At UMSI, the future of AI is human-centered
Tuesday, 08/04/2026
Last Updated: Tuesday, 08/04/2026
By Noor HindiArtificial intelligence already shapes the ordinary systems people rely on every day. It influences how they get medical care, do their jobs and make sense of the world around them. But as AI moves deeper into everyday life, the most important questions are not just technical. They are human.
Human-centered AI is the practice of designing AI around human benefit, values and understanding, rather than model performance alone.
At the University of Michigan School of Information, researchers ask the questions that will define the future of AI: Who is this system for? Who is missing from its design? How can human and machine judgment complement each other? What values are hidden inside its measurements? And how will AI change the way people understand information, each other and themselves?
For UMSI researchers, human-centered AI is not simply about building more powerful models. It is about studying how AI enters people’s lives: a job application, a doctor’s visit, a search result, a chatbot. The question is what those systems make possible, what they obscure and who they leave behind.
“Keeping humans in the loop is not enough,” says Qiaozhu Mei, UMSI associate dean for research and innovation and a professor of information. “We have to make sure we are putting humans at the center.”
What does it mean to go beyond better models, and what should AI automate?
Across UMSI, AI research spans everything from natural language processing and data visualization to accessibility, ethics and community-engaged research. What ties that work together is not one method or one discipline, but a shared commitment to understanding AI in relation to the people it will most affect.
“The hardest technology problems remain stubbornly human, and the pace of change continues to accelerate,” says UMSI dean Andrea Forte. “We aim to prepare information leaders who can take this work beyond the university into institutions, industries and public life.”
That means studying AI not only as a technical system, but as something that shapes decisions, relationships, institutions and access to information. It means asking what people need from AI, where the technology can help, where it can cause harm and what human judgment should never be designed out.
The question of what should never be designed out is at the heart of UMSI professor Eytan Adar’s work.
Adar studies how automation can be used strategically. His research looks for places where AI and automation can help people without taking over the parts of a task that are meaningful, creative or necessary for learning.
“We do not like to automate everything,” Adar says. “There is a common philosophy in computer science that there is no problem you can’t make go away with more automation.”
In many domains, he says, full automation is not the goal. In education, if a system does everything for a student, the student may not learn. In art, if every part of the process is automated, the value and meaning of the creative act can change.
Adar points to his research with comic book artists.. Computer scientists had assumed artists wanted help automatically coloring line art. But after talking with artists, researchers realized the problem was more specific. Artists did not want to give up coloring, where they made creative decisions about light, mood and emotion. What they disliked was “flatting,” the tedious process of blocking out shapes before color is applied.
So the researchers automated that part.
“The desire to automate everything is clearly not the right instinct,” Adar says. “There is a strategic way of doing it where everybody can be happy. The artists get the parts they do not like done for them, but they still get to make creative choices.”
For Adar, that is the heart of human-centered AI: identifying real problems, understanding what people want and do not want, and aligning emerging technologies with human needs. He warns that risks come when systems are deployed too quickly and without enough attention to their consequences.
Who gets left out of the norm?
For UMSI associate professor of information Robin Brewer, human-centered AI begins with a deceptively simple principle: “Not forgetting the human in the AI process.”
Brewer studies accessibility, aging and technology use. In the context of disability, she sees possibilities for AI to make previously inaccessible tasks easier: image recognition for blind and low-vision users, or speech tools that support communication.
But her work also points to a deeper problem in AI design: systems are often built around norms. In disability and aging contexts, that means people outside those norms are frequently treated as outliers.
“AI is fancy statistics,” Brewer says. “The issue with statistics is that it is about norms and normalizing.”
If researchers and companies design around an imagined “average” user, Brewer’s research shows how disabled people, older adults and others at the margins may be poorly represented in training data, represented inaccurately in AI output, overlooked in design and excluded from the benefits of new tools.
The challenge for human-centered AI, Brewer says, is that making AI more accessible and inclusive often falls to smaller communities of accessibility and aging researchers, even though the consequences are much broader.
“Not profitable does not mean we should not work on it,” Brewer says.
How do we know if AI is fair or accountable?
UMSI assistant professor Abigail Jacobs studies what happens when human decisions are hidden inside technical systems.
AI tools are often described as neutral, objective or purely data-driven. But Jacobs’ research shows that these systems are shaped by choices: What data is used? What gets measured? Who defines success? Who decides whether a system is fair, safe or useful?
For Jacobs, those questions are central to human-centered AI.
“Human-centeredness means we need to think about what the human is interacting with, under what conditions, and what that means,” Jacobs says.
Her work looks at how AI systems can shift power and responsibility, especially when they are used by companies, governments or other institutions. In areas like hiring, healthcare, public services or access to information, technical decisions can become decisions that shape people’s lives.
One area of Jacobs’ recent research focuses on AI evaluations and benchmarks, which are the tests companies use to claim that systems are smart, fair, safe or capable. Those claims can sound objective–and indeed policymakers increasingly take these marketing claims as scientific and objective–but Jacobs says they often depend on choices about what counts as evidence.
“What we call things matters,” Jacobs says. “If you actually want to measure these things, and if you actually want to claim that you are making progress toward something, you should take that project seriously.”
That is why studying AI responsibly requires more than technical expertise alone. It requires researchers who can examine technology alongside law, sociology, economics, public health and the lived experiences of the people these systems affect.
For Jacobs, that is what makes UMSI different.
“Human-centered AI can only be done in a truly interdisciplinary environment,” Jacobs says.
At UMSI, she says, researchers can build bridges across sociology, organizational studies, law, business and health: connections essential for studying AI’s real-world impact.
How does AI change the way we communicate?
For UMSI associate professor David Jurgens, human-centered AI begins with language.
Jurgens studies natural language processing which focuses on how machines understand language.
“It is human-centered from the start because we are looking at text that people are writing, and most text is written by people to other people,” Jurgens says. “It is the way we communicate our ideas.”
His research builds models that can better understand the social information embedded in language: how people relate to one another, how they adapt their speech to different contexts, and how norms around power, status, age and emotion shape communication.
For AI systems to communicate effectively with people, Jurgens says, they need to understand more than words. They need to recognize frustration, confusion, anger and social expectations.
The concept is often called theory of mind: the ability to form a sense of what someone else may be thinking or feeling, and to respond in a way that fits the situation. In human conversation, people do this constantly. They adjust their tone, language and response based on whether someone seems confused, upset, uncertain or angry.
Often, Jurgens says, researchers want social models to have some version of that ability: to understand not just what a person said, but what they may be feeling or needing in the moment.
But Jurgens also sees one of the most profound risks of AI in a different possibility: that AI systems could begin to substitute for human relationships.
“I think the biggest risk is not AI replacing our jobs,” Jurgens says. “I think it is AI replacing our friends.”
As chatbots become more common sources of companionship, advice and emotional support, Jurgens says researchers need to ask how to build socially responsible AI without replacing the human relationships people need.
“The problem people have with information these days is that there is too much of it, and it is not clear what to trust,” he says. “It is urgent that we design responsible tools that use AI to help people wade through the thicket of information, find what they can trust, understand it and take action based on it.”
For students interested in AI, Jurgens says technical skill alone is not enough. UMSI’s emphasis on how these systems affect people helps students understand not only how to build AI tools, but how to use them responsibly.
“If you are just learning AI and not learning anything about the people you are helping serve, that is the gap,” he says.
What makes UMSI ready for this moment?
Together, the researchers show why UMSI is positioned to lead in human-centered AI. The school brings together technical expertise, social science, design, ethics and policy: fields often separated elsewhere, but deeply connected in the real world.
AI does not live only inside code. It lives in classrooms, hospitals, workplaces, search results and everyday decisions. Understanding it requires more than one discipline.
That interdisciplinarity is also what drew Adar to UMSI rather than a traditional computer science department.
“It is hard to find places that have a true intersection of those things,” Adar says. “If you care about a topic, you can tackle it from different directions.”
For Mei, that combination — technical, human and community-driven — is what makes UMSI distinctive.
“It is really hard to find a unit with all these technical, human and community-driven aspects,” Mei says.
RELATED
Check out UMSI’s undergraduate minor in human-centered AI and an AI concentration in our online Master of Applied Data Science degree.
