As agentic AI systems grow increasingly capable of operating for hours or days at a time, users’ prompts are transforming into highly elaborate specifications for the AI to autonomously work on. While prompting for bounded, single-turn tasks has been extensively studied, less is known about how people communicate specifications for long-horizon tasks. In this work, we conducted a qualitative study in which 16 professionals drafted specifications for both a human colleague and an AI, revealing a core divergence: participants treated human delegation as a “compass,” offering high-level intent to encourage flexible exploration. In contrast, communication with AI resembled painstakingly laying down “railway tracks”: rigid, exhaustive instructions to minimize ambiguity and deviation. This reflected a perception that current AI struggles to infer intent, prioritize, and make judgments on its own. When envisioning an ideal AI collaborator, users expressed a desire for a hybrid : a collaborator blending AI’s efficiency and large context window with the critical thinking and agency of a human colleague. We discuss design implications for future AI systems, proposing that they align on outcomes through generated rough drafts, verify feasibility via end-to-end “test runs,” and monitor execution through intelligent check-ins—ultimately transforming AI from a passive instruction-follower into a reliable collaborator for ambiguous, long-horizon tasks.

Multimodal large language models (MLLMs), with their expansive world knowledge and reasoning capabilities, present a unique opportunity for end-users to create personalized AI sensors capable of reasoning about complex situations. A user could describe a desired sensing task in natural language (e.g., "alert if my toddler is getting into mischief"), with the MLLM analyzing the camera feed and responding within seconds. In a formative study, we found that users saw substantial value in defining their own sensors, yet struggled to articulate their unique personal requirements and debug the sensors through prompting alone. To address these challenges, we developed Gensors, a system that empowers users to define customized sensors supported by the reasoning capabilities of MLLMs. Gensors 1) assists users in eliciting requirements through both automatically-generated and manually created sensor criteria, 2) facilitates debugging by allowing users to isolate and test individual criteria in parallel, 3) suggests additional criteria based on user-provided images, and 4) proposes test cases to help users "stress test" sensors on potentially unforeseen scenarios. In a user study, participants reported significantly greater sense of control, understanding, and ease of communication when defining sensors using Gensors. Beyond addressing model limitations, Gensors supported users in debugging, eliciting requirements, and expressing unique personal requirements to the sensor through criteria-based reasoning; it also helped uncover users' "blind spots" by exposing overlooked criteria and revealing unanticipated failure modes. Finally, we discuss how unique characteristics of MLLMs--such as hallucinations and inconsistent responses--can impact the sensor-creation process. These findings contribute to the design of future intelligent sensing systems that are intuitive and customizable by everyday users.

Prototyping AI applications is notoriously difficult. While large language model (LLM) prompting has dramatically lowered the barriers to AI prototyping, designers are still prototyping AI functionality and UI separately. We investigate how coupling prompt and UI design affects designers’ workflows. Grounding this research, we developed PromptInfuser, a Figma plugin that enables users to create semi-functional mockups, by connecting UI elements to the inputs and outputs of prompts. In a study with 14 designers, we compare PromptInfuser to designers’ current AI-prototyping workflow. PromptInfuser was perceived to be significantly more useful for communicating product ideas, more capable of producing prototypes that realistically represent the envisioned artifact, more efficient for prototyping, and more helpful for anticipating UI issues and technical constraints. PromptInfuser encouraged iteration over prompt and UI together, which helped designers identify UI and prompt incompatibilities and reflect upon their total solution. Together, these findings inform future systems for prototyping AI applications.

Large language model (LLM) prompting is a promising new approach for users to create and customize their own chatbots. However, current methods for steering a chatbot’s outputs, such as prompt engineering and fine-tuning, do not support users in converting their natural feedback on the model’s outputs to changes in the prompt or model. In this work, we explore how to enable users to interactively refine model outputs through their feedback, by helping them convert their feedback into a set of principles (i.e. a constitution) that dictate the model’s behavior. From a formative study, we (1) found that users needed support converting their feedback into principles for the chatbot and (2) classified the different principle types desired by users. Inspired by these findings, we developed ConstitutionMaker, an interactive tool for converting user feedback into principles, to steer LLM-based chatbots. With ConstitutionMaker, users can provide either positive or negative feedback in natural language, select auto-generated feedback, or rewrite the chatbot’s response; each mode of feedback automatically generates a principle that is inserted into the chatbot’s prompt. In a user study with 14 participants, we compare ConstitutionMaker to an ablated version, where users write their own principles. With ConstitutionMaker, participants felt that their principles could better guide the chatbot, that they could more easily convert their feedback into principles, and that they could write principles more efficiently, with less mental demand. ConstitutionMaker helped users identify ways to improve the chatbot, formulate their intuitive responses to the model into feedback, and convert this feedback into specific and clear principles. Together, these findings inform future tools that support the interactive critiquing of LLM outputs.

Recent advances in multimodal large language models (LLMs) have made it easier to rapidly prototype AI-powered features, especially for mobile use cases. However, gathering early, mobile-situated user feedback on these AI prototypes remains challenging. The broad scope and flexibility of LLMs means that, for a given use-case-specific prototype, there is a crucial need to understand the wide range of in-the-wild input users are likely to provide and their in-context expectations for the AI’s behavior. To explore the concept of in situ AI prototyping and testing, we created MobileMaker: a platform that enables designers to rapidly create and test mobile AI prototypes directly on devices. This tool also enables testers to make on-device, in-the-field revisions of prototypes using natural language. In an exploratory study with 16 participants, we explored how user feedback on prototypes created with MobileMaker compares to that of existing prototyping tools (e.g., Figma, prompt editors). Our findings suggest that MobileMaker prototypes enabled more serendipitous discovery of: model input edge cases, discrepancies between AI’s and user’s in-context interpretation of the task, and contextual signals missed by the AI. Furthermore, we learned that while the ability to make in-the-wild revisions led users to feel more fulfilled as active participants in the design process, it might also constrain their feedback to the subset of changes perceived as more actionable or implementable by the prototyping tool.

This paper examines the art practices, artwork, and motivations of prolific users of the latest generation of text-to-image models. Through interviews, observations, and a user survey, we present a sampling of the artistic styles and describe the developed community of practice around generative AI. We find that: 1) artists hold the text prompt and the resulting image can be considered collectively as a form of artistic expression (prompts as art), and 2) prompt templates (prompts with “slots” for others to fill in with their own words) are developed to create generative art styles. We discover that the value placed by this community on unique outputs leads to artists seeking specialized vocabulary to produce distinctive art pieces (e.g., by reading architectural blogs to find phrases to describe images). We also find that some artists use “glitches” in the model that can be turned into artistic styles of their own right. From these findings, we outline specific implications for design regarding future prompting and image editing options.

Believable proxies of human behavior can empower interactive applications ranging from immersive environments to rehearsal spaces for interpersonal communication to prototyping tools. In this paper, we introduce generative agents: computational software agents that simulate believable human behavior. Generative agents wake up, cook breakfast, and head to work; artists paint, while authors write; they form opinions, notice each other, and initiate conversations; they remember and reflect on days past as they plan the next day. To enable generative agents, we describe an architecture that extends a large language model to store a complete record of the agent’s experiences using natural language, synthesize those memories over time into higher-level reflections, and retrieve them dynamically to plan behavior. We instantiate generative agents to populate an interactive sandbox environment inspired by The Sims, where end users can interact with a small town of twenty-five agents using natural language. In an evaluation, these generative agents produce believable individual and emergent social behaviors. For example, starting with only a single user-specified notion that one agent wants to throw a Valentine’s Day party, the agents autonomously spread invitations to the party over the next two days, make new acquaintances, ask each other out on dates to the party, and coordinate to show up for the party together at the right time. We demonstrate through ablation that the components of our agent architecture—observation, planning, and reflection—each contribute critically to the believability of agent behavior. By fusing large language models with computational interactive agents, this work introduces architectural and interaction patterns for enabling believable simulations of human behavior.

In this paper, we present a natural language code synthesis tool, GenLine, backed by 1) a large generative language model and 2) a set of task-specific prompts that create or change code. To understand the user experience of natural language code synthesis with these new types of models, we conducted a user study in which participants applied GenLine to two programming tasks. Our results indicate that while natural language code synthesis can sometimes provide a magical experience, participants still faced challenges. In particular, participants felt that they needed to learn the model’s “syntax,” despite their input being natural language. Participants also struggled to form an accurate mental model of the types of requests the model can reliably translate and developed a set of strategies to debug model input. From these findings, we discuss design implications for future natural language code synthesis tools built using large generative language models.

Although large language models (LLMs) have demonstrated impressive potential on simple tasks, their breadth of scope, lack of transparency, and insufficient controllability can make them less effective when assisting humans on more complex tasks. In response, we introduce the concept of Chaining LLM steps together, where the output of one step becomes the input for the next, thus aggregating the gains per step. We first define a set of LLM primitive operations useful for Chain construction, then present an interactive system where users can modify these Chains, along with their intermediate results, in a modular way. In a 20-person user study, we found that Chaining not only improved the quality of task outcomes, but also significantly enhanced system transparency, controllability, and sense of collaboration. Additionally, we saw that users developed new ways of interacting with LLMs through Chains: they leveraged sub-tasks to calibrate model expectations, compared and contrasted alternative strategies by observing parallel downstream effects, and debugged unexpected model outputs by “unit-testing” sub-components of a Chain. In two case studies, we further explore how LLM Chains may be used in future applications.

Social computing prototypes probe the social behaviors that may arise in an envisioned system design. This prototyping practice is currently limited to recruiting small groups of people. Unfortunately, many challenges do not arise until a system is populated at a larger scale. Can a designer understand how a social system might behave when populated, and make adjustments to the design before the system falls prey to such challenges? We introduce social simulacra, a prototyping technique that generates a breadth of realistic social interactions that may emerge when a social computing system is populated. Social simulacra take as input the designer’s description of a community’s design—goal, rules, and member personas—and produce as output an instance of that design with simulated behavior, including posts, replies, and anti-social behaviors. We demonstrate that social simulacra shift the behaviors that they generate appropriately in response to design changes, and that they enable exploration of “what if?” scenarios where community members or moderators intervene. To power social simulacra, we contribute techniques for prompting a large language model to generate thousands of distinct community members and their social interactions with each other; these techniques are enabled by the observation that large language models’ training data already includes a wide variety of positive and negative behavior on social media platforms. In evaluations, we show that participants are often unable to distinguish social simulacra from actual community behavior and that social computing designers successfully refine their social computing designs when using social simulacra.

This paper describes results from a study examining how musicians have been affected by the restrictive environments imposed by the COVID-19 pandemic, and the corresponding technological implications. Through a survey of 29 musicians and additional interviews with 7 professional improvisation musicians, we observe the challenges musicians face in 1) finding the technological infrastructure and shared spaces necessary for dynamic, remote creative collaborations, 2) producing the mindset, free time, and sources of inspiration necessary for initiating creativity during times of stress, and 3) maintaining (and potentially expanding) the social connections and creative community vital to creative practice. We also report on how some musicians have creatively leveraged the radical changes in their lives to drive new artistic practices and styles of music. Collectively, these data illustrate the resilience of certain existing remote collaboration and creative tools, while suggesting ways other digital tools could more flexibly accommodate new use cases. The data also suggest the value of tools and services that help people creatively cope with radical changes to their lives and livelihoods.

Recent advances in deep generative neural networks have made it possible for artificial intelligence to actively collaborate with human beings in co-creating novel content (e.g. music, art). While substantial research focuses on (individual) human-AI collaborations, comparatively less research examines how AI can play a role in human-human collaborations during co-creation. In a qualitative lab study, we observed 30 participants (15 pairs) compose a musical phrase in pairs, both with and without AI. Our findings reveal that AI may play important roles in influencing human social dynamics during creativity, including: 1) implicitly seeding a common ground at the start of collaboration, 2) acting as a psychological safety net in creative risk-taking, 3) providing a force for group progress, 4) mitigating interpersonal stalling and friction, and 5) altering users’ collaborative and creative roles. This work contributes to the future of generative AI in social creativity by providing implications for how AI could enrich, impede, or alter creative social dynamics in the years to come.

While generative deep neural networks (DNNs) have demonstrated their capacity for creating novel musical compositions, less attention has been paid to the challenges and potential of co-creating with these musical AIs, especially for novices. In a needfinding study with a widely used, interactive musical AI, we found that the AI can overwhelm users with the amount of musical content it generates, and frustrate them with its non-deterministic output. To better match co-creation needs, we developed AI-steering tools, consisting of Voice Lanes that restrict content generation to particular voices; Example-Based Sliders to control the similarity of generated content to an existing example; Semantic Sliders to nudge music generation in high-level directions (happy/sad, conventional/surprising); and Multiple Alternatives of generated content to audition and choose from. In a summative study (N=21), we discovered the tools not only increased users' trust, control, comprehension, and sense of collaboration with the AI, but also contributed to a greater sense of self-efficacy and ownership of the composition relative to the AI.

Medical data labeling workflows critically depend on accurate assessments from human experts. Yet human assessments can vary markedly, even among medical experts. Prior research has demonstrated benefits of labeler training on performance. Here we utilized two types of labeler training feedback: highlighting incorrect labels for difficult cases ("individual performance" feedback), and expert discussions from adjudication of these cases. We presented ten generalist eye care professionals with either individual performance alone, or individual performance and expert discussions from specialists. Compared to performance feedback alone, seeing expert discussions significantly improved generalists' understanding of the rationale behind the correct diagnosis while motivating changes in their own labeling approach; and also significantly improved average accuracy on one of four pathologies in a held-out test set. This work suggests that image adjudication may provide benefits beyond developing trusted consensus labels, and that exposure to specialist discussions can be an effective training intervention for medical diagnosis.

Machine learning (ML) is increasingly being used in image retrieval systems for medical decision making. One application of ML is to retrieve visually similar medical images from past patients (e.g. tissue from biopsies) to reference when making a medical decision with a new patient. However, no algorithm can perfectly capture an expert's ideal notion of similarity for every case: an image that is algorithmically determined to be similar may not be medically relevant to a doctor's specific diagnostic needs. In this paper, we identified the needs of pathologists when searching for similar images retrieved using a deep learning algorithm, and developed tools that empower users to cope with the search algorithm on-the-fly, communicating what types of similarity are most important at different moments in time. In two evaluations with pathologists, we found that these tools increased the diagnostic utility of images found and increased user trust in the algorithm. The tools were preferred over a traditional interface, without a loss in diagnostic accuracy. We also observed that users adopted new strategies when using refinement tools, re-purposing them to test and understand the underlying algorithm and to disambiguate ML errors from their own errors. Taken together, these findings inform future human-ML collaborative systems for expert decision-making.

The black-box nature of machine learning algorithms can make their predictions difficult to understand and explain to end-users. In this paper, we propose and evaluate two kinds of example-based explanations in the visual domain, normative explanations and comparative explanations (Figure 1), which automatically surface examples from the training set of a deep neural net sketch-recognition algorithm. To investigate their effects, we deployed these explanations to 1150 users on QuickDraw, an online platform where users draw images and see whether a recognizer has correctly guessed the intended drawing. When the algorithm failed to recognize the drawing, those who received normative explanations felt they had a better understanding of the system, and perceived the system to have higher capability. However, comparative explanations did not always improve perceptions of the algorithm, possibly because they sometimes exposed limitations of the algorithm and may have led to surprise. These findings suggest that examples can serve as a vehicle for explaining algorithmic behavior, but point to relative advantages and disadvantages of using different kinds of examples, depending on the goal.

The growing popularity of machine learning (ML) has attracted more software developers to now want to adopt ML into their own practices, through tinkering with and learning from ML framework websites and online code examples. To investigate the motivations, hurdles, and desires of these software developers, we deployed a survey to the website of the TensorFlow.js ML framework. We found via 645 responses that many wanted to learn ML for aspirational reasons rather than for immediate job needs. Critically, developers faced hurdles due to a perceived lack of mathematical and theoretical background. They desired frameworks to provide more basic ML conceptual support, such as a curated corpus of best practices, conceptual tutorials, and a de-mystification of mathematical jargon into practical tips. These findings inform the design of ML frameworks and informal learning resources to broaden the base of people acquiring this increasingly important skill set.

Although rapid advances in machine learning have made it increasingly applicable to expert decision-making, the delivery of accurate algorithmic predictions alone is insufficient for effective human-AI collaboration. In this work, we investigate the key types of information medical experts desire when they are first introduced to a diagnostic AI assistant. In a qualitative lab study, we interviewed 21 pathologists before, during, and after being presented deep neural network (DNN) predictions for prostate cancer diagnosis, to learn the types of information that they desired about the AI assistant. Our findings reveal that, far beyond understanding the local, case-specific reasoning behind any model decision, clinicians desired upfront information about basic, global properties of the model, such as its known strengths and limitations, its subjective point-of-view, and its overall design objective--what it's designed to be optimized for. Participants compared these information needs to the collaborative mental models they develop of their medical colleagues when seeking a second opinion: the medical perspectives and standards that those colleagues embody, and the compatibility of those perspectives with their own diagnostic patterns. These findings broaden and enrich discussions surrounding AI transparency for collaborative decision-making, providing a richer understanding of what experts find important in their introduction to AI assistants before integrating them into routine practice.

The interpretation of deep learning models is a challenge due to their size, complexity, and often opaque internal state. In addition, many systems, such as image classifiers, operate on low-level features rather than high-level concepts. To address these challenges, we introduce Concept Activation Vectors (CAVs), which provide an interpretation of a neural net’s internal state in terms of human-friendly concepts. The key idea is to view the high-dimensional internal state of a neural net as an aid, not an obstacle. We show how to use CAVs as part of a technique, Testing with CAVs (TCAV), that uses directional derivatives to quantify the degree to which a user-defined concept is important to a classification result–for example, how sensitive a prediction of “zebra” is to the presence of stripes. Using the domain of image classification as a testing ground, we describe how CAVs may be used to explore hypotheses and generate insights for a standard image classification network as well as a medical application.

The busyness of daily life makes it difficult to find time for informal learning. Yet, learning requires significant time and effort, with repeated exposures to educational content on a recurring basis. Despite the struggle to find time, there are numerous moments in a day that are typically wasted due to waiting, such as while waiting for the elevator to arrive, wifi to connect, or an instant message to arrive. We introduce the concept of wait-learning : automatically detecting wait time and inviting people to learn while waiting. Our approach is to design seamless interactions that augment existing wait time with productive opportunities. Combining wait time with productive work opens up a new class of software systems that overcome the problem of limited time. In this article, we establish a design space for wait-learning and explore this design space by creating WaitSuite, a suite of five different wait-learning apps that each uses a different kind of waiting. For one of these apps, we conducted a feasibility study to evaluate learning and to understand how exercises should be timed during waiting periods. Subsequently, we evaluated multiple kinds of wait-learning in a two-week field study of WaitSuite with 25 people. We present design implications for wait-learning, and a theoretical framework that describes how wait time, ease of accessing the learning task, and competing demands impact the effectiveness of wait-learning in different waiting scenarios. These findings provide insight into how wait-learning can be designed to minimize interruption to ongoing tasks and maximize engagement with learning.

Microtasks are small units of work designed to be completed individually, eventually contributing to a larger goal. Although microtasks can be performed in isolation, in practice people often complete a chain of microtasks within a single session. Through a series of crowd-based studies, we look at how various microtasks can be chained together to improve efficiency and minimize mental demand, focusing on the writing domain. We find that participants completed low-complexity microtasks faster when they were preceded by the same type of microtask, whereas they found high-complexity microtasks less mentally demanding when pre-ceded by microtasks on the same content. Furthermore, participants were faster at starting high-complexity microtasks after completing lower-complexity microtasks, but completion time and quality were not affected. These findings provide insight into how microtasks can be ordered to optimize transitions from one microtask to another.

Personalized support for students is a gold standard in education, but it scales poorly with the number of students. Prior work on learnersourcing presented an approach for learners to engage in human computation tasks while trying to learn a new skill. Our key insight is that students, through their own experience struggling with a particular problem, can become experts on the particular optimizations they implement or bugs they resolve. These students can then generate hints for fellow students based on their new expertise. We present workflows that harvest and organize students' collective knowledge and advice for helping fellow novices through design problems in engineering. Systems embodying each workflow were evaluated in the context of a college-level computer architecture class with an enrollment of more than two hundred students each semester. We show that, given our design choices, students can create helpful hints for their peers that augment or even replace teachers' personalized assistance, when that assistance is not available.

Competing priorities in daily life make it difficult for those with a casual interest in learning to set aside time for regular practice. In this paper, we explore wait-learning: leveraging brief moments of waiting during a person's existing conversations for second language vocabulary practice, even if the conversation happens in the native language. We present an augmented version of instant messaging, WaitChatter, that supports the notion of wait-learning by displaying contextually relevant foreign language vocabulary and micro-quizzes just-in-time while the user awaits a response from her conversant. Through a two week field study of WaitChatter with 20 people, we found that users were able to learn 57 new words on average during casual instant messaging. Furthermore, we found that users were most receptive to learning opportunities immediately after sending a chat message, and that this timing may be critical given user tendency to multi-task during waiting periods.

The Experience Sampling Method (ESM) enables researchers to capture information about participants' experiences in the moment. Adding an end-of-day retrospective survey also allows participants to elaborate on those experiences. Although the use of photos in retrospective interviews and surveys for memory elicitation is well known, little research has investigated the use of photos in ESM studies. As smartphone adoption increases facilitating ESM studies and making photo sharing easier, researchers need to continuously evaluate the method and investigate the role of photos in such studies. We conducted a large-scale ESM and retrospective survey study via Android smartphones with more than 1,000 US participants, and analyzed participants' photo submissions, including how photo use correlated with participants' data quality and what, if any, value photos added for researchers. Our study sheds light on the role of photos in ESM and retrospective studies that researchers can reference when constructing future study designs.

With an unprecedented scale of learners watching educational videos on online platforms such as MOOCs and YouTube, there is an opportunity to incorporate data generated from their interactions into the design of novel video interaction techniques. Interaction data has the potential to help not only instructors to improve their videos, but also to enrich the learning experience of educational video watchers. This paper explores the design space of data-driven interaction techniques for educational video navigation. We introduce a set of techniques that augment existing video interface widgets, including: a 2D video timeline with an embedded visualization of collective navigation traces; dynamic and non-linear timeline scrubbing; data-enhanced transcript search and keyword summary; automatic display of relevant still frames next to the video; and a visual summary representing points with high learner activity. To evaluate the feasibility of the techniques, we ran a laboratory user study with simulated learning tasks. Participants rated watching lecture videos with interaction data to be efficient and useful in completing the tasks. However, no significant differences were found in task performance, suggesting that interaction data may not always align with moment-by-moment information needs during the tasks.