2022 MIT IDE Annual Conference

• Devin Cook – Associate Director, MIT IDE

How can we build data and AI ecosystems in which all partners — citizens, companies, and government — are winners? 

Data and AI are now primary means of production, along with capital and labor. Scalable, sustainable, and socially acceptable data and AI ecosystems require new ways to think about data ownership, new ways to quantify impact, and new ways to manage trade. The Data-Driven Societies Research Group aims to explore and establish equitable frameworks and infrastructure upon which our digitally-enabled world is based.

Alex “Sandy” Pentland leads this Research Group.

• Esteban Moro – Visiting Professor at MIT Media Lab and MIT Connection Science, IDSS

Reliable and timely information on socio-economic status and divides is critical to social and economic research and policy. Novel data sources from mobile communication platforms have enabled new cost-effective approaches and models to investigate social disparity, but their lack of interpretability, accuracy, or scale has limited their relevance to date. With access to 3.7 billion time-stamped and geo-referenced mobile traffic records in a major European country, we find profound geographical unevenness in mobile service usage—especially on news, e-mail, social media consumption, and audio/video streaming. The digital usage gap is so large that we can accurately infer the socio-economic status of a small area only from aggregated data traffic. Our results make the case for an inexpensive, privacy-preserving, real-time, and scalable way to understand the digital usage divide and, in turn, poverty, unemployment, or economic growth in our societies through mobile phone data.

• Renée Richardson Gosline – Human/AI Interface Research Group Lead, MIT IDE

How do we optimize the human and AI division of decision-making? When and where should we trust machines? How do we prevent bias in algorithms?

As organizations experience rapid digital transformation, the effective adoption of AI by employees, partners, and customers, becomes critical for competitive advantage. It’s imperative to examine when humans trust algorithms and how to develop mutual learning between humans and algorithms for optimal decision-making. We are exploring what symbiosis exists between human and AI and how the balance varies by task. Using experiments and mixed-method approaches, the Human-Machine Interface Research Group will proffer strategic, policy, and behavioral solutions.

Renée Richardson Gosline leads this Research Group.

• Renée Richardson Gosline – Human/AI Interface Research Group Lead, MIT IDE

As Algorithms, “artificial Intelligence” (AI), and Machine Learning transform the way we work and live, it is imperative that we investigate the impact on human judgment. Building upon research experiments at the IDE on trust in algorithms and bias in machine learning, we will discuss the promise and peril of automaticity in the use of AI. How can human bias interact with algorithmic bias, and what systems can we put in place to be more human-centered amidst digital transformations in our organizations? We will discuss current research and works in progress.

• Jerry Zhang – Ph.D. Candidate – Marketing, MIT Sloan School of Management

Technological advancements have provided consumers the option to choose services provided by “artificial Intelligence” (AI) or human beings. Previous research has shown that consumers display “algorithmic aversion” when AIs err. We further explore this phenomenon by testing the boundaries of this effect, and how it compares to human errors in a series of experiments. We find that participants quite rationally update their preferences according to cues about the advisor’s accuracy. The degree of algorithm aversion varies with the perceived size of the error made by the AI relative to the human forecaster. Overall, our findings suggest that, although users tolerate AI’s imperfection to a degree greater than previously thought, firms should strive to improve the accuracy of their AI as this affects algorithmic acceptance.

Enjoy a short break

• Dean Eckles – Social Networks & Digital Experimentation Research Group Lead, MIT IDE

How do new technologies and new means of social interaction and contagion impact productivity, consumer demand, political mobilization, and public health? How is human persuasion and behavior change caused by interactive technologies?

The IDE examines the profound impact social behavior and digital experimentation has on our lives, including how it affects aspects such as political mobilization, consumer demand, information sharing, worker productivity, and even stock prices. Through our research, we show how social networks can improve social and business policy through a dramatic new scientific understanding of human behavior. Recent projects include a study of how social media can help proactively manage epidemics through social media, as well as how mobilization and tracking impact voter democratization. We conduct our research using applied statistics, experimental design, machine learning, and causal inference.

Dean Eckles leads this Research Group.

• Maja Svanberg – MIT Graduate Student, Technology and Policy Program

Releases of data and statistics increasingly incorporate methods from differential privacy in order to preserve privacy. These releases are then used to make decisions. Because achieving differential privacy involves adding random noise to data, this can result in some units being randomly assigned to interventions, even in the presence of an ostensibly deterministic policy. For example, some jurisdictions provide voter information in non-English languages, based on certain population counts. Due to noise added to the United States Census data release, the jurisdiction’s counts may be pushed just over or just under the threshold for the policy. Here we develop methods for using this randomness for causal inference about the effects of those interventions. In particular, we characterize deterministic policies applied to differentially private data as corresponding to stochastic policies applied to the original private data. We develop inverse-probability weighted estimators, which make use of the probability a unit is randomized to treatments by the privacy mechanism. We then describe techniques that can make these estimators themselves differentially private. Using simulations based on the U.S. Census Bureau’s American Community Survey, we evaluate how these estimators can be used to estimate how translated voter information affects voter turnout.

• Eaman Jahani – Postdoctoral Associate, UC-Berkeley

• Yuan Yuan – Professor, Purdue University and Research Fellow, MIT IDE

Mobile devices and applications provide new opportunities to promote healthy physical activity. Existing applications and research trials have made use of financial, social, and prosocial incentives for physical activity. However, little is known about the effects of prosocial incentives in particular. Here we provide evidence on effects of prosocial incentives on promoting physical activity. In a field experiment with more than 40 million users on a giant social platform in Asia, we remind users to interact with a “step donation” feature that associates users’ step counts with third party donations to charities; we estimate that interacting with this feature causes users to take over 1,000 additional steps in the short run. To supplement these experimental results, we further conduct matching on observational data to compare the effects of prosocial (step donation) and social incentives (via a feature where users may be encouraged by social comparisons of steps among friends), yielding evidence that the prosocial incentive may have larger effects in our setting. Finally, we discuss plausible prosocial motivations that drive step donation engagement and use a regression discontinuity design to identify users likely driven by warm glow giving. Overall, our study shows the potential of leveraging prosocial incentives to encourage physical activity and sheds light on possible further improvements on the design of prosocial incentives.

• David Rand – Misinformation & Fake News Research Group Lead, MIT IDE

Why do people share inaccurate content online? What can be done to reduce the spread of misinformation and fake news?

Cognitive science research and digital field experiments are tools to advance the fight against online misinformation. We shed light on the basic science of how people decide what to believe and share, and leverage these insights to design anti-misinformation interventions. We then run experiments to test the impact of these interventions, and collaborate with technology companies like Google and Facebook to translate our science into real-world impact. Through the Misinformation & Fake News Research Pillar, we drive transparency in our digital lives.

David Rand leads this Research Group.

• Jennifer Allen – PhD Candidate, MIT Sloan School of Management

There is a great deal of interest in the role that partisanship, and cross-party animosity in particular, plays in interactions on social media. Most prior research, however, must infer users’ judgments of others’ posts from engagement data. Here, we leverage data from Birdwatch, Twitter’s crowdsourced fact-checking pilot program, to directly measure judgments of whether other users’ tweets are misleading, and whether other users’ free-text evaluations of third-party tweets are helpful. For both sets of judgments, we find that contextual features – in particular, the partisanship of the users – are far more predictive of judgments than the content of the tweets and evaluations themselves. Specifically, users are more likely to write negative evaluations of tweets from counter-partisans; and are more likely to rate evaluations from counter-partisans as unhelpful. Our findings provide clear evidence that Birdwatch users preferentially challenge content from those with whom they disagree politically. While not necessarily indicating that Birdwatch is ineffective for identifying misleading content, these results demonstrate the important role that partisanship can play in content evaluation. Platform designers must consider the ramifications of partisanship when implementing crowdsourcing programs. For more, see paper PDF and tweet thread.

• Adam Berinsky – Professor, MIT and Director, MIT Political Experiments Research Lab

The spread of misinformation online is a global problem that requires global solutions. To that end, we conducted an experiment in 16 countries across 6 continents (N = 33,480) to investigate predictors of susceptibility to misinformation and interventions to combat misinformation. In every country, participants with a more analytic cognitive style and stronger accuracy-related motivations were better at discerning truth from falsehood; valuing democracy was also associated with greater truth discernment whereas political conservatism was negatively associated with truth discernment in most countries. Subtly prompting people to think about accuracy was broadly effective at improving the veracity of news that people were willing to share, as were minimal digital literacy tips. Finally, crowdsourced accuracy evaluation was able to differentiate true from false headlines with high accuracy in all countries. The consistent patterns we observe suggest that the psychological factors underlying the misinformation challenge are similar across the globe, and that similar solutions may be broadly effective. For more, see paper PDF and tweet thread.

Enjoy lunch and networking

• John Horton – Labor Economics & Online Marketplaces Research Group Lead, MIT IDE

Professor Horton’s research focuses on the intersection of labor economics, market design, and information systems. He is particularly interested in improving the efficiency and equity of matching markets. This session is dedicated to those topics, with members of our research team presenting their primary research projects.

• Hong Yi Tu Yi – PhD Candidate, MIT Sloan School of Management

While there is widespread recognition that the remote work rate surged during the coronavirus pandemic, there is disagreement about the extent of this change. To address this limitation, we field a new, nationally-representative survey: the Remote Life Survey (RLS). After constraining the sample to working respondents who were employed before and during the pandemic, we find that in October, 2020, 31.6% of this continuously employed workforce always worked from home and 22.8% sometimes or rarely worked from home, totalling 53.6%. We compare our results with alternative measurement approaches, focusing on five factors: (a) differences in the selection of respondents among mail versus web-based surveys, (b) differences in the inclusion of self-employed workers, (c) ambiguity that arises from the classification of remote versus non-remote work into discrete categories, (d) the industry mix of the sample, and (e) the exclusion of people who were already remote pre-pandemic. We find that explanation (e) explains the bulk of the difference in estimates between the Current Population Survey and other measures of remote work, underestimating the remote work rate by up to 33 percentage points. Overall, we estimate that about half of the US workforce currently works remotely at least one day each week.

• Emma van Inwegen – PhD Candidate, MIT Sloan School of Management

There are two ways to help employers hire: (a) expand their choice set by attracting more applicants or (b) help them choose among that choice set. We report the results of an experiment where employers in a large online labor market were given hiring assistance that could take either form, based on the determination of the helper. Broadly speaking, job openings with few applicants were given recruiting help, while applicants with many applicants were given advice on who to hire. All were given general advice on the hiring process. We find that employers of treated job posts were over 10% more likely to make a hire than those in the control group. In decomposing the reasons for the increased hiring, we find evidence that both (a) and (b) were important, but with recruiting help being about twice as important than selection help. The hiring assistance program was remarkably cost-effective and a planner that could tax the wage bill at even just 2% could fund the intervention.

• Wang Jin – Research Scientist, MIT IDE

Using novel measures that capture the investment in cloud technology based on firms’ labor demand, we show that cloud adoption is associated with significant productivity gains for a large sample of US public firms between 2010 and 2019. This finding is robust to a range of cloud-related measures, samples, and specifications. Our research shows that the adoption of cloud-based technologies is associated with a 2.3-6.9% increase in sales. This effect is statistically significant, economically large, and increases over time, indicating potential organizational learning and complementary investment. This study is among the first to provide large-scale firm-level evidence concerning both short-term and long-term effects of cloud implementations on firm performance across a wide range of sectors.

Enjoy a short break

• Andrew McAfee – Co-director, MIT IDE

How can technology continue to drive positive progress in our world? Do we need to constrain growth or embrace it in order to improve both the human condition and the state of nature?

Technology is the most disruptive force in human history, ushering in a new era of unprecedented health, convenience, and prosperity. At the same time, we acknowledge the potential for peril as we innovate. We need to constantly reframe the conversation and mindfully reexamine our tools and our goals, not only asking “What will technology do to our economy, society, and environment?” but rather, “What will we do with technology?” The Tech for Good research group amplifies the IDE’s ongoing position that we need faster, more inclusive technology progress and access– not a slowdown –to meet today’s needs and those of the future. By examining the technologies, business models, and organizations that are shaping the future of the world, we aim to accelerate and ensure the broad, positive progress that innovation drives in our economy, society, and environment.

Andrew McAfee leads this research group.

• Erez Yoeli – Research Scientist and Director, Applied Cooperation Team (ACT), MIT Sloan School of Management

Tuberculosis (TB) is the number one killer amongst infectious diseases, having killed an estimated 1% of all humans, ever. Today, despite the fact that it has had a cure for over 70 years, it continues to infect over 10 million people a year, and kill over a million. Together with the digital health startup Keheala, we designed and tested a digital health platform to better support TB patients as they undergo treatment. In two separate randomized control trials (RCTs) with over 15,000 patients in Kenya, our platform reduced TB treatment non-completion by one-third to two-thirds, while meaningfully reducing deaths and improving medication adherence. I’ll tell you about our ongoing collaborations with machine learning (ML) experts at MIT Sloan to further improve our interventions by incorporating ML algorithms that will allow us to prioritize access and outreach to those patients who need it most.

• Avinash Collis – Assistant Professor, University of Texas at Austin and Digital Fellow, MIT IDE

Policy and communication responses to COVID-19 can benefit from a better understanding of people’s baseline and resulting beliefs, behaviors, and norms. From July 2020 to March 2021, researchers at the IDE, in collaboration with JHU, WHO, and Meta, fielded a global survey on these topics in 67 countries yielding over 2.0 million responses. This talk provides some results from the survey. For example, we find that providing normative information (how many others in your country are getting the COVID-19 vaccine) in vaccine communication increases intentions to accept a COVID-19 vaccine.

• Neil Thompson – Research Scientist, MIT CSAIL and MIT IDE

As Moore’s Law slows, companies are increasingly turning to algorithm improvement as a way to fuel new applications. But how much scope for improvement is there? In this talk, I will discuss the first quantitative estimates made for the overall pace of algorithm progress. I will also discuss the future of algorithm improvement, including how it will shape quantum computing.

• Sinan Aral, Director, MIT IDE

• Sinan Aral, Director, MIT IDE
• David Verrill, Executive Director, MIT IDE