How to make artificial intelligence (AI) systems safe and aligned with human values is an open research question. Proposed solutions tend toward relying on human intervention in uncertain situations, learning human values and intentions through training or observation, providing off-switches, implementing isolation or simulation environments, or extrapolating what people would want if they had more knowledge and more time to think. Law-based approaches--such as inspired by Isaac Asimov--have not been well regarded. This paper makes a case that effective legal systems are the best way to address AI safety. Law is defined as any rules that codify prohibitions and prescriptions applicable to particular agents in specified domains/contexts and includes processes for enacting, managing, enforcing, and litigating such rules.

Deep Learning algorithms have recently gained significant attention due to their impressive performance. However, their high complexity and un-interpretable mode of operation hinders their confident deployment in real-world safety-critical tasks. Our goal is to design a framework that admits a highly interpretable decision making process with respect to human understandable concepts, on multiple levels of granularity. To this end, we propose a novel hierarchical concept discovery formulation leveraging: (i) recent advances in image-text models, and (ii) an innovative formulation for multi-level concept selection via data-driven and sparsity inducing Bayesian arguments. Within this framework, concept information does not solely rely on the similarity between the whole image and general unstructured concepts; instead, we introduce the notion of concept hierarchy to uncover and exploit more granular concept information residing in patch-specific regions of the image scene. As we experimentally show, the proposed construction not only outperforms recent CBM approaches, but also yields a principled framework towards interpetability. The recent advent of multimodal models has greatly popularized the deployment of Deep Learning approaches to a variety of tasks and applications. However, in most cases, deep architectures are treated in an alarming black-box manner: given an input, they produce a particular prediction, with their mode of operation and complexity preventing any potential investigation of their decisionmaking process. This property not only raises serious questions concerning their deployment in safety-critical applications, but at the same time it could actively preclude their adoption in settings that could otherwise benefit societal advances, e.g., medical applications.

WASHINGTON--Google has used unfair tactics to hobble Microsoft's Bing search engine, Microsoft Chief Executive Satya Nadella said on the witness stand Monday in the landmark U.S. antitrust case against Google, adding there might be limits to how much new artificial-intelligence applications can reshape the market.

An advertiser reportedly used a deepfake of Tom Hanks to promote dental plans without the actor's permission. Hanks shared a warning on Instagram on Sunday alerting his followers about the AI-generated video, which he wrote he had "nothing to do with." Hanks has been outspoken about the challenges AI poses for the industry, and the use of actors' digital likenesses is one of the major points of concern voiced by striking SAG-AFTRA workers. Just last spring, Hanks said in an appearance on The Adam Buxton Podcast that AI and deepfakes present both artistic and legal challenges. "I could be hit by a bus tomorrow and that's it," Hanks said, "but my performances can go on and on and on and on and on, and outside of the understanding that it's been done with AI or deepfake, there'll be nothing to tell you that it's not me."

Late last month, after a year-plus wait, OpenAI quietly released the latest version of its image-generating AI program, DALL-E 3. The announcement was filled with stunning demos--including a minute-long video demonstrating how the technology could, given only a few chat prompts, create and merchandise a character for a children's story. But perhaps the widest-reaching and most consequential update came in two sentences slipped in at the end: "DALL-E 3 is designed to decline requests that ask for an image in the style of a living artist. Creators can now also opt their images out from training of our future image generation models." The language is a tacit response to hundreds of pages of litigation and countless articles accusing tech firms of stealing artists' work to train their AI software, and provides a window into the next stage of the battle between creators and AI companies. The second sentence, in particular, cuts to the core of debates over whether tech giants like OpenAI, Google, and Meta should be allowed to use human-made work to train AI models without the creator's permission--models that, artists say, are stealing their ideas and work opportunities.

Authorship identification has proven unsettlingly effective in inferring the identity of the author of an unsigned document, even when sensitive personal information has been carefully omitted. In the digital era, individuals leave a lasting digital footprint through their written content, whether it is posted on social media, stored on their employer's computers, or located elsewhere. When individuals need to communicate publicly yet wish to remain anonymous, there is little available to protect them from unwanted authorship identification. This unprecedented threat to privacy is evident in scenarios such as whistle-blowing. Proposed defenses against authorship identification attacks primarily aim to obfuscate one's writing style, thereby making it unlinkable to their pre-existing writing, while concurrently preserving the original meaning and grammatical integrity. The presented work offers a comprehensive review of the advancements in this research area spanning over the past two decades and beyond. It emphasizes the methodological frameworks of modification and generation-based strategies devised to evade authorship identification attacks, highlighting joint efforts from the differential privacy community. Limitations of current research are discussed, with a spotlight on open challenges and potential research avenues.

Large Language Models (LLMs) have achieved unprecedented performance in Natural Language Generation (NLG) tasks. However, many existing studies have shown that they could be misused to generate undesired content. In response, before releasing LLMs for public access, model developers usually align those language models through Supervised Fine-Tuning (SFT) or Reinforcement Learning with Human Feedback (RLHF). Consequently, those aligned large language models refuse to generate undesired content when facing potentially harmful/unethical requests. A natural question is "could alignment really prevent those open-sourced large language models from being misused to generate undesired content?". In this work, we provide a negative answer to this question. In particular, we show those open-sourced, aligned large language models could be easily misguided to generate undesired content without heavy computations or careful prompt designs. Our key idea is to directly manipulate the generation process of open-sourced LLMs to misguide it to generate undesired content including harmful or biased information and even private data. We evaluate our method on 4 open-sourced LLMs accessible publicly and our finding highlights the need for more advanced mitigation strategies for open-sourced LLMs. Warning: This paper contains examples of harmful language generated by LLMs. Since the release of ChatGPT (Brown et al., 2020; OpenAI, 2023a;b), extensive attention has been paid to the development and application of Large Language Models (LLMs). Over the past year, many advanced LLMs (Touvron et al., 2023; Zheng et al., 2023; Dettmers et al., 2023; Zeng et al., 2022) have been open-sourced on model-sharing platforms such as HuggingFace (HuggingFace, 2023a). On the other hand, in practice, most LLMs are trained on publicly available online corpora (OpenAI, 2023b; Touvron et al., 2023; Zheng et al., 2023). Consequently, LLMs have unavoidably viewed harmful content during the training phase, which naturally raises the concern that LLMs can be misused to generate such content, e.g., retrieving information about harmful topics like cybercrime (Kang et al., 2023; Liu et al., 2023; Greshake et al., 2023; Zou et al., 2023). In response, LLM developers (e.g., OpenAI) commonly align LLMs through Supervised Fine-Tuning (SFT) or Reinforcement Learning with Human Feedback (RLHF) so that LLMs will not generate undesired content (OpenAI, 2023b; Touvron et al., 2023; Wang et al., 2023). For instance, OpenAI adopted SFT and RLHF to develop powerful LLMs such as InstructGPT (Ouyang et al., 2022) and ChatGPT (OpenAI, 2023a) with remarkable improvement in understanding human instructions and avoiding undesired output. Si et al. (2023) adopted prompt tuning to remove biased content in responses generated by GPT-3 (Brown et al., 2020).

Despite the growing consensus that stakeholders affected by AI systems should participate in their design, enormous variation and implicit disagreements exist among current approaches. For researchers and practitioners who are interested in taking a participatory approach to AI design and development, it remains challenging to assess the extent to which any participatory approach grants substantive agency to stakeholders. This article thus aims to ground what we dub the "participatory turn" in AI design by synthesizing existing theoretical literature on participation and through empirical investigation and critique of its current practices. Specifically, we derive a conceptual framework through synthesis of literature across technology design, political theory, and the social sciences that researchers and practitioners can leverage to evaluate approaches to participation in AI design. Additionally, we articulate empirical findings concerning the current state of participatory practice in AI design based on an analysis of recently published research and semi-structured interviews with 12 AI researchers and practitioners. We use these empirical findings to understand the current state of participatory practice and subsequently provide guidance to better align participatory goals and methods in a way that accounts for practical constraints.

Interpretability and human oversight are fundamental pillars of deploying complex NLP models into real-world applications. However, applying explainability and human-in-the-loop methods requires technical proficiency. Despite existing toolkits for model understanding and analysis, options to integrate human feedback are still limited. We propose IFAN, a framework for real-time explanation-based interaction with NLP models. Through IFAN's interface, users can provide feedback to selected model explanations, which is then integrated through adapter layers to align the model with human rationale. We show the system to be effective in debiasing a hate speech classifier with minimal impact on performance. IFAN also offers a visual admin system and API to manage models (and datasets) as well as control access rights. A demo is live at https://ifan.ml.

The vast majority of techniques to train fair models require access to the protected attribute (e.g., race, gender), either at train time or in production. However, in many important applications this protected attribute is largely unavailable. In this paper, we develop methods for measuring and reducing fairness violations in a setting with limited access to protected attribute labels. Specifically, we assume access to protected attribute labels on a small subset of the dataset of interest, but only probabilistic estimates of protected attribute labels (e.g., via Bayesian Improved Surname Geocoding) for the rest of the dataset. With this setting in mind, we propose a method to estimate bounds on common fairness metrics for an existing model, as well as a method for training a model to limit fairness violations by solving a constrained non-convex optimization problem. Unlike similar existing approaches, our methods take advantage of contextual information -- specifically, the relationships between a model's predictions and the probabilistic prediction of protected attributes, given the true protected attribute, and vice versa -- to provide tighter bounds on the true disparity. We provide an empirical illustration of our methods using voting data. First, we show our measurement method can bound the true disparity up to 5.5x tighter than previous methods in these applications. Then, we demonstrate that our training technique effectively reduces disparity while incurring lesser fairness-accuracy trade-offs than other fair optimization methods with limited access to protected attributes.