Although current deep models for face tasks surpass human performance on some benchmarks, we do not understand how they work. Thus, we cannot predict how it will react to novel inputs, resulting in catastrophic failures and unwanted biases in the algorithms. Explainable AI helps bridge the gap, but currently, there are very few visualization algorithms designed for faces. This work undertakes a first-of-its-kind meta-analysis of explainability algorithms in the face domain. We explore the nuances and caveats of adapting general-purpose visualization algorithms to the face domain, illustrated by computing visualizations on popular face models. We review existing face explainability works and reveal valuable insights into the structure and hierarchy of face networks. We also determine the design considerations for practical face visualizations accessible to AI practitioners by conducting a user study on the utility of various explainability algorithms.

We consider the problem of learning from data corrupted by underrepresentation bias, where positive examples are filtered from the data at different, unknown rates for a fixed number of sensitive groups. We show that with a small amount of unbiased data, we can efficiently estimate the group-wise drop-out parameters, even in settings where intersectional group membership makes learning each intersectional rate computationally infeasible. Using this estimate for the group-wise drop-out rate, we construct a re-weighting scheme that allows us to approximate the loss of any hypothesis on the true distribution, even if we only observe the empirical error on a biased sample. Finally, we present an algorithm encapsulating this learning and re-weighting process, and we provide strong PAC-style guarantees that, with high probability, our estimate of the risk of the hypothesis over the true distribution will be arbitrarily close to the true risk.

As part of the deal, Amazon, the world's largest provider of cloud infrastructure services through its AWS unit, will become the primary provider of computational processing power, also called compute, for Anthropic. The process of training and running state-of-the-art AI models requires vast amounts of compute, and many analysts expect future AI models to require increasing amounts of compute. In return, Amazon will acquire a minority ownership position in Anthropic, and Amazon's engineers will be able to incorporate Anthropic's AI models into their products and services such as Amazon's personal assistant, Alexa. Anthropic has also committed to offering its models via Bedrock, Amazon's online platform on which it hosts foundation models--broadly capable AI models that can be adapted for different tasks. Anthropic was founded in 2021, after a group of OpenAI employees left over differences in their approach to AI safety.

The US Federal Bureau of Investigation (FBI) has done tens of thousands of face recognition searches using software from outside providers in recent years. Yet only 5 percent of the 200 agents with access to the technology have taken the bureau's three-day training course on how to use it, a report from the Government Accountability Office (GAO) this month reveals. The bureau has no policy for face recognition use in place to protect privacy, civil rights, or civil liberties. Lawmakers and others concerned about face recognition have said that adequate training on the technology and how to interpret its output is needed to reduce improper use or errors, although some experts say training can lull law enforcement and the public into thinking face recognition is low risk. Since the false arrest of Robert Williams near Detroit in 2020, multiple instances have surfaced in the US of arrests after a face recognition model wrongly identified a person.

Editor's note: This article is part of The Atlantic's series on Books3. You can search the database for yourself here, and read about its origins here. This summer, I reported on a data set of more than 191,000 books that were used without permission to train generative-AI systems by Meta, Bloomberg, and others. "Books3," as it's called, was based on a collection of pirated ebooks that includes travel guides, self-published erotic fiction, novels by Stephen King and Margaret Atwood, and a lot more. Books play a crucial role in the training of generative-AI systems.

Editor's note: This searchable database is part of The Atlantic's series on Books3. You can read about the origins of the database here, and an analysis of what's in it here. This summer, I acquired a data set of more than 191,000 books that were used without permission to train generative-AI systems by Meta, Bloomberg, and others. I wrote in The Atlantic about how the data set, known as "Books3," was based on a collection of pirated ebooks, most of them published in the past 20 years. Since my article appeared, I've heard from several authors wanting to know if their work is in Books3.

"Fundamentally, it's trained; it's clean. It's viable for businesses to use. We'll stand behind that claim," says Craig Peters, the CEO of Getty Images. The past year has seen a boom in generative AI systems that produce images and text. Earlier this year, Getty Images announced it was suing Stability AI for using millions of its images, without permission, to train its open-source image-generation AI Stable Diffusion.

The AI Insight Forums were announced a few months ago by Senate Majority Leader Chuck Schumer as part of his "SAFE Innovation" initiative, which is really a set of principles for AI legislation in the United States. The invite list was heavily skewed toward Big Tech execs, including CEOs of AI companies, though a few civil society and AI ethics researchers were included too. Coverage of the meeting thus far has put a particular emphasis on the reportedly unanimous agreement about the need for AI regulation, and on issues raised by Elon Musk and others about the "civilizational risks" created by AI. (This tracker from Tech Policy Press is pretty handy if you want to know more.) But to really dig below the surface, I caught up with one of the other attendees, Inioluwa Deborah Raji, who gave me an inside look at how the first meeting went, the pernicious myths she needed to debunk, and where disagreements could be felt in the room. Raji is a researcher at the University of California, Berkeley, and a fellow at Mozilla.

Segmenting long videos into chapters enables users to quickly navigate to the information of their interest. This important topic has been understudied due to the lack of publicly released datasets. To address this issue, we present VidChapters-7M, a dataset of 817K user-chaptered videos including 7M chapters in total. VidChapters-7M is automatically created from videos online in a scalable manner by scraping user-annotated chapters and hence without any additional manual annotation. We introduce the following three tasks based on this data. First, the video chapter generation task consists of temporally segmenting the video and generating a chapter title for each segment. To further dissect the problem, we also define two variants of this task: video chapter generation given ground-truth boundaries, which requires generating a chapter title given an annotated video segment, and video chapter grounding, which requires temporally localizing a chapter given its annotated title. We benchmark both simple baselines and state-of-the-art video-language models for these three tasks. We also show that pretraining on VidChapters-7M transfers well to dense video captioning tasks in both zero-shot and finetuning settings, largely improving the state of the art on the YouCook2 and ViTT benchmarks. Finally, our experiments reveal that downstream performance scales well with the size of the pretraining dataset. Our dataset, code, and models are publicly available at https://antoyang.github.io/vidchapters.html.

Employers are adopting algorithmic hiring technology throughout the recruitment pipeline. Algorithmic fairness is especially applicable in this domain due to its high stakes and structural inequalities. Unfortunately, most work in this space provides partial treatment, often constrained by two competing narratives, optimistically focused on replacing biased recruiter decisions or pessimistically pointing to the automation of discrimination. Whether, and more importantly what types of, algorithmic hiring can be less biased and more beneficial to society than low-tech alternatives currently remains unanswered, to the detriment of trustworthiness. This multidisciplinary survey caters to practitioners and researchers with a balanced and integrated coverage of systems, biases, measures, mitigation strategies, datasets, and legal aspects of algorithmic hiring and fairness. Our work supports a contextualized understanding and governance of this technology by highlighting current opportunities and limitations, providing recommendations for future work to ensure shared benefits for all stakeholders.