Media manipulation, POTATO and pandemic parenting: UMSI research roundup
University of Michigan School of Information faculty and PhD students are creating and sharing knowledge that helps build a better world. Here are some of their recent publications.
“Challenges and Opportunities in Information Manipulation Detection: An Examination of Wartime Russian Media”
arXiv, October 2022
Chan Young Park, Julia Mendelsohn, Anjalie Field and Yulia Tsvetkov
NLP research on public opinion manipulation campaigns has primarily focused on detecting overt strategies such as fake news and disinformation. However, information manipulation in the ongoing Russia-Ukraine war exemplifies how governments and media also employ more nuanced strategies. We release a new dataset, VoynaSlov, containing 38M+ posts from Russian media outlets on Twitter and VKontakte, as well as public activity and responses, immediately preceding and during the 2022 Russia-Ukraine war. We apply standard and recently developed NLP models on VoynaSlov to examine agenda setting, framing, and priming, several strategies underlying information manipulation, and reveal variation across media outlet control, social media platform, and time. Our examination of these media effects and extensive discussion of current approaches’ limitations encourage further development of NLP models for understanding information manipulation in emerging crises, as well as other real-world and interdisciplinary tasks.
“POTATO: The Portable Text Annotation Tool”
We present POTATO, the portable text annotation tool, a free, fully open-sourced annotation system that 1) supports labeling many types of text and multimodal data; 2) offers easy-to-configure features to maximize the productivity of both deployers and annotators (convenient templates for common ML/NLP tasks, active learning, keypress shortcuts, keyword highlights, tooltips); and 3) supports a high degree of customization (editable UI, inserting pre-screening questions, attention and qualification tests). Experiments over two annotation tasks suggest that POTATO improves labeling speed through its specially designed productivity features, especially for long documents and complex tasks. POTATO is available at this https URL and will continue to be updated.
“Subgroup formation in human–robot teams: A multi-study mixed-method approach with implications for theory and practice”
Human–robot teams represent a challenging work application of artificial intelligence (AI). Building strong emotional bonds with robots is one solution to promoting teamwork in such teams, but does this come at a cost in the form of subgroups? Subgroups – smaller divisions within teams – in all human teams can undermine teamwork. Despite the importance of this question, it has received little attention. We employed a mixed-methods approach by conducting a lab experiment and a qualitative online survey. We (a) examined the formation and impact of subgroups in human–robot teams and (b) obtained insights from workers currently adapting to robots in the workplace on mitigating impacts of subgroups. The experimental study (Study 1) with 44 human–robot teams found that robot identification (RID) and team identification (TID) are associated with increases and decreases in the likelihood of a subgroup formation, respectively. RID and TID moderated the impacts of subgroups on teamwork quality and subsequent performance in human–robot teams. Study 2 was a qualitative study with 112 managers and employees who worked collaboratively with robots. We derived practical insights from this study that help situate and translate what was learned in Study 1 into actual work practices.
“Advancing digital patient-centered measurement methods for team-based care”
Objectives: To conceptualize new methods for integrating patient-centered measurement into team-based care.
Methods: A standalone portal was introduced into a rural clinic to support conceptualization of new methods for integration of patient-centered measurement in team-based care. The portal housed mental health-related online resources, three patient-reported measures and a self-action plan. Six providers and four patients used the portal for four months. Our data collection techniques included clinic discussions, one-on-one interviews, workflow diagrams and data generated through the portal. Analysis was supported through coding interview transcripts, looking across multiple sources of research data and research team discussions.
Results: Our research team conceptualized five team-based patient-centered measurement methods through this study. Patient-centered measurement Team Mapping offers a technique to provide greater clarity of care-team roles and responsibilities in data collected through patient-centered measurement. Longitudinal Care Alignment can guide the care-team on incorporating patient-centered measurement into ongoing provider–patient interactions. Digital Tool Exploration can be used to evaluate a team's readiness toward digital tool adoption, and the impact of these tools. Team-based quality improvement serves as a framework for engaging teams in patient-centered quality improvement. Shared learning is a method that promotes patient-provider interactions that validate patient's perspectives of their care.
Conclusion: The portal illuminated new methods for the integration of patient-centered measurement in team-based care. The first three proposed patient-centered measurement methods provide ways to assess how a clinic can incorporate patient-centered measurement methods into team-based care. The latter two methods focus on the aim of patient-generated data in which patient's values and perspectives are represented and quality of patient-centered care can be evaluated. Further testing is needed to assess the utility of these patient-centered measurement methods across different clinical settings and domains.
“Anti-racism for Freethinkers: Cultivating a Mindset for Curiosity and Scientific Inquiry in the Context of Racial Equity and Social Justice”
TechTrends, December 2022
The term “freethinking” originated in the 17th century to describe inquiry into beliefs which were accepted unquestioningly. Feminists such as Mary Wolstonecraft, abolitionists like Frederick Douglass, and novelists such as Mark Twain and Zora Neal Hurston are among the many who dared to simultaneously challenge religious dogma, patriarchal convention and racialized boundaries. But today the concept has been appropriated by the alt-right. A broad spectrum ranging from hardened white supremacists to those with more centrist tendencies have developed a discourse that objects to any form of antiracism on the grounds that it runs counter to individualism and freethought. In this essay we suggest that this critique from the alt-right should not be dismissed. Rather, it should be the impetus to revitalize the connections between antiracism and the principles of freethinking. We map out some of the history in which these connections were previously established; the reason the connection was weakened, and the principles by which the confluence could be restored. We recount some initial experiments using educational technologies to support this framework.
“False Consensus, Information Theory, and Prediction Markets”
arXiv, November 2022
We study a setting where Bayesian agents with a common prior have private information related to an event's outcome and sequentially make public announcements relating to their information. Our main result shows that when agents' private information is independent conditioning on the event's outcome whenever agents have similar beliefs about the outcome, their information is aggregated. That is, there is no false consensus.
Our main result has a short proof based on a natural information theoretic framework. A key ingredient of the framework is the equivalence between the sign of the “interaction information” and a super/sub-additive property of the value of people's information. This provides an intuitive interpretation and an interesting application of the interaction information, which measures the amount of information shared by three random variables.
We illustrate the power of this information theoretic framework by reproving two additional results within it: 1) that agents quickly agree when announcing (summaries of) beliefs in round robin fashion [Aaronson 2005]; and 2) results from [Chen et al 2010] on when prediction market agents should release information to maximize their payment. We also interpret the information theoretic framework and the above results in prediction markets by proving that the expected reward of revealing information is the conditional mutual information of the information revealed.
“Wearing a High Heel and a House Shoe at the Same Time": Parents' Information Needs While Navigating Change in their Child's Behavioral Care”
Proceedings of the ACM on Human-Computer Interaction, CSCW, October 2022
Change is an inevitable part of a parent's role, whether due to their child's development, family life, or external events. To understand the information needs of parents navigating change, we studied the effects of the COVID-19 pandemic as a widely experienced disruption in the lives of parents and children. We interviewed 16 parents about their caregiving experience as the COVID-19 pandemic collapsed boundaries between home, school, and work. In particular, we asked about adjustments to behavioral care, or the social learning, supports, and interventions through which children develop social and emotional skills. We focused on parents of children already receiving accommodations and behavioral support from their school, to understand how disruptions in these services affected the role of the parent in meeting their child's individual needs. Applying role theory and the Kübler-Ross change curve, we describe the coping mechanisms that parents used to navigate the stages of change, as well as the information needs that remained unmet, despite their efforts. We discuss how practitioner-initiated and parent-centered supports can be designed around the lived experience of change, by accommodating a parent's capacity to accept and use help at different stages.
“An Approximation of Freedom: On-demand Therapy and the Feminization of Labor”
Proceedings of the ACM on Human-Computer Interaction, CSCW, October 2022
Platform labor and gig work have become key sites for understanding a nascent "future of work" hallmarked by informalization and digitization. A growing body of research emphasizes how experiences of platform work are mediated not only by algorithms and user interfaces, but also by gender, race, local cultures as well as labor hierarchies. Drawing from ongoing ethnographic research on the digital transformation of healthcare, we show how therapists' experiences of platform labor are centrally shaped by the historical and ongoing feminization of mental health work. Platforms reinscribe feminized labor conditions that are pervasive in the healthcare industry, and yet platform labor appears as “useful” to some therapists as they navigate a set of precarious career choices fundamentally structured by feminization. We use the analytic of the stopgap to describe platforms' two-fold reproduction of the status quo: first by offering an approximation of freedom to individual workers, helping to forestall a crisis of unsustainable work conditions; and second by reinscribing the same logics of exploitation in order to make labor scalable. This stopgap analytic reorients the focus away from the impact of the platforms technologies as such, toward the conditions that make stopgap solutions necessary for survival. It also points toward the importance of intervening in the conditions of exclusion and exploitation that help to create a market for platform stopgaps.
“Affective Learning Objectives for Communicative Visualizations”
IEEE Transactions on Visualization and Computer Graphics, September 2022
* Selected as a best paper of the Vis 2022 conference.
When designing communicative visualizations, we often focus on goals that seek to convey patterns, relations, or comparisons (cognitive learning objectives). We pay less attention to affective intents – those that seek to influence or leverage the audience's opinions, attitudes, or values in some way. Affective objectives may range in outcomes from making the viewer care about the subject, strengthening a stance on an opinion, or leading them to take further action. Because such goals are often considered a violation of perceived “neutrality” or are “political,” designers may resist or be unable to describe these intents, let alone formalize them as learning objectives. While there are notable exceptions – such as advocacy visualizations or persuasive cartography – we find that visualization designers rarely acknowledge or formalize affective objectives. Through interviews with visualization designers, we expand on prior work on using learning objectives as a framework for describing and assessing communicative intent. Specifically, we extend and revise the framework to include a set of affective learning objectives. This structured taxonomy can help designers identify and declare their goals and compare and assess designs in a more principled way. Additionally, the taxonomy can enable external critique and analysis of visualizations. We illustrate the use of the taxonomy with a critical analysis of an affective visualization.
“Brief amici curiae of Information Science Scholars filed”
Supreme Court Case Gonzalez vs. Google, January 2023
Nicholas J. Belkin & W. Bruce Croft; Pascal R. Chesnais, Matthew J. Mucklo & Jonathan A. Sheena; M.C. Davis, M.D. Linsky & M.V. Zelkowitz; Jon M. Kleinberg; Yumao Lu, Fuchun Peng, Xing Wei & Benoit Dumoulin; Thomas W. Malone, Kenneth R. Grant, Franklyn A. Turbak, Stephen A. Brobst & Michael D. Cohen; Paul Resnick, Neophytos Iacovou, Mitesh Suchak, Peter Bergstrom & John Riedl; J.J. Rocchio; Barry Schwartz; Alok Sinha; Bin Tan, Xuehua Shen & ChengXiang Zhai.
In enacting the Communications Decency Act, Congress found that the internet and the services it enables “represent an extraordinary advance in the availability of educational and informational resources to our citizens.” Yet access to all the world’s information has created a need for ways to find the right information.2 Section 230 encourages that process by providing certain immunities to interactive computer services when they provide users with access to information from others on their servers.
Section 230 provides that one type of interactive computer service is an “access software provider,” which it defines to include providers of tools that filter, pick, choose, search, subset, or organize content. Recommender systems do just that, by helping to identify content that might serve a user’s needs. These systems have long sought to provide personalized services, for example by relying on the interests of a particular user, their past actions, or the preferences of other users who appear to have similar interests. Work on such systems predates the Communications Decency Act, and modern recommender systems are based on similar principles. Because recommender systems filter, pick, and choose content to recommend to a user, the providers of the recommendations are access software providers.
Moreover, contrary to Petitioners’ position, Section 230 does not draw a distinction between computer systems that rely on explicit user requests for information and those that rely on implicit requests via a user’s actions. Many common functions protected by Section 230 are based on implicit requests. In a directly analogous way, YouTube provides users with access to its computer servers when it responds to the signals contained in users’ actions, regardless of whether they have formulated explicit search queries. Under the standard understanding of client-server architectures, a computer can act as a server even when a human user is not making explicit requests to it, and acts as an interactive “server” whenever it is receiving requests from a “client” computer program, which in the case of YouTube could be a web browser or a smartphone app.
Finally, search engines, too, provide recommendations. There is fundamentally no distinction between the rankings that search engines perform and the operations that recommendation systems perform: by ranking the search results provided in response to a query, a search engine recommends some results more highly than others. And at times the recommendations in these search rankings may depend on information that isn’t part of the user’s query. Nothing in Section 230, or in the way these systems are designed, supports distinguishing between liability for recommendations made by search engines and recommendations made by YouTube.
“#ActuallyAutistic Twitter as a Site for Epistemic Resistance”
ACM Transactions on Computer-Human Interaction, October 2022
The Internet has, for several decades, played a critical role in autistic self-advocacy and community building. This semi-autoethnographic, interpretivist study turns to #ActuallyAutistic Twitter to examine autistic concerns about autism research, how these concerns differ from those of autism researchers, and how autistics interact with autism research and researchers. I find that #ActuallyAutistic Twitter discourses align with the neurodiversity paradigm, while dominant autism discourses in the academy align with the medical model of disability. Though both orientations toward autism research sometimes share research priorities, they represent fundamentally irreconcilable approaches to these priorities and autism, more broadly. I explore how autistics on Twitter interact with non-autistic researchers and how the tenor of these interactions varies according to which research paradigm a particular researcher subscribes. I conclude with a discussion of how HCI researchers interested in autism can operationalize these findings by approaching their work through the framework of crip technoscience.