Search

Two substantial technological advances have reshaped the public square in recent decades: first with the advent of the internet and second with the recent introduction of large language models (LLMs). LLMs offer opportunities for a paradigm shift towards more decentralized, participatory online spaces that can be used to facilitate deliberative dialogues at scale, but also create risks of exacerbating societal schisms. Here, we explore four applications of LLMs to improve digital public squares: collective dialogue systems, bridging systems, community moderation, and proof-of-humanity systems. Building on the input from over 70 civil society experts and technologists, we argue that LLMs both afford promising opportunities to shift the paradigm for conversations at scale and pose distinct risks for digital public squares. We lay out an agenda for future research and investments in AI that will strengthen digital public squares and safeguard against potential misuses of AI., Comment: 40 pages, 5 figures

Describing skills in natural language has the potential to provide an accessible way to inject human knowledge about decision-making into an AI system. We present MaestroMotif, a method for AI-assisted skill design, which yields high-performing and adaptable agents. MaestroMotif leverages the capabilities of Large Language Models (LLMs) to effectively create and reuse skills. It first uses an LLM's feedback to automatically design rewards corresponding to each skill, starting from their natural language description. Then, it employs an LLM's code generation abilities, together with reinforcement learning, for training the skills and combining them to implement complex behaviors specified in language. We evaluate MaestroMotif using a suite of complex tasks in the NetHack Learning Environment (NLE), demonstrating that it surpasses existing approaches in both performance and usability.

Rewards remain an uninterpretable way to specify tasks for Reinforcement Learning, as humans are often unable to predict the optimal behavior of any given reward function, leading to poor reward design and reward hacking. Language presents an appealing way to communicate intent to agents and bypass reward design, but prior efforts to do so have been limited by costly and unscalable labeling efforts. In this work, we propose a method for a completely unsupervised alternative to grounding language instructions in a zero-shot manner to obtain policies. We present a solution that takes the form of imagine, project, and imitate: The agent imagines the observation sequence corresponding to the language description of a task, projects the imagined sequence to our target domain, and grounds it to a policy. Video-language models allow us to imagine task descriptions that leverage knowledge of tasks learned from internet-scale video-text mappings. The challenge remains to ground these generations to a policy. In this work, we show that we can achieve a zero-shot language-to-behavior policy by first grounding the imagined sequences in real observations of an unsupervised RL agent and using a closed-form solution to imitation learning that allows the RL agent to mimic the grounded observations. Our method, RLZero, is the first to our knowledge to show zero-shot language to behavior generation abilities without any supervision on a variety of tasks on simulated domains. We further show that RLZero can also generate policies zero-shot from cross-embodied videos such as those scraped from YouTube., Comment: 27 pages

Having explored an environment, intelligent agents should be able to transfer their knowledge to most downstream tasks within that environment. Referred to as "zero-shot learning," this ability remains elusive for general-purpose reinforcement learning algorithms. While recent works have attempted to produce zero-shot RL agents, they make assumptions about the nature of the tasks or the structure of the MDP. We present \emph{Proto Successor Measure}: the basis set for all possible solutions of Reinforcement Learning in a dynamical system. We provably show that any possible policy can be represented using an affine combination of these policy independent basis functions. Given a reward function at test time, we simply need to find the right set of linear weights to combine these basis corresponding to the optimal policy. We derive a practical algorithm to learn these basis functions using only interaction data from the environment and show that our approach can produce the optimal policy at test time for any given reward function without additional environmental interactions. Project page: https://agarwalsiddhant10.github.io/projects/psm.html., Comment: Under submission, 23 pages

Responding to misinformation online can be an exhausting and thankless task. It takes time and energy to write effective content, puts users at risk of online harassment, and strains personal relationships. Despite these challenges, there are people who voluntarily respond to misinformation online, and some have established communities on platforms such as Reddit, Discord, and X (formerly Twitter) dedicated to these efforts. In this work, we interviewed 8 people who participate in such communities to understand the type of support they receive from each other in these discussion spaces. Interviewees described that their communities helped them sustain motivation, save time, and improve their communication skills. Common practices included sharing sources and citations, providing emotional support, giving others advice, and signaling positive feedback. We present our findings as three case studies and discuss opportunities for future work to support collaborative practices in online communities dedicated to misinformation response. Our work surfaces how resource sharing, social motivation, and decentralization can make misinformation correction more sustainable, rewarding, and effective for online citizens.

Language models trained on diverse datasets unlock generalization by in-context learning. Reinforcement Learning (RL) policies can achieve a similar effect by meta-learning within the memory of a sequence model. However, meta-RL research primarily focuses on adapting to minor variations of a single task. It is difficult to scale towards more general behavior without confronting challenges in multi-task optimization, and few solutions are compatible with meta-RL's goal of learning from large training sets of unlabeled tasks. To address this challenge, we revisit the idea that multi-task RL is bottlenecked by imbalanced training losses created by uneven return scales across different tasks. We build upon recent advancements in Transformer-based (in-context) meta-RL and evaluate a simple yet scalable solution where both an agent's actor and critic objectives are converted to classification terms that decouple optimization from the current scale of returns. Large-scale comparisons in Meta-World ML45, Multi-Game Procgen, Multi-Task POPGym, Multi-Game Atari, and BabyAI find that this design unlocks significant progress in online multi-task adaptation and memory problems without explicit task labels., Comment: NeurIPS 2024

Alignment of large language models (LLMs) to societal values should account for pluralistic values from diverse groups. One technique uses in-context learning for inference-time alignment, but only considers similarity when drawing few-shot examples, not accounting for cross-group differences in value prioritization. We propose SPICA, a framework for pluralistic alignment that accounts for group-level differences during in-context example retrieval. SPICA introduces three designs to facilitate pluralistic alignment: scenario banks, group-informed metrics, and in-context alignment prompts. From an evaluation of SPICA on an alignment task collecting inputs from four demographic groups ($n = 544$), our metrics retrieve in-context examples that more closely match observed preferences, with the best prompt configuration using multiple contrastive responses to demonstrate examples. In an end-to-end evaluation ($n = 80$), we observe that SPICA-aligned models are higher rated than a baseline similarity-only retrieval approach, with groups seeing up to a +0.16 point improvement on a 5 point scale. Additionally, gains from SPICA were more uniform, with all groups benefiting from alignment rather than only some. Finally, we find that while a group-agnostic approach can effectively align to aggregated values, it is not most suited for aligning to divergent groups.

We introduce a method to measure the alignment between public will and language model (LM) behavior that can be applied to fine-tuning, online oversight, and pre-release safety checks. Our `chain of alignment' (CoA) approach produces a rule based reward (RBR) by creating model behavior $\textit{rules}$ aligned to normative $\textit{objectives}$ aligned to $\textit{public will}$. This factoring enables a nonexpert public to directly specify their will through the normative objectives, while expert intelligence is used to figure out rules entailing model behavior that best achieves those objectives. We validate our approach by applying it across three different domains of LM prompts related to mental health. We demonstrate a public input process built on collective dialogues and bridging-based ranking that reliably produces normative objectives supported by at least $96\% \pm 2\%$ of the US public. We then show that rules developed by mental health experts to achieve those objectives enable a RBR that evaluates an LM response's alignment with the objectives similarly to human experts (Pearson's $r=0.841$, $AUC=0.964$). By measuring alignment with objectives that have near unanimous public support, these CoA RBRs provide an approximate measure of alignment between LM behavior and public will., Comment: Pluralistic Alignment Workshop at NeurIPS 2024

Creative writers have a love for their craft, yet AI systems using large language models (LLMs) offer the automation of significant parts of the writing process. So why do some creative writers choose to integrate AI into their workflows? To explore this, we interview and observe a writing session with 18 creative writers who already use AI regularly in their writing practice. Our findings reveal that creative writers are intentional about how they incorporate AI, making many deliberate decisions about when and how to engage AI based on the core values they hold about writing. These values, such as authenticity and craftsmanship, alongside writers' relationships with and use of AI influence the parts of writing over which they wish to maintain control. Through our analysis, we contribute a taxonomy of writer values, writer relationships with AI, and integration strategies, and discuss how these three elements interrelate.

AI agents are increasingly tasked with making proactive suggestions in online spaces where groups collaborate, but can be unhelpful or even annoying, due to not fitting the group's preferences or behaving in socially inappropriate ways. Fortunately, group spaces have a rich history of prior social interactions and affordances for social feedback to support creating agents that align to a group's interests and norms. We present Social-RAG, a workflow for grounding agents to social information about a group, which retrieves from prior group interactions, selects relevant social signals, and then feeds the context into a large language model to generate messages to the group. We implement this into PaperPing, our system that posts academic paper recommendations in group chat, leveraging social signals determined from formative studies with 39 researchers. From a three-month deployment in 18 channels, we observed PaperPing posted relevant messages in groups without disrupting their existing social practices, fostering group common ground.

The rise of large language models (LLMs) has led many researchers to consider their usage for scientific work. Some have found benefits using LLMs to augment or automate aspects of their research pipeline, while others have urged caution due to risks and ethical concerns. Yet little work has sought to quantify and characterize how researchers use LLMs and why. We present the first large-scale survey of 816 verified research article authors to understand how the research community leverages and perceives LLMs as research tools. We examine participants' self-reported LLM usage, finding that 81% of researchers have already incorporated LLMs into different aspects of their research workflow. We also find that traditionally disadvantaged groups in academia (non-White, junior, and non-native English speaking researchers) report higher LLM usage and perceived benefits, suggesting potential for improved research equity. However, women, non-binary, and senior researchers have greater ethical concerns, potentially hindering adoption., Comment: 30 pages, 5 figures

Automatically synthesizing dense rewards from natural language descriptions is a promising paradigm in reinforcement learning (RL), with applications to sparse reward problems, open-ended exploration, and hierarchical skill design. Recent works have made promising steps by exploiting the prior knowledge of large language models (LLMs). However, these approaches suffer from important limitations: they are either not scalable to problems requiring billions of environment samples; or are limited to reward functions expressible by compact code, which may require source code and have difficulty capturing nuanced semantics; or require a diverse offline dataset, which may not exist or be impossible to collect. In this work, we address these limitations through a combination of algorithmic and systems-level contributions. We propose ONI, a distributed architecture that simultaneously learns an RL policy and an intrinsic reward function using LLM feedback. Our approach annotates the agent's collected experience via an asynchronous LLM server, which is then distilled into an intrinsic reward model. We explore a range of algorithmic choices for reward modeling with varying complexity, including hashing, classification, and ranking models. By studying their relative tradeoffs, we shed light on questions regarding intrinsic reward design for sparse reward problems. Our approach achieves state-of-the-art performance across a range of challenging, sparse reward tasks from the NetHack Learning Environment in a simple unified process, solely using the agent's gathered experience, without requiring external datasets nor source code. We make our code available at \url{URL} (coming soon).

Unsupervised skill discovery carries the promise that an intelligent agent can learn reusable skills through autonomous, reward-free environment interaction. Existing unsupervised skill discovery methods learn skills by encouraging distinguishable behaviors that cover diverse states. However, in complex environments with many state factors (e.g., household environments with many objects), learning skills that cover all possible states is impossible, and naively encouraging state diversity often leads to simple skills that are not ideal for solving downstream tasks. This work introduces Skill Discovery from Local Dependencies (Skild), which leverages state factorization as a natural inductive bias to guide the skill learning process. The key intuition guiding Skild is that skills that induce diverse interactions between state factors are often more valuable for solving downstream tasks. To this end, Skild develops a novel skill learning objective that explicitly encourages the mastering of skills that effectively induce different interactions within an environment. We evaluate Skild in several domains with challenging, long-horizon sparse reward tasks including a realistic simulated household robot domain, where Skild successfully learns skills with clear semantic meaning and shows superior performance compared to existing unsupervised reinforcement learning methods that only maximize state coverage.

Human-Computer Interaction (HCI) and Computer Supported Collaborative Work (CSCW) have a longstanding tradition of interrogating the values that underlie systems in order to create novel and accessible experiences. In this work, we use a neurodiversity framing to examine how people with ways of thinking, speaking, and being that differ from normative assumptions are perceived by researchers seeking to study and design social computing systems for neurodivergent people. From a critical analysis of 84 publications systematically gathered across a decade of social computing research, we determine that research into social computing with neurodiverse participants is largely medicalized, adheres to historical stereotypes of neurodivergent children and their families, and is insensitive to the wide spectrum of neurodivergent people that are potential users of social technologies. When social computing systems designed for neurodivergent people rely upon a conception of disability that restricts expression for the sake of preserving existing norms surrounding social experience, the result is often simplistic and restrictive systems that prevent users from "being social" in a way that feels natural and enjoyable. We argue that a neurodiversity perspective informed by critical disability theory allows us to engage with alternative forms of sociality as meaningful and desirable rather than a deficit to be compensated for. We conclude by identifying opportunities for researchers to collaborate with neurodivergent users and their communities, including the creation of spectrum-conscious social systems and the embedding of double empathy into systems for more equitable design., Comment: 29 Pages, 5 Figures, Accepted to CSCW 2025

In order to train agents that can quickly adapt to new objectives or reward functions, efficient unsupervised representation learning in sequential decision-making environments can be important. Frameworks such as the Exogenous Block Markov Decision Process (Ex-BMDP) have been proposed to formalize this representation-learning problem (Efroni et al., 2022b). In the Ex-BMDP framework, the agent's high-dimensional observations of the environment have two latent factors: a controllable factor, which evolves deterministically within a small state space according to the agent's actions, and an exogenous factor, which represents time-correlated noise, and can be highly complex. The goal of the representation learning problem is to learn an encoder that maps from observations into the controllable latent space, as well as the dynamics of this space. Efroni et al. (2022b) has shown that this is possible with a sample complexity that depends only on the size of the controllable latent space, and not on the size of the noise factor. However, this prior work has focused on the episodic setting, where the controllable latent state resets to a specific start state after a finite horizon. By contrast, if the agent can only interact with the environment in a single continuous trajectory, prior works have not established sample-complexity bounds. We propose STEEL, the first provably sample-efficient algorithm for learning the controllable dynamics of an Ex-BMDP from a single trajectory, in the function approximation setting. STEEL has a sample complexity that depends only on the sizes of the controllable latent space and the encoder function class, and (at worst linearly) on the mixing time of the exogenous noise factor. We prove that STEEL is correct and sample-efficient, and demonstrate STEEL on two toy problems. Code is available at: https://github.com/midi-lab/steel.

Community and organizational policies are typically designed in a top-down, centralized fashion, with limited input from impacted stakeholders. This can result in policies that are misaligned with community needs or perceived as illegitimate. How can we support more collaborative, participatory approaches to policy design? In this paper, we present PolicyCraft, a system that structures collaborative policy design through case-grounded deliberation. Building on past research that highlights the value of concrete cases in establishing common ground, PolicyCraft supports users in collaboratively proposing, critiquing, and revising policies through discussion and voting on cases. A field study across two university courses showed that students using PolicyCraft reached greater consensus and developed better-supported course policies, compared with those using a baseline system that did not scaffold their use of concrete cases. Reflecting on our findings, we discuss opportunities for future HCI systems to help groups more effectively bridge between abstract policies and concrete cases.

Emerging efforts in AI alignment seek to broaden participation in shaping model behavior by eliciting and integrating collective input into a policy for model finetuning. While pluralistic, these processes are often linear and do not allow participating stakeholders to confirm whether potential outcomes of their contributions are indeed consistent with their intentions. Design prototyping has long advocated for rapid iteration using tight feedback loops of ideation, experimentation, and evaluation to mitigate these issues. We thus propose policy prototyping for LLMs, a new process that draws inspiration from prototyping practices to enable stakeholders to collaboratively and interactively draft LLM policies. Through learnings from a real-world LLM policymaking initiative at an industrial AI lab, we motivate our approach and characterize policy prototyping with four guiding principles. Because policy prototyping emphasizes a contrasting set of priorities compared to previous approaches, we envision our approach to be a valuable addition to the methodological repertoire for pluralistic alignment.

Existing tools for laypeople to create personal classifiers often assume a motivated user working uninterrupted in a single, lengthy session. However, users tend to engage with social media casually, with many short sessions on an ongoing, daily basis. To make creating personal classifiers for content curation easier for such users, tools should support rapid initialization and iterative refinement. In this work, we compare three strategies -- (1) example labeling, (2) rule writing, and (3) large language model (LLM) prompting -- for end users to build personal content classifiers. From an experiment with 37 non-programmers tasked with creating personalized comment moderation filters, we found that with LLM prompting, participants reached 95\% of peak performance in 5 minutes, beating other strategies due to higher recall, but all strategies struggled with iterative refinement. Despite LLM prompting's better performance, participants preferred different strategies in different contexts and, even when prompting, provided examples or wrote rule-like prompts, suggesting hybrid approaches.

Transfer learning in reinforcement learning (RL) has become a pivotal strategy for improving data efficiency in new, unseen tasks by utilizing knowledge from previously learned tasks. This approach is especially beneficial in real-world deployment scenarios where computational resources are constrained and agents must adapt rapidly to novel environments. However, current state-of-the-art methods often fall short in ensuring safety during the transfer process, particularly when unforeseen risks emerge in the deployment phase. In this work, we address these limitations by introducing a novel Caution-Aware Transfer Learning (CAT) framework. Unlike traditional approaches that limit risk considerations to mean-variance, we define "caution" as a more generalized and comprehensive notion of risk. Our core innovation lies in optimizing a weighted sum of reward return and caution-based on state-action occupancy measures-during the transfer process, allowing for a rich representation of diverse risk factors. To the best of our knowledge, this is the first work to explore the optimization of such a generalized risk notion within the context of transfer RL. Our contributions are threefold: (1) We propose a Caution-Aware Transfer (CAT) framework that evaluates source policies within the test environment and constructs a new policy that balances reward maximization and caution. (2) We derive theoretical sub-optimality bounds for our method, providing rigorous guarantees of its efficacy. (3) We empirically validate CAT, demonstrating that it consistently outperforms existing methods by delivering safer policies under varying risk conditions in the test tasks.

One important property of DIstribution Correction Estimation (DICE) methods is that the solution is the optimal stationary distribution ratio between the optimized and data collection policy. In this work, we show that DICE-based methods can be viewed as a transformation from the behavior distribution to the optimal policy distribution. Based on this, we propose a novel approach, Diffusion-DICE, that directly performs this transformation using diffusion models. We find that the optimal policy's score function can be decomposed into two terms: the behavior policy's score function and the gradient of a guidance term which depends on the optimal distribution ratio. The first term can be obtained from a diffusion model trained on the dataset and we propose an in-sample learning objective to learn the second term. Due to the multi-modality contained in the optimal policy distribution, the transformation in Diffusion-DICE may guide towards those local-optimal modes. We thus generate a few candidate actions and carefully select from them to approach global-optimum. Different from all other diffusion-based offline RL methods, the guide-then-select paradigm in Diffusion-DICE only uses in-sample actions for training and brings minimal error exploitation in the value function. We use a didatic toycase example to show how previous diffusion-based methods fail to generate optimal actions due to leveraging these errors and how Diffusion-DICE successfully avoids that. We then conduct extensive experiments on benchmark datasets to show the strong performance of Diffusion-DICE. Project page at https://ryanxhr.github.io/Diffusion-DICE/., Comment: NeurIPS 2024, first two authors contribute equally

Social technology research today largely focuses on mitigating the negative impacts of technology and, therefore, often misses the potential of technology to enhance human connections and well-being. However, we see a potential to shift towards a holistic view of social technology's impact on human flourishing. We introduce Positive Social Technology (Positech), a framework that shifts emphasis toward leveraging social technologies to support and augment human flourishing. This workshop is organized around three themes relevant to Positech: 1) "Exploring Relevant and Adjacent Research" to define and widen the Positech scope with insights from related fields, 2) "Projecting the Landscape of Positech" for participants to outline the domain's key aspects and 3) "Envisioning the Future of Positech," anchored around strategic planning towards a sustainable research community. Ultimately, this workshop will serve as a platform to shift the narrative of social technology research towards a more positive, human-centric approach. It will foster research that goes beyond fixing technologies to protect humans from harm, to also pursue enriching human experiences and connections through technology.

The increasing harms caused by hate, harassment, and other forms of abuse online have motivated major platforms to explore hierarchical governance. The idea is to allow communities to have designated members take on moderation and leadership duties; meanwhile, members can still escalate issues to the platform. But these promising approaches have only been explored in plaintext settings where community content is public to the platform. It is unclear how one can realize hierarchical governance in the huge and increasing number of online communities that utilize end-to-end encrypted (E2EE) messaging for privacy. We propose private hierarchical governance systems. These should enable similar levels of community governance as in plaintext settings, while maintaining cryptographic privacy of content and governance actions not reported to the platform. We design the first such system, taking a layered approach that adds governance logic on top of an encrypted messaging protocol; we show how an extension to the message layer security (MLS) protocol suffices for achieving a rich set of governance policies. Our approach allows developers to rapidly prototype new governance features, taking inspiration from a plaintext system called PolicyKit. We build a prototype E2EE messaging system called MlsGov that supports content-based community and platform moderation, elections of community moderators, votes to remove abusive users, and more., Comment: Published in IEEE Security and Privacy 2024

At the core of both successful generative and self-supervised representation learning models there is a reconstruction objective that incorporates some form of image corruption. Diffusion models implement this approach through a scheduled Gaussian corruption process, while masked auto-encoder models do so by masking patches of the image. Despite their different approaches, the underlying similarity in their methodologies suggests a promising avenue for an auto-encoder capable of both de-noising tasks. We propose a unified self-supervised objective, dubbed Unified Masked Diffusion (UMD), that combines patch-based and noise-based corruption techniques within a single auto-encoding framework. Specifically, UMD modifies the diffusion transformer (DiT) training process by introducing an additional noise-free, high masking representation step in the diffusion noising schedule, and utilizes a mixed masked and noised image for subsequent timesteps. By integrating features useful for diffusion modeling and for predicting masked patch tokens, UMD achieves strong performance in downstream generative and representation learning tasks, including linear probing and class-conditional generation. This is achieved without the need for heavy data augmentations, multiple views, or additional encoders. Furthermore, UMD improves over the computational efficiency of prior diffusion based methods in total training time. We release our code at https://github.com/philippe-eecs/small-vision., Comment: 19 Pages, 8 Figures, 3Tables

Demonstrations are an effective alternative to task specification for learning agents in settings where designing a reward function is difficult. However, demonstrating expert behavior in the action space of the agent becomes unwieldy when robots have complex, unintuitive morphologies. We consider the practical setting where an agent has a dataset of prior interactions with the environment and is provided with observation-only expert demonstrations. Typical learning from observations approaches have required either learning an inverse dynamics model or a discriminator as intermediate steps of training. Errors in these intermediate one-step models compound during downstream policy learning or deployment. We overcome these limitations by directly learning a multi-step utility function that quantifies how each action impacts the agent's divergence from the expert's visitation distribution. Using the principle of duality, we derive DILO (Dual Imitation Learning from Observations), an algorithm that can leverage arbitrary suboptimal data to learn imitating policies without requiring expert actions. DILO reduces the learning from observations problem to that of simply learning an actor and a critic, bearing similar complexity to vanilla offline RL. This allows DILO to gracefully scale to high dimensional observations, and demonstrate improved performance across the board. Project page (code and videos): $\href{https://hari-sikchi.github.io/dilo/}{\text{hari-sikchi.github.io/dilo/}}$, Comment: 8th Conference on Robot Learning (CoRL 2024), Munich, Germany. 23 pages

Reinforcement Learning is a promising tool for learning complex policies even in fast-moving and object-interactive domains where human teleoperation or hard-coded policies might fail. To effectively reflect this challenging category of tasks, we introduce a dynamic, interactive RL testbed based on robot air hockey. By augmenting air hockey with a large family of tasks ranging from easy tasks like reaching, to challenging ones like pushing a block by hitting it with a puck, as well as goal-based and human-interactive tasks, our testbed allows a varied assessment of RL capabilities. The robot air hockey testbed also supports sim-to-real transfer with three domains: two simulators of increasing fidelity and a real robot system. Using a dataset of demonstration data gathered through two teleoperation systems: a virtualized control environment, and human shadowing, we assess the testbed with behavior cloning, offline RL, and RL from scratch.

Reinforcement learning (RL) algorithms often struggle to learn policies that generalize to novel situations due to issues such as causal confusion, overfitting to irrelevant factors, and failure to isolate control of state factors. These issues stem from a common source: a failure to accurately identify and exploit state-specific causal relationships in the environment. While some prior works in RL aim to identify these relationships explicitly, they rely on informal domain-specific heuristics such as spatial and temporal proximity. Actual causality offers a principled and general framework for determining the causes of particular events. However, existing definitions of actual cause often attribute causality to a large number of events, even if many of them rarely influence the outcome. Prior work on actual causality proposes normality as a solution to this problem, but its existing implementations are challenging to scale to complex and continuous-valued RL environments. This paper introduces functional actual cause (FAC), a framework that uses context-specific independencies in the environment to restrict the set of actual causes. We additionally introduce Joint Optimization for Actual Cause Inference (JACI), an algorithm that learns from observational data to infer functional actual causes. We demonstrate empirically that FAC agrees with known results on a suite of examples from the actual causality literature, and JACI identifies actual causes with significantly higher accuracy than existing heuristic methods in a set of complex, continuous-valued environments.

Current work in language models (LMs) helps us speed up or even skip thinking by accelerating and automating cognitive work. But can LMs help us with critical thinking -- thinking in deeper, more reflective ways which challenge assumptions, clarify ideas, and engineer new concepts? We treat philosophy as a case study in critical thinking, and interview 21 professional philosophers about how they engage in critical thinking and on their experiences with LMs. We find that philosophers do not find LMs to be useful because they lack a sense of selfhood (memory, beliefs, consistency) and initiative (curiosity, proactivity). We propose the selfhood-initiative model for critical thinking tools to characterize this gap. Using the model, we formulate three roles LMs could play as critical thinking tools: the Interlocutor, the Monitor, and the Respondent. We hope that our work inspires LM researchers to further develop LMs as critical thinking tools and philosophers and other 'critical thinkers' to imagine intellectually substantive uses of LMs.

Robust reinforcement learning agents using high-dimensional observations must be able to identify relevant state features amidst many exogeneous distractors. A representation that captures controllability identifies these state elements by determining what affects agent control. While methods such as inverse dynamics and mutual information capture controllability for a limited number of timesteps, capturing long-horizon elements remains a challenging problem. Myopic controllability can capture the moment right before an agent crashes into a wall, but not the control-relevance of the wall while the agent is still some distance away. To address this we introduce action-bisimulation encoding, a method inspired by the bisimulation invariance pseudometric, that extends single-step controllability with a recursive invariance constraint. By doing this, action-bisimulation learns a multi-step controllability metric that smoothly discounts distant state features that are relevant for control. We demonstrate that action-bisimulation pretraining on reward-free, uniformly random data improves sample efficiency in several environments, including a photorealistic 3D simulation domain, Habitat. Additionally, we provide theoretical analysis and qualitative results demonstrating the information captured by action-bisimulation., Comment: Published at the Reinforcement Learning Conference 2024

In real-world control settings, the observation space is often unnecessarily high-dimensional and subject to time-correlated noise. However, the controllable dynamics of the system are often far simpler than the dynamics of the raw observations. It is therefore desirable to learn an encoder to map the observation space to a simpler space of control-relevant variables. In this work, we consider the Ex-BMDP model, first proposed by Efroni et al. (2022), which formalizes control problems where observations can be factorized into an action-dependent latent state which evolves deterministically, and action-independent time-correlated noise. Lamb et al. (2022) proposes the "AC-State" method for learning an encoder to extract a complete action-dependent latent state representation from the observations in such problems. AC-State is a multistep-inverse method, in that it uses the encoding of the the first and last state in a path to predict the first action in the path. However, we identify cases where AC-State will fail to learn a correct latent representation of the agent-controllable factor of the state. We therefore propose a new algorithm, ACDF, which combines multistep-inverse prediction with a latent forward model. ACDF is guaranteed to correctly infer an action-dependent latent state encoder for a large class of Ex-BMDP models. We demonstrate the effectiveness of ACDF on tabular Ex-BMDPs through numerical simulations; as well as high-dimensional environments using neural-network-based encoders. Code is available at https://github.com/midi-lab/acdf., Comment: RLC 2024

Real-world misinformation, often multimodal, can be partially or fully factual but misleading using diverse tactics like conflating correlation with causation. Such misinformation is severely understudied, challenging to address, and harms various social domains, particularly on social media, where it can spread rapidly. High-quality and timely correction of misinformation that identifies and explains its (in)accuracies effectively reduces false beliefs. Despite the wide acceptance of manual correction, it is difficult to be timely and scalable. While LLMs have versatile capabilities that could accelerate misinformation correction, they struggle due to a lack of recent information, a tendency to produce false content, and limitations in addressing multimodal information. We propose MUSE, an LLM augmented with access to and credibility evaluation of up-to-date information. By retrieving evidence as refutations or supporting context, MUSE identifies and explains content (in)accuracies with references. It conducts multimodal retrieval and interprets visual content to verify and correct multimodal content. Given the absence of a comprehensive evaluation approach, we propose 13 dimensions of misinformation correction quality. Then, fact-checking experts evaluate responses to social media content that are not presupposed to be misinformation but broadly include (partially) incorrect and correct posts that may (not) be misleading. Results demonstrate MUSE's ability to write high-quality responses to potential misinformation--across modalities, tactics, domains, political leanings, and for information that has not previously been fact-checked online--within minutes of its appearance on social media. Overall, MUSE outperforms GPT-4 by 37% and even high-quality responses from laypeople by 29%. Our work provides a general methodological and evaluative framework to correct misinformation at scale., Comment: 50 pages

League of Legends (LoL) has been a dominant esport for a decade, yet the inherent complexity of the game has stymied the creation of analytical measures of player skill and performance. Current industry standards are limited to easy-to-procure individual player statistics that are incomplete and lacking context as they do not take into account teamplay or game state. We present a unified performance model for League of Legends which blends together measures of a player's contribution within the context of their team, insights from traditional sports metrics such as the Plus-Minus model, and the intricacies of LoL as a complex team invasion sport. Using hierarchical Bayesian models, we outline the use of gold and damage dealt as a measure of skill, detailing players' impact on their own-, their allies'- and their enemies' statistics throughout the course of the game. Our results showcase the model's increased efficacy in separating professional players when compared to a Plus-Minus model and to current esports industry standards, while metric quality is rigorously assessed for discrimination, independence, and stability. Readers might also find additional qualitative analytics which explore champion proficiency and the impact of collaborative team-play. Future work is proposed to refine and expand the SIDO performance model, offering a comprehensive framework for esports analytics in team performance management, scouting and research realms.

As more users turn to video-sharing platforms like YouTube as an information source, they may consume misinformation despite their best efforts. In this work, we investigate ways that users can better assess the credibility of videos by first exploring how users currently determine credibility using existing signals on platforms and then by introducing and evaluating new credibility-based signals. We conducted 12 contextual inquiry interviews with YouTube users, determining that participants used a combination of existing signals, such as the channel name, the production quality, and prior knowledge, to evaluate credibility, yet sometimes stumbled in their efforts to do so. We then developed Viblio, a prototype system that enables YouTube users to view and add citations and related information while watching a video based on our participants' needs. From an evaluation with 12 people, all participants found Viblio to be intuitive and useful in the process of evaluating a video's credibility and could see themselves using Viblio in the future.

In communities with social hierarchies, fear of judgment can discourage communication. While anonymity may alleviate some social pressure, fully anonymous spaces enable toxic behavior and hide the social context that motivates people to participate and helps them tailor their communication. We explore a design space of meronymous communication, where people can reveal carefully chosen aspects of their identity and also leverage trusted endorsers to gain credibility. We implemented these ideas in a system for scholars to meronymously seek and receive paper recommendations on Twitter and Mastodon. A formative study with 20 scholars confirmed that scholars see benefits to participating but are deterred due to social anxiety. From a month-long public deployment, we found that with meronymity, junior scholars could comfortably ask ``newbie'' questions and get responses from senior scholars who they normally found intimidating. Responses were also tailored to the aspects about themselves that junior scholars chose to reveal., Comment: Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI '24), May 11--16, 2024, Honolulu, HI, USA

Social media systems are as varied as they are pervasive. They have been almost universally adopted for a broad range of purposes including work, entertainment, activism, and decision making. As a result, they have also diversified, with many distinct designs differing in content type, organization, delivery mechanism, access control, and many other dimensions. In this work, we aim to characterize and then distill a concise design space of social media systems that can help us understand similarities and differences, recognize potential consequences of design choices, and identify spaces for innovation. Our model, which we call Form-From, characterizes social media based on (1) the form of the content, either threaded or flat, and (2) from where or from whom one might receive content, ranging from spaces to networks to the commons. We derive Form-From inductively from a larger set of 62 dimensions organized into 10 categories. To demonstrate the utility of our model, we trace the history of social media systems as they traverse the Form-From space over time, and we identify common design patterns within cells of the model.

We introduce Diffusion World Model (DWM), a conditional diffusion model capable of predicting multistep future states and rewards concurrently. As opposed to traditional one-step dynamics models, DWM offers long-horizon predictions in a single forward pass, eliminating the need for recursive queries. We integrate DWM into model-based value estimation, where the short-term return is simulated by future trajectories sampled from DWM. In the context of offline reinforcement learning, DWM can be viewed as a conservative value regularization through generative modeling. Alternatively, it can be seen as a data source that enables offline Q-learning with synthetic data. Our experiments on the D4RL dataset confirm the robustness of DWM to long-horizon simulation. In terms of absolute performance, DWM significantly surpasses one-step dynamics models with a $44\%$ performance gain, and is comparable to or slightly surpassing their model-free counterparts.

A recent surge in remote meetings has led to complaints of ``Zoom fatigue'' and ``collaboration overload,'' negatively impacting worker productivity and well-being. One way to alleviate the burden of meetings is to de-emphasize their synchronous participation by shifting work to and enabling sensemaking during post-meeting asynchronous activities. Towards this goal, we propose the design concept of meeting bridges, or information artifacts that can encapsulate meeting information towards bridging to and facilitating post-meeting activities. Through 13 interviews and a survey of 198 information workers, we learn how people use online meeting information after meetings are over, finding five main uses: as an archive, as task reminders, to onboard or support inclusion, for group sensemaking, and as a launching point for follow-on collaboration. However, we also find that current common meeting artifacts, such as notes and recordings, present challenges in serving as meeting bridges. After conducting co-design sessions with 16 participants, we distill key principles for the design of meeting bridges to optimally support asynchronous collaboration goals. Overall, our findings point to the opportunity of designing information artifacts that not only support users to access but also continue to transform and engage in meeting information post-meeting., Comment: accepted to CSCW 2024

Large language models (LLMs) are increasingly capable of providing users with advice in a wide range of professional domains, including legal advice. However, relying on LLMs for legal queries raises concerns due to the significant expertise required and the potential real-world consequences of the advice. To explore \textit{when} and \textit{why} LLMs should or should not provide advice to users, we conducted workshops with 20 legal experts using methods inspired by case-based reasoning. The provided realistic queries ("cases") allowed experts to examine granular, situation-specific concerns and overarching technical and legal constraints, producing a concrete set of contextual considerations for LLM developers. By synthesizing the factors that impacted LLM response appropriateness, we present a 4-dimension framework: (1) User attributes and behaviors, (2) Nature of queries, (3) AI capabilities, and (4) Social impacts. We share experts' recommendations for LLM response strategies, which center around helping users identify `right questions to ask' and relevant information rather than providing definitive legal judgments. Our findings reveal novel legal considerations, such as unauthorized practice of law, confidentiality, and liability for inaccurate advice, that have been overlooked in the literature. The case-based deliberation method enabled us to elicit fine-grained, practice-informed insights that surpass those from de-contextualized surveys or speculative principles. These findings underscore the applicability of our method for translating domain-specific professional knowledge and practices into policies that can guide LLM behavior in a more responsible direction., Comment: 14 pages

Although reinforcement learning (RL) can solve many challenging sequential decision making problems, achieving zero-shot transfer across related tasks remains a challenge. The difficulty lies in finding a good representation for the current task so that the agent understands how it relates to previously seen tasks. To achieve zero-shot transfer, we introduce the function encoder, a representation learning algorithm which represents a function as a weighted combination of learned, non-linear basis functions. By using a function encoder to represent the reward function or the transition function, the agent has information on how the current task relates to previously seen tasks via a coherent vector representation. Thus, the agent is able to achieve transfer between related tasks at run time with no additional training. We demonstrate state-of-the-art data efficiency, asymptotic performance, and training stability in three RL fields by augmenting basic RL algorithms with a function encoder task representation.

Millions of online communities are governed by volunteer moderators, who shape their communities by setting and enforcing rules, recruiting additional moderators, and participating in the community themselves. These moderators must regularly make decisions about how to govern, yet measuring the 'success' of governance is complex and nuanced, making it challenging to determine what governance strategies are most successful. Furthermore, prior work has shown that communities have differing values, suggesting that 'one-size-fits-all' approaches to governance are unlikely to serve all communities well. In this work, we assess governance practices on reddit by classifying the sentiment of community members' public discussion of their own moderators. We label 1.89 million posts and comments made on reddit over an 18 month period. We relate these perceptions to characteristics of community governance, and to different actions that community moderators take. We identify types of communities where moderators are perceived particularly positively and negatively, and highlight promising strategies for moderator teams. Amongst other findings, we show that strict rule enforcement is linked to more favorable perceptions of moderators of communities dedicated to certain topics, such as news communities, than others. We investigate what kinds of moderators are associated with improved community perceptions upon their addition to a mod team, and find that moderators who are active community members before and during their mod tenures are seen more favorably. We make all our models, datasets, and code public.

Social media feeds are deeply personal spaces that reflect individual values and preferences. However, top-down, platform-wide content algorithms can reduce users' sense of agency and fail to account for nuanced experiences and values. Drawing on the paradigm of interactive machine teaching (IMT), an interaction framework for non-expert algorithmic adaptation, we map out a design space for teachable social media feed experiences to empower agential, personalized feed curation. To do so, we conducted a think-aloud study (N=24) featuring four social media platforms -- Instagram, Mastodon, TikTok, and Twitter -- to understand key signals users leveraged to determine the value of a post in their feed. We synthesized users' signals into taxonomies that, when combined with user interviews, inform five design principles that extend IMT into the social media setting. We finally embodied our principles into three feed designs that we present as sensitizing concepts for teachable feed experiences moving forward., Comment: CHI 2024

Advancements in large language models (LLMs) are sparking a proliferation of LLM-powered user experiences (UX). In product teams, designers often craft UX to meet user needs, but it is unclear how they engage with LLMs as a novel design material. Through a formative study with 12 designers, we find that designers seek a translational mechanism that enables design requirements to shape and be shaped by LLM behavior, motivating a need for designerly adaptation to facilitate this translation. We then built Canvil, a Figma widget that operationalizes designerly adaptation. We used Canvil as a technology probe in a group-based design study (6 groups, N=17), finding that designers constructively iterated on both adaptation approaches and interface designs to enhance end-user interaction with LLMs. Furthermore, designers identified promising collaborative workflows for designerly adaptation. Our work opens new avenues for processes and tools that foreground designers' user-centered expertise in LLM-powered applications. Canvil is available for public use at https://www.figma.com/community/widget/1277396720888327660.

In this work, we examine how fact-checkers prioritize which claims to fact-check and what tools may assist them in their efforts. Through a series of interviews with 23 professional fact-checkers from around the world, we validate that harm assessment is a central component of how fact-checkers triage their work. We also clarify the processes behind fact-checking prioritization, finding that they are typically ad hoc, and gather suggestions for tools that could help with these processes. To address the needs articulated by fact-checkers, we present a structured framework of questions to help fact-checkers negotiate the priority of claims through assessing potential harms. Our FABLE Framework of Misinformation Harms incorporates five dimensions of magnitude -- (social) Fragmentation, Actionability, Believability, Likelihood of spread, and Exploitativeness -- that can help determine the potential urgency of a specific message or claim when considering misinformation as harm. The result is a practical and conceptual tool to support fact-checkers and others as they make strategic decisions to prioritize their efforts. We conclude with a discussion of computational approaches to support structured prioritization, as well as applications beyond fact-checking to content moderation and curation., Comment: Accepted to CSCW 2024, with clean up for typos and figures

Case studies commonly form the pedagogical backbone in law, ethics, and many other domains that face complex and ambiguous societal questions informed by human values. Similar complexities and ambiguities arise when we consider how AI should be aligned in practice: when faced with vast quantities of diverse (and sometimes conflicting) values from different individuals and communities, with whose values is AI to align, and how should AI do so? We propose a complementary approach to constitutional AI alignment, grounded in ideas from case-based reasoning (CBR), that focuses on the construction of policies through judgments on a set of cases. We present a process to assemble such a case repository by: 1) gathering a set of ``seed'' cases -- questions one may ask an AI system -- in a particular domain from discussions in online communities, 2) eliciting domain-specific key dimensions for cases through workshops with domain experts, 3) using LLMs to generate variations of cases not seen in the wild, and 4) engaging with the public to judge and improve cases. We then discuss how such a case repository could assist in AI alignment, both through directly acting as precedents to ground acceptable behaviors, and as a medium for individuals and communities to engage in moral reasoning around AI., Comment: MP2 workshop @ NeurIPS 2023

Learning to solve tasks from a sparse reward signal is a major challenge for standard reinforcement learning (RL) algorithms. However, in the real world, agents rarely need to solve sparse reward tasks entirely from scratch. More often, we might possess prior experience to draw on that provides considerable guidance about which actions and outcomes are possible in the world, which we can use to explore more effectively for new tasks. In this work, we study how prior data without reward labels may be used to guide and accelerate exploration for an agent solving a new sparse reward task. We propose a simple approach that learns a reward model from online experience, labels the unlabeled prior data with optimistic rewards, and then uses it concurrently alongside the online data for downstream policy and critic optimization. This general formula leads to rapid exploration in several challenging sparse-reward domains where tabula rasa exploration is insufficient, including the AntMaze domain, Adroit hand manipulation domain, and a visual simulated robotic manipulation domain. Our results highlight the ease of incorporating unlabeled prior data into existing online RL algorithms, and the (perhaps surprising) effectiveness of doing so., Comment: 25 pages, 16 figures, 37th Conference on Neural Information Processing Systems (NeurIPS 2023)

Offline Goal-Conditioned Reinforcement Learning (GCRL) is tasked with learning to achieve multiple goals in an environment purely from offline datasets using sparse reward functions. Offline GCRL is pivotal for developing generalist agents capable of leveraging pre-existing datasets to learn diverse and reusable skills without hand-engineering reward functions. However, contemporary approaches to GCRL based on supervised learning and contrastive learning are often suboptimal in the offline setting. An alternative perspective on GCRL optimizes for occupancy matching, but necessitates learning a discriminator, which subsequently serves as a pseudo-reward for downstream RL. Inaccuracies in the learned discriminator can cascade, negatively influencing the resulting policy. We present a novel approach to GCRL under a new lens of mixture-distribution matching, leading to our discriminator-free method: SMORe. The key insight is combining the occupancy matching perspective of GCRL with a convex dual formulation to derive a learning objective that can better leverage suboptimal offline data. SMORe learns scores or unnormalized densities representing the importance of taking an action at a state for reaching a particular goal. SMORe is principled and our extensive experiments on the fully offline GCRL benchmark composed of robot manipulation and locomotion tasks, including high-dimensional observations, show that SMORe can outperform state-of-the-art baselines by a significant margin., Comment: Published at International Conference of Learning Representations (ICLR) 2024. 26 pages

Computer vision often treats human perception as homogeneous: an implicit assumption that visual stimuli are perceived similarly by everyone. This assumption is reflected in the way researchers collect datasets and train vision models. By contrast, literature in cross-cultural psychology and linguistics has provided evidence that people from different cultural backgrounds observe vastly different concepts even when viewing the same visual stimuli. In this paper, we study how these differences manifest themselves in vision-language datasets and models, using language as a proxy for culture. By comparing textual descriptions generated across 7 languages for the same images, we find significant differences in the semantic content and linguistic expression. When datasets are multilingual as opposed to monolingual, descriptions have higher semantic coverage on average, where coverage is measured using scene graphs, model embeddings, and linguistic taxonomies. For example, multilingual descriptions have on average 29.9% more objects, 24.5% more relations, and 46.0% more attributes than a set of monolingual captions. When prompted to describe images in different languages, popular models (e.g. LLaVA) inherit this bias and describe different parts of the image. Moreover, finetuning models on captions from one language performs best on corresponding test data from that language, while finetuning on multilingual data performs consistently well across all test data compositions. Our work points towards the need to account for and embrace the diversity of human perception in the computer vision community.

We examine in detail the provisioning process used by many common, consumer-grade Internet of Things (IoT) devices. We find that this provisioning process involves the IoT device, the vendor's cloud-based server, and a vendor-provided mobile app. In order to better understand this process, we develop two toolkits. IoT-Dissect I enables us to decrypt and examine the messages exchanged between the IoT device and the vendor's server, and between the vendor's server and a vendor-provided mobile app. IoT-Dissect II permits us to reverse engineer the vendor's mobile app and observe its operation in detail. We find several potential security issues with the provisioning process and recommend ways to mitigate these potential problems. Further, based on these observations, we conclude that it is likely feasible to construct a vendor-agnostic IoT home gateway that will automate this largely manual provisioning process, isolate IoT devices on their own network, and perhaps open the tight association between an IoT device and the vendor's server., Comment: 7 pages, 11 figures

Offline reinforcement learning (RL) presents a promising approach for learning reinforced policies from offline datasets without the need for costly or unsafe interactions with the environment. However, datasets collected by humans in real-world environments are often noisy and may even be maliciously corrupted, which can significantly degrade the performance of offline RL. In this work, we first investigate the performance of current offline RL algorithms under comprehensive data corruption, including states, actions, rewards, and dynamics. Our extensive experiments reveal that implicit Q-learning (IQL) demonstrates remarkable resilience to data corruption among various offline RL algorithms. Furthermore, we conduct both empirical and theoretical analyses to understand IQL's robust performance, identifying its supervised policy learning scheme as the key factor. Despite its relative robustness, IQL still suffers from heavy-tail targets of Q functions under dynamics corruption. To tackle this challenge, we draw inspiration from robust statistics to employ the Huber loss to handle the heavy-tailedness and utilize quantile estimators to balance penalization for corrupted data and learning stability. By incorporating these simple yet effective modifications into IQL, we propose a more robust offline RL approach named Robust IQL (RIQL). Extensive experiments demonstrate that RIQL exhibits highly robust performance when subjected to diverse data corruption scenarios., Comment: Accepted by ICLR 2024

Navigating the vast scientific literature often starts with browsing a paper's abstract. However, when a reader seeks additional information, not present in the abstract, they face a costly cognitive chasm during their dive into the full text. To bridge this gap, we introduce recursively expandable abstracts, a novel interaction paradigm that dynamically expands abstracts by progressively incorporating additional information from the papers' full text. This lightweight interaction allows scholars to specify their information needs by quickly brushing over the abstract or selecting AI-suggested expandable entities. Relevant information is synthesized using a retrieval-augmented generation approach, presented as a fluid, threaded expansion of the abstract, and made efficiently verifiable via attribution to relevant source-passages in the paper. Through a series of user studies, we demonstrate the utility of recursively expandable abstracts and identify future opportunities to support low-effort and just-in-time exploration of long-form information contexts through LLM-powered interactions., Comment: 21 pages, 10 figures, 4 tables. arXiv admin note: text overlap with arXiv:2305.14314 by other authors