Syllabus

---

Every reading gets a short commentary, due at noon on the day of its lecture. Write and submit them here, and read how to write one before your first. Off campus, the Stanford Library proxy gets you past paywalls.

Unit 1Ubiquitous Computing

Wed Sep 23Week 1

Ubiquitous and Tangible ComputingSlides

No readings.

Tutorial signupsdue Sunday, Sep 27, 9:00pm

Mon Sep 28Week 2

Input and InteractionSlides

The Computer for the 21st Century Mark Weiser. Scientific American. September 1991, pp. 94-104. [Submit Commentary]

This is one of the most influential, highly-cited papers in all of human-computer interaction. Weiser's vision of ubiquitous computing articulated a vision in which computing recedes into the background rather than stands as a focal point of our attention. Researchers and practitioners have chased this vision since 1991. We read this article because it is the foundation upon which much of the Ubicomp component of the class sits. What do you think about Weiser's differentiation between calm computation that recedes into infrastructure, and simply spreading computers around everywhere in the environment? What are the implications of such a vision?

Tangible Bits: Towards Seamless Interfaces Between People, Bits and Atoms Hiroshi Ishii, Brygg Ullmer. CHI 1997. [Submit Commentary]

The Tangible Bits paper sits alongside Weiser's Ubiquitous Computing paper as one of the foundational works of modern HCI. It came out several years after Weiser's article, but in many ways, was the innovation that made Weiser's vision actionable. Tangible bits takes a strong stance that input and output need to be aligned in concrete physical artifacts, whereas Weiser allowed digital outputs such as screens and projectors. This paper has spawned many, many projects and even entire conferences. What tradeoffs does Tangible Bits make through its strict adherence to input/output co-location?

Wed Sep 30

Activity SensingSlides

Opioid Overdose Detection Using Smartphones Rajalakshmi Nandakumar, Shyamnath Gollakota, Jacob E. Sunshine. Science Translational Medicine 2019. [Submit Commentary]

This paper offers a sensing pipeline around human activity. Here, they're focused on detecting opioid overdose. Focus on the overall pipeline: what are the sensors and features? What could go wrong here? How do they defend against the false positives? What are the ethics of deploying this kind of technology? What ought we to deploy?

Bloom: Designing for LLM-Augmented Behavior Change Interactions Matthew Jörke, Defne Genç, Valentin Teutschbein, Shardul Sapkota, Sarah Chung, Paul Schmiedmayer, Maria Ines Campero, Abby C. King, Emma Brunskill, James A. Landay. CHI 2026. [Submit Commentary]

Bloom keeps the activity sensing of a fitness tracker but adds a language-model coach, and it tests the result in a four-week randomized field study. Do you agree that we should be adding a conversational agent to these behavior change and sensing pipelines? What would be needed for this effect to persist and work at scale?

Unit 2Design

Mon Oct 5Week 3

Design CognitionSlides

The Design of Everyday Things (Chapter 2) Don Norman. Basic Books, revised and expanded edition, 2013; first published 1988. [Submit Commentary]

This book is one of the best known treatises on design and HCI. Norman describes interfaces and the problems with them from the perspective of a cognitive psychologist. Chapter Two is about the psychology of everyday interactions with systems. It describes how people work with systems to get things done and more importantly what people think about as they are working with the systems. The descriptions of the gulf of execution and evaluation, the seven stages of action and the corresponding seven principles of design will recur throughout this course. What other interface design problems have you encountered where they might have been relevant?

Quiz 1end of class, after the lecture; bring pencil

Wed Oct 7

Design ProcessSlides

'Yours is better!': participant response bias in HCI Nicola Dell, Vidya Vaidyanathan, Indrani Medhi, Edward Cutrell, William Thies. CHI 2012. [Submit Commentary]

The design process leans heavily on asking people what they think of your design. Dell and her colleagues ran a series of experiments in Bangalore showing how far the answers bend toward what participants think the researcher wants to hear: participants were roughly 2.5 times as likely to prefer a design when they believed the interviewer had made it, and roughly 5 times as likely when a foreign researcher asked through a translator. Which parts of the design process are most exposed to this kind of demand characteristic? Which of the interventions we discuss in class actually address it, and which only paper over it? Is this a problem of method, or of the power relationship between designer and participant?

Deconstructing Community-Based Collaborative Design: Towards More Equitable Participatory Design Engagements Christina Harrington, Sheena Erete, Anne Marie Piper. CSCW 2019. [Submit Commentary]

As we will discuss in class, participatory design is a method for bringing community stakeholders in directly to the full design process, rather than only allowing the designer to engage in the concept generation and selection stages. But, in this work, Harrington and her colleagues demonstrate through a series of co-design (participatory) workshops that bringing people into the room does not mean that they have, or view themselves as having, full status and authority. What do you think we ought to do about this problem? Does it suggest deeper issues with participatory design and its goals? Does it suggest alternative design processes that we might consider?

Mon Oct 12Week 4

Design ToolsSlides

The Reflective Practitioner (Chapter 3) Donald Schön. Basic Books, 1983. [Submit Commentary]

Optional: read Chapter 2, which sets up Schön's critique of rationality inproblem solving and lays out his theory of reflection-in-action.

This book provides foundational theory that we draw on in HCI design: what is core to the activity of design, and what does expertise in design really mean? Schön's answer — reflection in action — is a frame that has proven useful, by articulating that the designer iterates by taking an action, reflecting on the result of that action, and using that reflection to gain insight into the problem in order to plan the next action. Schön's argument that design cannot be planned and is instead a process of iteration and reflection is foundational reasoning in HCI as to why our field works the way it does, and why we teach the way we do. Do you agree with his position? What are the theory's consequences for the goals of the tools and processes that we develop for design?

Generating Automatic Feedback on UI Mockups with Large Language Models Peitong Duan, Jeremy Warner, Yang Li, Björn Hartmann. CHI 2024. [Submit Commentary]

This paper offers an AI-powered opportunity to get rapid feedback on your design. Focus on their decision to stay anchored to known design principles, rather than give open-ended feedback. Do you want AI feedback on your design? What role ought it to play in the design feedback process?

Unit 3Social Computing

Wed Oct 14

Social MediaSlides

Beyond Being There Jim Hollan, Scott Stornetta. CHI 1992. [Submit Commentary]

A classic of social computing, which was originally called CSCW (Computer-Supported Cooperative Work). It essentially argues both that remote collaboration will never be as good, and that it could be even better. The question of "does this satisfy the Beyond Being There criteria?" is a good gut check for any social computing system. What implications do you see of this idea, or what alternatives might you pitch? What are the implications of the claim, if we take it in its strongest instantiation that we should never even try to mirror offline interactions?

Groupware and social dynamics: eight challenges for developers Jonathan Grudin. Communications of the ACM 1994. [Submit Commentary]

A second classic of social computing (Grudin calls it `groupware'). In this article in the Communications of the ACM, Jonathan Grudin lays out all the reasons why social applications fail to get over the cold start problem. These problems remain resiliently, frustratingly challenging today. Which do you feel are the most core of the problems that he articulates, and why are they the most central? Are there any core problems that he overlooked?

Mon Oct 19Week 5

CollaborationSlides

Distance Matters Gary Olson, Judith Olson. Human-Computer Interaction 2000. [Submit Commentary]

Rounding out a week of classics. CSCW ("computer-supported cooperative work") is a vibrant field of HCI focused on collaboration, and is the field that gave rise to social computing. (Social computing broadened CSCW out from office-based collaboration environments to user-contributed content online.) The Distance Matters paper is part of the CSCW canon. This paper raised an important question as to how effective collaboration software could really ever be, and why. Do you think the Distance Matters limit is fundamental? If so, why? If not, what could change the situation?

Quiz 2end of class, after the lecture; bring pencil

Wed Oct 21

Societal ComputingSlides

Feminist HCI: taking stock and outlining an agenda for design Shaowen Bardzell. CHI 2010. [Submit Commentary]

Do we err by assigning genderless constructs such as 'the user' in HCI? Do our visions of HCI make implicit assumptions about what kinds of sensing and tracking are desirable, what kinds of social interactions ought to be designed for, or what kinds of problems are interesting, based on this notion of the prototypical user? As you read this article about refocusing attention from default users to the marginalized, ask: where else do these gaps arise, and what might we do about them?

Reranking partisan animosity in algorithmic social media feeds alters affective polarization Tiziano Piccardi, Martin Saveski, Chenyan Jia, Jeffrey T. Hancock, Jeanne L. Tsai, Michael S. Bernstein. Science 2025. [Submit Commentary]

Could a social media feed be ranked for a societal value rather than for engagement? This paper stops arguing and tests it: a browser extension reranks people's real X feeds in real time, with a language model scoring each post for antidemocratic attitudes and partisan animosity, in a preregistered ten-day field experiment with 1,256 participants during the 2024 presidential campaign. Pushing that content down or up shifted feelings toward the other party by more than two points on a 100-point thermometer, in both directions and regardless of party. Focus on the method as much as the result: the platform never cooperated. What does it mean that outside researchers can now run experiments on a feed? Which other values could a feed be reranked for, and who should get to decide?

Unit 4Software and Tools

Mon Oct 26Week 6

Human-centered AI

Direct Manipulation Vs. Interface Agents Ben Shneiderman and Pattie Maes. Interactions 1997. [Submit Commentary]

In a pair of debates in 1997 researchers Pattie Maes and Ben Shneiderman argued about the way in which future interfaces would work. Would users delegate tasks to proactive software agents (Maes' position) or would users always have full control even as automation increases (Shneiderman position). One could argue that until recently the direct manipulation paradigm has been dominant. But with the advent of LLMs and generative AI are we moving to a world of autonomous software agents?

Just-In-Time Objectives: A General Approach for Specialized AI Interactions Michelle S. Lam, Omar Shaikh, Hallie Xu, Alice Guo, Diyi Yang, Jeffrey Heer, James A. Landay, Michael S. Bernstein. CHI 2026. [Submit Commentary]

Non-experts struggle to steer large language models by writing better prompts. This paper takes a different route: instead of asking the person to specify what they want, the system infers what they are trying to do right now, then optimizes its interface and responses for that one objective on the fly. Is inferring the objective the right move, or does it take agency away from the user at exactly the moment they most need it? What would you want the system to show you about the objective it inferred, and how would you correct it? Where does this land in the direct manipulation versus agents debate from the other reading today?

Wed Oct 28

Programming and Toolkits

AI Chains: Transparent and Controllable Human-AI Interaction by Chaining Large Language Model Prompts Tongshuang Wu, Michael Terry, Carrie J. Cai. CHI 2022. [Submit Commentary]

Should we program sequences of LLM prompts? AI chains offer a toolkit for having more control over a large AI pipeline. It is something of a paper of its moment, as today's agentic models obviously go farther. However: do you think there is benefit to this kind of explicit authorship of the workflow? If so, how should it be controlled by us?

Professional Software Developers Don't Vibe, They Control: AI Agent Use for Coding in 2025 Ruanqianqian (Lisa) Huang, Avery Reyna, Sorin Lerner, Haijun Xia, Brian Hempel. arXiv preprint 2025. [Submit Commentary]

Here we stand in The Year of Our Claude 2026, with professional programmers figuring out what their role is. This is a joint survey+interview study of developers who are using agentic coding environments. What do you expect to be a durable shift here? What is just a temporary mirage?

Mon Nov 2Week 7

TBA (taught by the course assistants; Michael at UIST)

No readings.

Wed Nov 4

No lecture: Quiz 3 (Michael at UIST)

No readings.

Quiz 3quiz only, no lecture this day; bring pencil

Mon Nov 9Week 8

Content Creation

Design principles for visual communication Maneesh Agrawala, Wilmot Li, and Floraine Berthouzoz. Communications of the ACM 2011. [Submit Commentary]

Maneesh Agrawala and his collaborators developed this approach to crafting diagrams, visualizations, and explanations. This article is a review of several of his group's projects, and an articulation of the general approach they take to their work. What role is computation playing, compared to what it would take to create these graphics manually?

Block and Detail: Scaffolding Sketch-to-Image Generation Vishnu Sarukkai, Lu Yuan, Mia Tang, Maneesh Agrawala, Kayvon Fatahalian. UIST 2024. [Submit Commentary]

Sketch-to-image models can turn a rough sketch into a detailed picture, but they skip the workflow artists actually use: block out the composition first, then add detail. This paper, from Maneesh Agrawala's group, identifies that gap and builds an interface around the artist's iterative refinement loop. It is in many ways the design-principles approach from the previous reading applied to generative AI. Which other content creation domains have a workflow that generative tools currently flatten, and what would it take to scaffold them the same way? The project page has the video.

Unit 5Cognition

Wed Nov 11

Visualization

Things that Make Us Smart (Chapter 3) Don Norman. Addison-Wesley, 1993. [Submit Commentary]

Choosing the 'right' representation for a problem can make the problem far easier to solve. The challenge is to match the representation to the problem or task. In visualization the goal is to represent numbers and data visually. The best visual representation is also dependent on the task. Is the task to identify outliers in a set of home price data? Is it to compare the heights of children in different grades? Is it to understand multiple dimensions and the correlations between them? How does the task affect your choice of representation?

Viral Visualizations: How Coronavirus Skeptics Use Orthodox Data Practices to Promote Unorthodox Science Online Crystal Lee, Tanya Yang, Gabrielle D. Inchoco, Graham M. Jones, and Arvind Satyanarayan. CHI 2021. [Submit Commentary]

This work presents a mixed quantitative-qualitative investigation of how coronavirus skeptics coopted the language and scientific rhetoric of information visualization to argue for conclusions that were counter to the medical scientific consensus. Information visualization has long recognized the role of narrative visualization, or the use of visualization to tell a story or convince an audience. This work dramatically expanded on this theory by demonstrating how a group of scientific skeptics used the rhetorics of science and visualization to 'follow the data' to its own conclusions. This work re-emphasized how artifacts have politics and how visualization cannot be argued as some sort of objective process that produces correct answers.

Mon Nov 16Week 9

Cognitive Models

The Humane Interface (Chapter 4) Jef Raskin. Addison-Wesley, 2000. [Submit Commentary]

Jef Raskin was an HCI expert who led the Macintosh project at Apple. In this chapter of his book he describes a set of methods developed by HCI researchers in the 1980s to quantify human performance based on models of human information processing. The keystroke-level model attempts to compute how long it might take the average person to perform a task using a GUI. The exact numbers produced by these methods are very rough. But they can be useful to obtain ballpark figures for how long it might take to perform a particular task. And they can help designers identify areas of the interface that might require too many clicks or keystrokes. But developing such low-level models is tedious and since the late 1990s these methods have fallen out of favor. Today they are little used in practice. Why do you think this might be? Are there issues today that could be addressed by building such models.

Generative Agents: Interactive Simulacra of Human Behavior Joon Sung Park, Joseph O'Brien, Carrie Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein. UIST 2023. [Submit Commentary]

How might we craft simulations of human individuals and societies that reflect our lives? Many of the core design challenges in human-computer interaction that we discussed in this class, from building computational tools to designing computer-mediated social interactions, must reckon with the complex nature of our world and the way individuals behave and interact with technology. Generative agents point to a future in which the power to simulate hypothetical worlds, where we can ask 'what if' counterfactual questions and paint concrete pictures of how a multiverse of different possibilities might unfold, promises an opportunity to navigate this complexity. This week, reflect on how generative agents extend the original visions of the foundational literature on cognitive models that we covered in this class (e.g., GOMS, KLM) in the era of generative AI models. How might these simulacra of human behaviors inform the design of our interactions, and what new interaction paradigms will these agents enable?

Wed Nov 18

Methodology

The Sciences of the Artificial (Chapter 5—The Science of Design: Creating the Artificial) Herbert A. Simon. MIT Press 1970. [Submit Commentary]

In The Sciences of the Artificial, Herb Simon argues for the status of design as a rigorous method in and of itself. In doing so, similar to Schön, he contrasts design with traditional engineering subjects and the ways in which seeking to cast all goals in terms of formalization of optimization objectives narrows our view. Does the argument convince? How else might we argue for design as a rigorous method?

Quiz 4end of class, after the lecture; bring pencil

Nov 23-27Thanksgiving

No class: Thanksgiving recess

Unit 6Global Citizenship

Mon Nov 30Week 10

Accessibility

Ability-Based Design: Concept, Principles and Examples Jacob O. Wobbrock, Shaun K. Kane, Krzysztof Z. Gajos, Susumu Harada, Jon Froehlich. ACM Transactions on Accessible Computing 2011. [Submit Commentary]

Wobbrock and colleagues critique the typical approach to accessibility design, which is to focus on the dis in disability---the deficit. Instead, they argue for ability-based design, which places the designer's attention on what the individual can do rather than what they cannot, and provide examples from their own research. What was your reaction to this inversion of the focus in accessibility? What kind of impact can, and can't, it have on the designs we have around us?

GenAssist: Making image generation accessible Mina Huh, Yi-Hao Peng, and Amy Pavel. Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology. 2023. [Submit Commentary]

How can blind and low vision users make better use of generative AI tools such as stable diffusion? GenAssist proposes a suite of automated checks to ensure that the generated image respects the desired caption. Obviously, this kind of approach cannot be 100% accurate in its checks. Given our discussions of AI overreliance, does this make it dangerous or useless? If not, why not? What design principles might we distill for these kinds of tools?

Wed Dec 2

Something Old, Something New

As We May Think Vannevar Bush. Atlantic Monthly 1945. [Submit Commentary]

Bush's article As We May Think is often pinpointed as the moment that catalyzed the ideas that gave rise to human-computer interaction as a field.

Pick a UIST paper Your choice. UIST 2026. [Submit Commentary]

Browse and pick one paper that appeared in the UIST 2026 program (UIST is this November, so these are the newest papers in the field), on a topic of your choice. We recommend browsing the Sessions list to find topics you're interested in, then picking a paper from that list. What does this paper speak to you about the future of HCI? Where should the field be heading?

Finalend of class, after the lecture; comprehensive, about 45 minutes