Home - Center for Human-driven AI Research and Methods (CHARM) at Harvard

Human-driven AI research that grows people and organizations, not replaces them.

For the public

See how human-driven AI research shapes everyday experiences with technology.

Learn more

For researchers

Explore CHARM’s research agenda and impact stories.

View our research publications

For industry partners

Partner with CHARM on human-centered AI tools, decision-support, and evaluation.

Partner with us

About CHARM

The Center for Human-Driven AI Research and Methods

The Center for Human-Driven AI Research and Methods (CHARM) is a multi-faculty center at the Harvard John A. Paulson School of Engineering and Applied Sciences. We focus on human–AI interaction design and computational methods that leaves people stronger, not just faster in the moment.

Why CHARM now?

AI is often framed as a race for speed: replace human labor with computation, produce more, finish faster. Yet this approach of optimizing for the immediate term risks losing out in the longer run.

At CHARM, we recognize that a single task is never the whole story. A great summary for a negotiator is just one step in their negotiation journey. A great data analysis for a patient is just one step in their health journey. A great writing support tool for a student is just one step of their learning journey. Each task is a moment in a longer human arc—learning, judgment, health, negotiation, creativity.

In addition to asking “Did the tool complete this step?” we also ask “Did the person leave stronger after using it?” or “Is the person better able to fulfill their needs and values?”

Our Mission & Strategy

Mission: To create AI tools that put people first: AI that expands human capability, leaving people better prepared for the next challenge.

Strategy: We make scientific contributions in human-computer interaction, visual computing, and machine learning to design tools that augment human capabilities and unlock what people and organizations can achieve. We move beyond proofs-of-concept to effective products by collaborating with institutions from around the world.

Who are we?

Faculty leading CHARM

CHARM brings together faculty from across Harvard SEAS, united by a focus on human-driven AI methods and interaction design.

Finale Doshi-Velez

Herchel Smith Professor of Computer Science

Reinforcement learning, AI in medicine, decision-making

Krzysztof Gajos

Gordon McKay professor of Computer Science

Human-computer interaction, accessible computing, intelligent interactive systems

Elena Glassman

Assistant Professor of Computer Science

AI-resilient interfaces, AI safety, human-computer interaction

Hanspeter Pfister

An Wang Professor of Computer Science

Visualization, computer graphics, computer vision

Meet the full CHARM team

Announcements

Latest from CHARM

Hongjin Lin is giving a talk with AI & Equality!

CHARM Member Hongjin Lin is giving a talk for AI & Equality on her paper “Funding AI for Good: A Call for Meaningful Engagement”.
You can see all of the details of her talk on the AI & Equality website!

Figure showing how reward model optimization shifts perceived race and gender

Grace Guo accepted into IEEE CVPR

CHARM member Grace Guo has had a paper she worked on “Bias at the End of the Score” accepted at IEEE CVPR 2026! Congratulations, Grace!

Read the full paper

Barbara Grosz discusses AI in today's society with CHARM

Barbara Grosz visits CHARM

We had the honor of hosting Barbara Grosz, Higgins Research Professor of Natural Sciences Emerita at CHARM on April 23rd, where CHARM members had the chance to discuss AI and its impact on society, plus how we can move forward using AI to aid us, while we as humans are still the decision-makers.

Previous announcements

Student Spotlight

Esther Brown on Wearable Technology · July, 2026

Nearly one in three Americans wear a fitness tracker every day, yet most people barely glance beyond their step count. These devices quietly collect a rich stream of physiological data: heart rate, sleep quality, stress indicators, and more. The question is how this information can be used to improve people’s health in more ways than just exercise.

Esther’s research takes that question seriously. She is building a visual framework that layers people’s physiological data alongside the events of their daily lives, whether that be a stressful meeting, a skipped lunch, or a late workout, using lightweight, user-driven tagging. The insight behind her work is that current tools ask people to input a lot of information while giving very little back: the result is generic graphs that are easy to ignore. Esther’s approach aims to make personal data feel genuinely personal, surfacing patterns that are meaningful and actionable.

The goal is to give people a clearer window into their own health, one where they can spot, for example, that their sleep suffers after late-night screen time, or that their stress peaks mid-week before they even feel it. By putting richer, more contextual insights directly in users’ hands, Esther’s work moves wearable technology closer to what it has always promised: helping people make real, informed decisions about their own wellbeing.

Previous spotlights

As generative AI produces information—code, essays, analysis, videos—at an unprecedented scale and speed, we face a critical bottleneck: the human capacity to verify alignment and make sense of it all. To address this, Ziwei is moving the paradigm from “AI that generates” toward “AI that co-creates.” A key theme of his research is AI-resiliency, the design of interfaces that ensure potential AI errors or contextually inappropriate decisions are easily noticed, judged, and recovered from.

This vision has taken shape in his recent publications. In his CHI 2024 work, Ziwei proposes GP-TSM, a novel LLM-powered text rendering technique that improves reading efficiency while minimizing the risks of AI-generated summaries. He further expanded this to large-scale document analysis with AbstractExplorer (UIST 2025), which leverages LLMs and human cognitive theories to help users navigate patterns across massive text corpora without losing the context of individual documents.

Currently, Ziwei is rethinking the fundamental interfaces of human-AI interaction. He is developing generative controls to allow users to intuitively audit and refine AI outputs. He is also extending the AI-resiliency framework to autonomous agents. By designing mechanisms to help users anticipate the consequences of an agent’s actions before they are finalized, he aims to empower users to remain in control in increasingly complex, agent-driven workflows.
Over the next decade, communities around the world will decide which neighborhoods get protected from floods, which farms get water in a drought, and which areas are left most exposed to heat waves. The World Meteorological Organization reports that the last 11 years have been the warmest on record, so the choices we make now about how to respond to climate change will shape our future for generations. AI is becoming more involved in these decisions– raising questions about whether we should trust these algorithms to help make choices that can change people’s lives and environments.

Hongjin’s research looks closely at what happens when AI is used “for social good” in real communities, especially around climate change. Instead of seeing AI as a neutral tool, she studies AI for Social Good (AI4SG) projects as socio‑technical systems: who designs them, who benefits, and whose voices get left out. Drawing on Human‑Computer Interaction for Development (HCI4D) and Data Feminism, she investigates how power imbalances can cause community organizations’ goals to be sidelined, even in projects that claim to help them.

To address this, Hongjin is looking to work with community groups to co‑design and co‑evaluate new technologies for social good. In practical terms, that means bringing educators, activists, and local residents into the design process, and asking what kinds of AI tools would actually serve their priorities. One direction explores the idea of giving “a voice to nature”: for example, AI systems that translate data about rivers, forests, or air quality into stories, visualizations, or conversations that people can understand and act on.

Recent Publications

BRIDGE: Borderless Reconfiguration for Inclusive and Diverse Gameplay Experience via Embodiment Transformation

Hayato Saiki, Chunggi Lee, Hikari Takahashi, Tica Lin, Hidetada Kishi, Kaori Tachibana, Yasuhiro Suzuki, Hanspeter Pfister, Kenji Suzuki
CHI ’26 Best Paper Award, April 13-17, 2026, Barcelona, Spain

Training resources for parasports are limited, reducing opportunities for athletes and coaches to engage with sport-specific movements and tactical coordination. To address this gap, we developed BRIDGE, a system that integrates a reconstruction pipeline, which detects and tracks players from broadcast video to generate 3D play sequences, with an embodiment-aware visualization framework that decomposes head, trunk, and wheelchair base orientations to represent attention, intent, and mobility. We evaluated BRIDGE in two controlled studies with 20 participants (10 national wheelchair basketball team players and 10 amateur players).The results showed that BRIDGE significantly enhanced the perceived naturalness of player postures and made tactical intentions easier to understand. In addition, it supported functional classification by realistically conveying players’ capabilities, which in turn improved participants’ sense of self-efficacy. This work advances inclusive sports learning and accessible coaching practices, contributing to more equitable access to tactical resources in parasports.

MnemoMaker’s co-creation workflow for generating and adapting AI-generated verbal and visual mnemonics.

MnemoMaker: Creator, Curator, or Something Else? Exploring Human-AI Mnemonic Co-Creation

Olivia Seow, Elena Sajno, Dongho Shin, Pattie Maes, Samantha W. T. Chan
CHI ’26, April 13-17, 2026, Barcelona, Spain

MnemoMaker is an interactive system enabling learners to co-create personalized verbal and visual mnemonics with AI assistance. Through iterative design with 87 participants across three phases,we found that likeability strongly predicts perceived memorability(𝑅2 = .588, 𝑝 < .001), with effective mnemonics characterized as creative, imaginable, and emotionally engaging. Users reported strong ownership over AI-assisted creations (5.20/7) despite attributing them nearly equally to themselves and the system (3.54 on a 1-7scale from “system’s contributions” to “my contributions”). How people engage with AI, not just what they create, shapes their sense of agency. Our demo invites CHI attendees to create vocabulary mnemonics in any language and discover their collaboration style(e.g., as Creators, Curators, Negotiators, Acceptors, or something in between). A live dashboard visualizes collaboration patterns across all attendees, turning individual experiences into collective insight about human-AI co-creation.

Bayesian Inverse Transition Learning: Learning Dynamics From Near-Optimal Trajectories

Leo Benac, Abhishek Sharma, Sonali Parbhoo, Finale Doshi-Velez
AISTATS ’26, May 2-5, 2026, Tangier, Morocco

We consider the problem of estimating the transition dynamics T∗ from near-optimal expert trajectories in the context of offline model-based reinforcement learning. We develop a novel constraint-based method, Inverse Transition Learning, that treats the limited coverage of the expert trajectories as a feature: we use the fact that the expert is near-optimal to inform our estimate of T∗. We integrate our constraints into a Bayesian approach. Across both synthetic environments and real healthcare scenarios like Intensive Care Unit (ICU) patient management in hypotension, we demonstrate not only significant improvements in decision making, but that our posterior can inform when transfer will be successful.

A Paradigm for Creative Ownership

Tejaswi Polimetla, Katy Ilonka Gero, Elena L. Glassman
CHI ’26, April 13–17, 2026, Barcelona, Spain

As generative AI tools become embedded in creative practice, questions of ownership in co-creative contexts are pressing. Yet studies of human-AI collaboration often invoke “ownership” without definition: sometimes conflating it with other concepts, and other times leaving interpretation to participants. This inconsistency makes findings difficult to compare across or even within studies. We introduce a framework of creative ownership comprising three dimensions- Person, Process, and System- each with three subdimensions, offering a shared language for both system design and HCI research. In semi-structured interviews with 21 creative professionals, we found that participants’ initial references to ownership (e.g., embodiment, control, concept) were fully encompassed by the framework, demonstrating its coverage. Once introduced, however, they also articulated and prioritized the remaining subdimensions, underscoring how the framework expands reflection and enables richer insights. Our contributions include 1) the framework, 2) a web-based visualization tool, and 3) empirical findings on its utility.

Funding AI for Good: A Call for Meaningful Engagement

Lin, Hongjin; Kawakami, Anna; D’Ignazio, Catherine; Holstein, Kenneth; Gajos, Krzysztof Z
CHI ’26, April 13-17, 2026, Barcelona, Spain

Artificial Intelligence for Social Good (AI4SG) is a growing area that explores AI’s potential to address social issues, such as public health. Yet prior work has shown limited evidence of its tangible benefits for intended communities, and projects frequently face real-world deployment and sustainability challenges. While existing HCI literature on AI4SG initiatives primarily focuses on the mechanisms of funded projects and their outcomes, much less attention has been given to the upstream funding agendas that influence project approaches. In this work, we conducted a reflexive thematic analysis of 35 funding documents, representing about $410 million USD in total investments. We uncovered a spectrum of conceptual framings of AI4SG and the approaches that funding rhetoric promoted: from biasing towards technology capacities (more techno-centric) to emphasizing contextual understanding of the social problems at hand alongside technology capacities (more balanced). Drawing on our f indings on how funding documents construct AI4SG, we offer recommendations for funders to embed more balanced approaches in future funding call designs. We further discuss implications for how the HCI community can positively shape AI4SG funding design processes.

View all publications

CHARM