One of the most important things you need to know about AI’s “ethics” is that it relates to value. Ethics doesn’t teach the right thing, and it offers a vocabulary of values - transparency, safety, justice – and a framework for prioritizing between them. This summer, we were able to understand AI values to lawmakers in the UK and the US and shape the future of AI regulation. This is where ethics shines. It’s about helping to pave the way forward when the law is not yet in place.
In line with embracing the core values of facial openness and accountability, we share a collection of what we said and done. This includes CEO Clem’s testimony to the US Congress and a statement at the US Senate AI Insight Forum. Advice on EU AI Law. Comments to NTIA on AI accountability. And our leading ethics scientist, Meg’s comments on the Democratic Caucus. What is common to many of these discussions is the question of why AI openness is beneficial, and I share a collection of answers to this question here.
In line with the core values of democratization, we also spent a lot of our time talking publicly, and we were able to speak with journalists to explain what is happening in the AI world today. This includes:
Comments from Sasha on AI’s energy use and carbon emissions (Atlantic, Guardian, (two), New Scientist, Weather Network, Wall Street Journal, (two)), and part of the Wall Street Journal Op-ed on topics. AI Doomsday Risk Ideas (Bloomberg, Times, Futurism, Sky News); Generated AI Bias Details (Bloomberg, NBC, VOX); Address how marginalized workers create AI (Glove and Mail, Atlantic) data. The emphasis on sexism in AI (supervisor). Provides insights from MIT technology reviews on AI text detection, open model releases, and AI transparency. Comments from Nathan on cutting-edge art on language models and open releases (Wired, VentureBeat, Business Insider, Fortune). Comments from Meg on AI and misinformation (CNN, Al Jazeera, The New York Times); The need to deal with artists’ work in the AI (Washington Post). The relationship between advances in AI produced and greater benefits (Washington Post, Venture Beat); How journalists report (CJR) better shape the evolution of AI. An explanation of the basic statistical concepts of AI (ARS Technica) confusion. Emphasis on patterns of sexism (fast companies). Comment from Irene on understanding the regulatory environment of AI (MIT Technology Review, Barron’s). Comments from Yacine on open source and AI laws (VentureBeat, Time) and copyright issues (VentureBeat). A comment from Giada on the concept of AI’s “singularity” (popular mechanics) and AI’s “sentience” (RFI, Radio France). Ideas on the dangers of artificial romance (Analytics India Magazine); description of value integrity (Hindus).
Our talk, released this summer, includes Giada’s TED presentation on whether “ethical” generator AI is possible (subtitles in automatic English translations are great!). Yacine’s presentation on Markkura Center’s Technology Ethics is open for applied ethics and responsible openness in the workshop on responsible open foundation models. Katie’s chat about Generating AI in Health. London Data Week presentations build open and better AI.
Of course, we are also making progress on regular work (“work at work”). As we have focused on how to shape AI in ways that are informed by society and human values that everyone feels welcome, the fundamental value of accessibility has emerged to our entire work. This includes new courses on AI audio, such as Maria. Katie’s resources on open access clinical language models. Naznian and other tutorials on responsible generation AI. Generated AI release gradients (video) and our constraint papers (video) on the clarification of ethical charters, legal tools, and technical documents of ML. Workshops on mapping risk surfaces of intertext AI with participatory, interdisciplinary approaches and workshops on assessing the impact of modalities and society-wide generation AI systems (video).
We also recently advanced the goals of fairness and justice with a bias and harm test applied to the new Hug multimodal model IDEFICS. We have worked on ways to operate responsibly and responsibly, including updating content policies (GIADA takes the lead). Using machine learning to improve metadata (Daniel’s lead), we advanced support for AI rigor by adding more descriptive statistics to our dataset (Daniel’s lead) to support the diversity of languages in the hub and to better understand what AI has learned and how it is evaluated.
From experiences from this past season, we currently offer a collection of many resources that are particularly useful in the current AI Ethics discourse.
Finally, we are amazed and pleased by the public perception of many of the social and ethics regulars, including the selection of both Irene and Sasha in the 35 innovators under the age of 35 (holding the ¼ of AI 35 under the age of 35!) Meg is included in the list of influential AI innovators (Wired, Fortune). Select less than 100 with Meg and Clem AI. It’s also very sad to say goodbye to my colleague Nathan. He has contributed to the work of linking ethics to reinforcement learning in AI systems. As a gift for his farewell, he provided further details on the challenges of operating ethical AI at RLHF.
Thank you for reading!
– Meg, hugging face ethics and social regulars
