When Mike Koval, the police chief of Madison, Wisconsin, abruptly resigned on a Sunday in September 2019, the community’s relationship with its men and women in blue was already strained. Use-of-force issues hung over the department after the killing of a Black teenager in 2015. Then, months before Koval left, another Black teenager, in the middle of a mental health crisis, was beaten on the head by an officer while being restrained by three others.
The process of selecting a new police chief followed a standard formula. A five-person team of mayor-appointed, city-council-approved commissioners would make the ultimate decision, allowing for public comment beforehand. But this time, the commissioners wanted that public input to involve more of the local community than just the folks who regularly appeared at town-hall-style meetings.
To gather more meaningful community feedback based on “lived experiences,” the commission took a new approach in which small groups of citizens—many from Madison’s most underheard neighborhoods—were brought together in a nonthreatening environment. Facilitators guided people who differed in age, ethnicity, gender, and socioeconomic status through intimate discussions on topics including what their own relationships with the police were like; whether they trusted or feared them; how they’d seen officers interact with kids and adults; and what type of training they thought police should receive to deal with stressful situations.
“The way we’re speaking with others is fundamentally broken. In every measurable way, things are getting more fractured and polarized.”
These conversations were recorded as part of an initiative called the Local Voices Network (LVN), which worked closely with the nonprofit Cortico and MIT’s Laboratory for Social Machines (LSM), headed by Professor Deb Roy. What made the process unique—and a potential model for other municipalities—was what happened next.
With help from machine-learning technology that Roy and an interdisciplinary team had developed over the past five years, MIT researchers sifted through hundreds of hours of audio to define topics and summarize larger conversations into snippets of text. By using this technology to augment human listening, the researchers were able to highlight parts of the conversations and identify the themes of greatest concern. The insights of 48 people in 31 different conversations were highlighted. The topics that emerged as common concerns became the basis for interview questions asked of the candidates to succeed Koval. Of the six final questions put before the four finalists, three came directly from the community conversations.
The facilitated work in Madison was a natural extension of Roy’s research in social media analytics. The scope of this work was further advanced when, in January 2021, MIT announced that the Laboratory for Social Machines would be expanded into an Institute-wide Center for Constructive Communication (CCC) based within the MIT Media Lab. The center will continue to work closely with Cortico, which Roy currently chairs. The two entities are now working hand in hand on building, as Roy says, “power tools” for democracy.
In Madison, thanks to tools like those, “we were able to actually uplift the specific concerns of a variety of members of the community,” says Colleen Butler, former director of capacity building at Cortico.
According to Roy, that’s how civic dialogue is supposed to work: various voices learning from each other to bridge divides and inform public policymaking. Instead, what he currently sees is a fragmented, reactive, angry world where vitriol and provocation score more points than conversation and understanding.
“The way we’re speaking with others is fundamentally broken,” he says. “In every measurable way, things are getting more fractured and polarized.”
For more than two decades, Roy has been deeply immersed in studying the complexity of human communication. Today, by combining that study with work on social-impact technology, he hopes to foster more constructive personal connections and enhance civic discourse. His aim is to find much-needed civility and common ground both in person and in social networks.
Most parents-to-be obsess over necessities like the crib, the bottles, and the pacifiers. Deb Roy had another item on his list: audio equipment.
In 2005, just before his son was born, Roy outfitted his home with 11 video cameras and 14 microphones. Over three years, he collected data—90,000 hours of video, 140,000 hours of audio—on how familial interactions affected his son’s speech development. Dubbed the Human Speechome Project, it built on Roy’s PhD dissertation, which focused on developing machine-learning models of human language. (He gave a TED talk about the experience in 2011.)
Roy’s key insight from the project was the notion of recurrent shared contexts. Parents don’t generally talk to their infants about objects or people not in the room. To foster language learning, it’s more helpful to use words in reference to something the infants and caregivers can perceive or participate in together. Roy wondered where else that sort of phenomenon might be found. Michael Fleischman, a PhD student in his research group, had an idea: the way people talk about TV. It was only a couple years after Twitter was founded, in 2006, that Roy and Fleischman discovered there were social media users who talk about television shows and commercials airing in real time, without even knowing each other.
“That’s how we ended up looking at tweets and other social media that were about what was on television,” says Roy. “You have this shared context. People tuned in to a live broadcast, and then talked to one another or just broadcasted, into the ether, reactions.”
He and Fleischman thought this was the basis for a good business idea. Advertisers have large research budgets for the purpose of figuring out how to help them connect with would-be consumers. In 2008, Roy took an extended leave from MIT, and the pair founded Bluefin Labs, a social analytics startup, to help companies analyze what everyday people were saying about television programs and advertising. Using algorithms, the startup could pick out millions of online comments made about a show or commercial in the hours immediately after it aired. Seeing that sort of information could then help networks and companies understand what was resonating with audiences, especially in the ever-growing online sphere.
“Companies that figure this out will thrive in the next 10 to 15 years. Companies that don’t will fail,” said a Nielsen executive quoted in a profile of the company published in MIT Technology Review in 2011.
Bluefin Labs was acquired by Twitter in 2013 for $100 million. For Roy, it served as a jumping-off point to his current work. He took a four-year role as Twitter’s chief media scientist, but he also went back to MIT.
“I knew that my long-term goal was to return to research,” he says. “My interest was to create a new kind of lab which could straddle the incredibly rich environment of doing explanatory and fundamental research with the skill set and all the things we did at Bluefin and Twitter.”
Forget analyzing the semantic patterns of the online world to figure out whether people liked a product being hawked during a commercial break: Roy wanted to take what he had learned at Bluefin, where he’d translated research into practical products and services, and apply those findings for noncommercial societal benefit. That’s when, in 2014, he set up the LSM at the Media Lab, with Twitter as a founding partner and main funder. He tapped Russell Stevens, a friend and previous advisor at Bluefin with a background in media and marketing, to help establish the lab.
What the researchers discovered this time when they examined tweets and other social media posts was something wholly different from what they’d seen in the world of entertainment TV: a crumbling social context instead of a cohesive one. After the Boston Marathon bombing, rumors spread like wildfire. During the 2016 presidential election, unverified reports were shared widely. Big news events came and went, playing out for all to see, but people reacted differently depending on what they heard and what they believed.
Through research at the lab, Roy, Stevens, and the LSM team tried to make sense of it—even going so far as to analyze millions of tweets to discern how false news spread through Twitter. (The resulting paper, which Roy coauthored, appeared on the cover of Sciencein 2018.) But to actually bridge those social divides, collaborators at the lab realized, they had to marry real-life conversations with the computational social science started at Bluefin and further developed at the LSM.
“If we really wanted to understand why we may be fragmenting into isolated tribes, we actually had to go talk to people,” Stevens says. “That’s the only solution.”
Finding common ground
Bringing conversations in the online world back to earth, so to speak, was Roy’s purpose in creating the Center for Constructive Communication. The announcement that introduced the new center characterized it as an “evolution” of the LSM. Unlike the LSM, though, it has a mandate to reach beyond academia—to bring the tools of data-driven analytics to bear on conversations about society, culture, and politics, and then to see where connections between people can be made.
“A democracy can’t function if the public is so divided and unable to listen to each other,” says Ceasar McDowell, the center’s associate director. “What we find out is that people aren’t as far apart as you think, but they don’t have the space where they feel that they will be heard and listened to in order to find that connection.”
That’s where Cortico comes in. Founded in 2016, with Roy and Stevens as two of the three cofounders, the nonprofit aimed primarily to facilitate on-the-ground conversations—first with the social tools that the LSM was developing, and now with interpersonal technologies being created by CCC and Cortico. CCC, which leads research in analytics and design research, partners with Cortico to develop prototype translations of research that can be tested with field partners—often local, grassroots organizations. Cortico then integrates findings from successful pilot programs into the LVN platform, which it independently develops and operates.
Can the marriage of real-life conversations with advanced digital technology put us on the road to becoming better citizens? Professor Deb Roy thinks so.
That platform, Cortico’s core initiative, is where the audio from these types of community conversations gets stored. Analytics tools—similar to what Bluefin Labs pioneered a decade ago—sift through the talk to find the common ground, and then to amplify those representative perspectives. Audio transcripts are made, and as the computer goes through the text, it picks out key points from conversations. Afterward, anyone can go back and listen to a particular segment to get the full context. CCC calls it “sense-making.”
To Jacquelyn Boggess, one of the commissioners involved in picking Madison’s police chief, the insights gained this way proved invaluable. Typically, the people who show up at town halls are telling commissioners which person to pick. The conversations with Madison’s citizens, she says, instead gave her a chance to hear how her decision might affect them.
“They’re not telling me who to choose. They’re telling me who they are and what they need,” Boggess says. “People told me stories of their lives and what goes on in their lives, as opposed to telling me who they think I should choose for police chief, and that was much more helpful.”
In late 2020, the LSM and Cortico used the LVN process to connect with citizens in Atlanta during the covid pandemic. As part of a collaboration with the Atlanta-based Task Force for Global Health, Cortico set up virtual group conversations of about six to eight people. They spoke about their fears of the new disease, the questions they had about staying safe, and their concerns about how covid testing was conducted. Cortico and LSM researchers (CCC was still a few weeks away from being announced) shared insights from those conversations with Black ministers, who they hoped could answer those questions for their congregations. In early 2021, LVN came in handy again as vaccines were being rolled out. “As the vaccine gained steam, we were able to tap into what folks were saying on the ground,” says Stevens. The platform gave residents a chance to express any concerns they had about receiving a vaccination; again, the team then spun up the results into messaging that could be distributed by trusted voices in various city neighborhoods.
Kick-starting a revolution
In the future, Roy hopes to expand the capabilities of CCC, Cortico, and LVN. Some of that will be accomplished through hardware designed to use during these group conversations: a portable recording device called a “digital hearth,” which is supposed to be a little more inviting than just a smartphone or microphone sitting in the center of a table. At the same time, Cortico is designing programs to train community organizers and volunteers on how to organize and facilitate local conversations.
“In general, online spaces, in order to meet certain design objectives and commercial objectives, tend to be disconnected from the in-person world,” Roy says. “We’re interested in weaving these back together.”
If a series of personal conversations could help Madison residents grapple with an issue as contentious as policing, and establish enough common ground to inform the questions asked in the official interviews, it seems to indicate that the process could work.
“I think it allows for greater transparency and community involvement—and, frankly, a more thoughtful process—than the more typical town hall type of meetings can offer,” says Butler.
Kick-starting a revolution in civic discourse is currently at the forefront of Roy’s mind. Right now, CCC is working on a new dashboard feature that would connect to information collected and organized in the LVN platform. A journalist set to moderate a public debate, for example, would be able to craft questions that address what’s on the minds of city residents as opposed to just picking a tweet or online comment at random. In fact, that’s exactly what is starting to happen with a new initiative in Boston.
Roy is careful to hedge his bets on how successful these new approaches can be. “The spaces for what we would call constructive conversation and constructive dialogue are shrinking,” he says. “I guess I know enough to realize it’d be naïve to think we’re going to fix that.”
Still, the tools he’s creating are unquestionably a start.
The gene-edited pig heart given to a dying patient was infected with a pig virus
The first transplant of a genetically-modified pig heart into a human may have ended prematurely because of a well-known—and avoidable—risk.
Saudi Arabia plans to spend $1 billion a year discovering treatments to slow aging
The oil kingdom fears that its population is aging at an accelerated rate and hopes to test drugs to reverse the problem. First up might be the diabetes drug metformin.
Yann LeCun has a bold new vision for the future of AI
One of the godfathers of deep learning pulls together old ideas to sketch out a fresh path for AI, but raises as many questions as he answers.
The dark secret behind those cute AI-generated animal images
Google Brain has revealed its own image-making AI, called Imagen. But don't expect to see anything that isn't wholesome.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.