mitigating a token-length side-channel attack in our AI products

• "it is possible to learn a surprisingly large amount of non-public information about an API-protected LLM from a relatively small number of API queries"

• Hackers can read private AI-assistant chats even though they’re encrypted

• Facial recognition technology and protests

Other

• The Guild of St. Luke: Reassessing Digital-Cultural Infrastructures

• Towards Optimizing Human-Centric Objectives in AI-Assisted Decision-Making With Offline Reinforcement Learning

• “Companies continue to promise to deliver the moon when it comes to AI and still provide moldy green cheese.”

• Artificial Intelligence and Machine Learning + Accessibility

• GPT 3.5 and 4 "have been globally exposed to ∼4.7M samples from 263 benchmarks"

• Dating Your (Potential) Executioner

• Let’s not make the same mistakes with AI that we made with social media

• Power and Play: Investigating "License to Critique" in Teams' AI Ethics Discussions

Compiled by Leif Hancox-Li