As the world experiments, innovates, and legislates around AI, we have gathered a reading list from the last month, the fourth in a new monthly series. (Read the last issue here.)
Scholarly Publishing
ChatGPT identifies gender disparities in scientific peer review: This article uses ChatGPT to analyze language use in peer review reports. The findings indicate subjectivity in the peer review process, including that female first authors receive less polite reviews than their male peers, indicating a gender bias in reviewing. This format could serve as a rubric for using generative AI to analyze the equitability of specialized content, including but not limited to peer reviews. (eLife, November 3, 2023)
We Can’t Let AI Generation Tools Take Away Our Own Training: This thinkpiece from NISO Executive Director Todd Carpenter argues that people still need to learn how to do the basics to teach LLMs how to conceptualize and improve on those topics. The more difficult tasks will be harder to conceptualize and improve upon: "Learning from our mistakes and improving on them is core to our humanity. It’s important to understand that language models lack this capacity to know the direction of 'better.' When AI tools are trained on a corpus of data that’s machine generated, the results become increasingly unreliable." (NISO, November 1, 2023)
Bringing Generative AI to the Web of Science: Clarivate is layering generative AI into Web of Science via their new AI Research Assistant, which will be available for beta testing in December 2023. Researchers will be able to ask questions and uncover answers based on Web of Science data, browse concise summaries of articles and results sets, and have conversations to speed up their research. (Clarivate, November 7, 2023)
‘ChatGPT detector’ catches AI-generated papers with unprecedented accuracy: A recent study published in Cell Reports Physical Science found that a new machine-learning tool can easily detect when chemistry papers are written using ChatGPT. This tool outperformed two already existing "ChatGPT" detectors, and could help academic publishers in identifying papers created by AI text generators. (Nature [paywall], November 6, 2023)
Large Language Model Advanced Data Analysis Abuse to Create a Fake Data Set in Medical Research: In the latest issue of JAMA Ophthalmology, a research letter reports on an experiment that tested the ability of an augmented version of GPT to generate a fake dataset suitable for scientific research. This version has the capability to perform statistical analysis and data visualization. (JAMA Ophthalmology, November 9, 2023)
The United States Copyright Office Notice of Inquiry on AI: A Quick Take: The CCC published its responses to the United States Copyright Office's request for comments on the topic of AI and copyright. The responses include high-level takes from the CCC on fair use, including their legal reasoning for the argument that AI models being trained on copyrighted content without the rights-holders' consent constitutes an infringement of copyright (rather than fair use, argued by some technologists). (The Scholarly Kitchen, November 28, 2023)
DEI, Ethics, Safety, + Privacy
Inside the Underground World of Black Market AI Chatbots: New bots are coming out that are designed to help cybercriminals get their jobs done. This article goes into some detail about how these nefarious chatbots work and what it could mean for the future of phishing and information security. (The Daily Beast, October 21, 2023)
Framework for Identifying Highly Consequential AI Use Cases: The Special Competitive Studies Project (SCSP) in collaboration with the John Hopkins University Applied Physics Laboratory (JHUAPL) developed the “Framework for Identifying Highly Consequential AI Use Cases.” It's intended to help regulators cut through the noise and objectively determine whether an AI use case is worth regulating or not. While the report doesn't suggest what kinds of regulatory actions should be taken for high impact AI use cases, it does make a few broad recommendations for how regulation should work, including sector-specific regulation and robust governance to fill in the gaps for non-critical use cases. (SCSP, November 7, 2023)
The Foundation Model Transparency Index: A comprehensive assessment of the transparency of foundation model developers. The top-scoring model scores only 54 out of 100. No major foundation model developer is close to providing adequate transparency, revealing a fundamental lack of transparency in the AI industry. The mean score is a just 37%. Yet, 82 of the indicators are satisfied by at least one developer, meaning that developers can significantly improve transparency by adopting best practices from their competitors. (Center for Research on Foundation Models, November 1, 2023)
Common Sense Media Launches First-Ever AI Ratings System: Common Sense Media released an AI ratings system that evaluates AI tools against a set of "Common Sense AI Principles." Their goal is to help the public to understand where these tools are using best practices, where they may compromise human rights or data privacy, and assess their potential to perpetuate misinformation and unfair bias. Notably, ChatGPT and Bard only got 3 out of 5 stars. (Common Sense Media, November 16, 2023)
Technology
New tools help artists fight AI by directly disrupting the systems: In addition to Nightshade, other tools are coming out to help artists combat plagiarism of their work, including Glaze and Kudurru. While Glaze is somewhat similar to Nightshade ("poisoning" LLMs to create confused outputs), Kudurru is a bit different: it tracks scrapers' IP addresses and blocks them or sends back unwanted content, such as an extended middle finger, or the classic "Rickroll" Internet trolling prank that spams unsuspecting users with a the music video for British singer Rick Astley's 1980s pop hit, "Never Gonna Give You Up." (NPR, November 3, 2023)
OpenAI Data Partnerships: OpenAI introduced OpenAI Data Partnerships, where they aim to collaborate with organizations to produce public and private datasets for training AI models. They're interested in large-scale and accessible datasets that "reflect human society" and "express human intention." They also state that they can work with any data modality, including text, images, audio, or video, and that they can use their in-house AI technology (including OCR and ASR) to help digitize, structure, and clean data. (OpenAI, November 9, 2023)
Legal & Politics
AI companies have all kinds of arguments against paying for copyrighted content: The responses to the US Copyright Office's call for public comment on potential new rules around generative AI’s use of copyrighted materials from Meta, Google, Microsoft, Adobe, Hugging Face, StabilityAI, Anthropic, and Apple have an underlying theme: they don't think they should have to pay for using copyrighted content. (The Verge, November 4, 2023)
Thune, Klobuchar release bipartisan AI bill: Senators John Thune (R-S.D.) and Amy Klobuchar (D-Minn.) introduced an artificial intelligence (AI) bill called Artificial Intelligence Research, Innovation and Accountability Act. This bill would require federal agencies to create standards aimed at providing transparency, accountability, content provenance, and detection frameworks for AI tools. (The Hill, November 15, 2023)
US, UK and a dozen more countries unveil pact to make AI ‘secure by design’: More than a dozen countries, including the UK and US, released a detailed international agreement on keeping artificial intelligence safe from rogue actors, pushing companies to create AI systems that are “secure by design”. The document stresses that AI needs to be developed and deployed to keep customers and the wider public safe from misuse. It's important to note, though, that the agreement is non-binding and carries mostly general recommendations such as monitoring AI systems for abuse, protecting data from tampering, and vetting software suppliers. (The Guardian, November 27, 2023)
The Bletchley Declaration by Countries Attending the AI Safety Summit: The "Bletchley Declaration" to reduce AI risks across borders was signed on October 31, 2023, by the U.S., EU, China, India, Japan, and nearly 20 other governments. The declaration lays out a shared intent to cooperate internationally on identifying AI risks, creating international risk policies to ensure safety, encouraging transparency, and creating a network for scientific research on AI safety that is aimed at policy-making and the public good. (Gov.uk, November 1, 2023)
Behind the Curtain: Myth of AI restraint: "Lots of people want to roll artificial intelligence out slowly, use it ethically, regulate it wisely. But everyone gets the joke: It defies all human logic and experience to think ethics will trump profit, power, prestige. Never has. Never will." This article suggests that there'll likely be no serious regulation of generative AI, which one day soon could spawn artificial general intelligence (AGI) — the one that could outthink our species." (Axios, November 21, 2023)
The 6 Types of Conversations with Generative AI: This article drills down into 6 key types of conversations that you can have with an AI chatbot: Search queries, funneling conversations, exploring conversations, chiseling conversations, expanding conversations, and pinpointing conversations. It also includes tips for users to have the most effective conversation depending on what you're looking for, such as not using ChatGPT for a search query, as well as for those designing AI bots. (Norman Nielsen Group, November 10, 2023)
