Cesar Hidalgo, Toulouse School of Economics, Director Center for Collective Learning, Founder of Datawheel.

To attend the colloquium talk please register using this link.

Why do people judge humans differently from machines? The role of agency and experience

People are known to judge artificial intelligence using a utilitarian moral philosophy and humans using a moral philosophy emphasizing perceived intentions. But why do people judge humans and machines differently? Psychology suggests that people may have different mind perception models for humans and machines, and thus, will treat human-like robots more similarly to the way they treat humans. Here we present a randomized experiment where we manipulated people’s perception of machines to explore whether people judge more human-like machines more similarly to the way they judge humans. We find that people’s judgments of machines become more similar to that of humans when they perceive machines as having more agency (e.g. ability to plan, act), but not more experience (e.g. ability to feel). Our findings indicate that people’s use of different moral philosophies to judge humans and machines can be explained by a progression of mind perception models where the perception of agency plays a prominent role. These findings add to the body of evidence suggesting that people’s judgment of machines becomes more similar to that of humans motivating further work on differences in the judgment of human and machine actions.

Time: 12:00 - 1:00 PM ET | 9:00 - 10:00 AM PT | 6:00 - 7:00 PM CET

Understanding how diverse policymakers govern biodiversity and cultural diversity is critical for addressing environmental challenges. In my talk, I will focus on Colombia, explaining the evolution of six decades of biodiversity governance (1959–2018) by analyzing policy mixes, actor roles, and ecosystem management. I will demonstrate how biodiversity has been mainstreamed into sectoral policies, particularly at the nexus of climate change and poverty, highlighting the need for effective coordination across sectors and actors. Additionally, I will explore Colombia’s biocultural diversity as a case study, showing how the synergies between nature-based and culture-based tourism offer a promising direction for ecotourism policies aimed at integrating biodiversity and cultural diversity. My research underscores the importance of integrating cultural and biological variables into tourism planning to create sustainable biocultural destinations that support both conservation and development goals, particularly in a megadiverse country like Colombia.

Time: 12:00 - 1:00 PM ET | 9:00 - 10:00 AM PT | 6:00 - 7:00 PM CET

This colloquium will explore new databases from Mexico on the link between environmental risk factors and breast cancer (BC) to reflect on how individuals, health specialists, business owners, and policymakers consider environmental risks in their decision-making processes to shape local and national public health policies. BC is the most frequent neoplasm in Mexico, contributing to almost 17% of malignant neoplasia in women. For the last 20 years the mortality rate of the disease has been increasing. Disparities in access to screening, diagnostic and medical treatment are structural determinants that impact on BC survival. The lack of population-based cancer registers in the country limits the possibility to know the behavior of the disease in younger women. About causal factors, less than 20% of the cases can be linked to inherited genes, suggesting that additional external exposures contribute to the burden of the disease. A recent research study reported novel connections between over 900 chemicals which are commonly found in food, air, water and personal care products, and BC. Mexico is a country in continuous industrialization, with flexible environmental regulations and dense population in urban-industrial neighborhoods. Several open questions remain regarding how to integrate this new evidence into BC prevention strategies, virtual assistance systems, risk management schemes, prediction algorithms, and policy design. The colloquium will outline the practical outcomes of the Social Hackathon initiative, which is aimed at addressing these issues. The Hackathon will take place in the lead-up to the EAAMO 24 conference in San Luis de Potosí.

Time: 12:00 - 1:00 PM ET | 9:00 - 10:00 AM PT | 6:00 - 7:00 PM CET

Recent advancements in natural language processing have led to the proliferation of large language models (LLMs). These models have been shown to yield good performance, using in-context learning, even on tasks and languages they are not trained on. However, their performance on African languages is largely understudied relative to high-resource languages. We present an analysis of four popular large language models(mT0, Aya, LLaMa 2, and GPT-4) on six tasks (topic classification, sentiment classification, machine translation, summarization, question answering, and named entity recognition) across 60 African languages, spanning different language families and geographical regions. Our results suggest that all LLMs produce lower performance for African languages, and there is a large gap in performance compared to high-resource languages (such as English) for most tasks. We find that GPT-4 has an average to good performance on classification tasks, yet its performance on generative tasks such as machine translation and summarization is significantly lacking. Surprisingly, we find that mT0 had the best overall performance for cross-lingual QA, better than the state-of-the-art supervised model (i.e. fine-tuned mT5) and GPT-4 on African languages.  Similarly, we find the recent Aya model to have comparable result to mT0 in almost all tasks except for topic classification where it outperform mT0. Overall, LLaMa 2 showed the worst performance, which we believe is due to its English and code-centric (around 98%) pre-training corpus. Our findings confirms that performance on African languages continues to remain a hurdle for the current LLMs, underscoring the need for additional efforts to close this gap.

Time: 12:00 - 1:00 PM ET | 9:00 - 10:00 AM PT | 6:00 - 7:00 PM CET.

Thousands of users consult digital archives daily, but the information they can access is unrepresentative of the diversity of documentary history. The sequence-to-sequence architecture typically used for optical character recognition (OCR) – which jointly learns a vision and language model - is poorly extensible to low-resource document collections, as learning a language-vision model requires extensive labeled sequences and compute. This study models OCR as a character level image retrieval problem, using a contrastively trained vision encoder. Because the model only learns characters’ visual features, it is more sample efficient and extensible than existing architectures, enabling accurate OCR in settings where existing solutions fail. Crucially, the model opens new avenues for community engagement in making digital history more representative of documentary history.

Time: 12:00 - 1:00 PM ET | 9:00 - 10:00 AM PT | 6:00 - 7:00 PM CET

The Asylum Seeker Advocacy Project (ASAP) is the largest membership organization of asylum seekers in history. Through ASAP’s participatory model, 600,000+ members have set collective priorities, pursued lawsuits, and led advocacy campaigns. And they have won—securing work permits for hundreds of thousands of immigrants! ASAP believes solutions will often have the highest impact when they are designed by the highly impacted. In this talk, Swapna Reddy, co-founder and Co-Executive Director of ASAP, will discuss opportunities, challenges, and lessons learned from engaging with a large member base to achieve meaningful change. Swapna will share concrete examples of how ASAP has engaged in participatory decision-making and collective action at scale.

Targeting is a central challenge in the design of humanitarian programs: given available data, how does one identify the individuals and households with the greatest need for assistance? Here we show that machine learning, applied to non-traditional data from satellites and mobile phones, can improve the targeting of anti-poverty programs. Our analysis is based on data from three field-based projects—in Togo, Afghanistan, and Kenya—that illustrate the promise, as well as some of the potential challenges, of this new approach to targeting. Collectively, the results highlight the potential for new data sources to improve humanitarian response efforts, particularly in crisis settings when traditional data are missing or out of date.

Cash transfer programs are the prime example of a social innovation that has scaled rapidly across the globe. Such programs often have multiple objectives, from providing short-term safety nets to the long-term ambition of breaking the intergenerational transmission of poverty. A lot of empirical evidence has been accumulated on the short term goals. 25 years after the start of the first national programs in Mexico and Brazil, we can also start taking stock of the long-term objectives. This presentation will first discuss the methodological challenges for studying long-term returns, focusing on common pitfalls and potential solutions. It will then lay out a framework to consider the different pathways and mechanisms towards long-term returns to clarify the type of outcomes and contextual variables to measure both on the intermediate and long run. Given the complexity and multi-dimensionality of the theory-of-change for long-term impacts, we then draw lessons from different studies, following the framework. We first highlight the evidence on long-term returns to conditional cash transfers (CCTs) that operate through returns on investments in education and skills. A lot of this evidence draws on the first generation of CCTs in Latin America, but we also link to the emerging evidence from other regions. We then turn to evidence on unconditional cash transfers (UCTs), which is particularly prominent for Sub-Saharan Africa, focusing on a wider set of investments and behavioral changes with evidence on impacts years after the end of the transfers.

While much social science to date on the digitization of agriculture has focused on large agribusiness, farms under two hectares account for 84% of farms worldwide and produce about 35% of the world's food. Many small farmers are located in the Global South, where land provides a source of sustenance and site of home. Understanding the implications of increasing digital connection is critical for understanding rural economies, societies, and politics. This talk draws on a decade of anthropological and participatory research with farmers, activists and policymakers in Myanmar to discuss how digital connection is shifting rural economics, culture, and politics, before and after the 2021 military coup.

Aurelio Nuño Mayer is a Visiting Scholar at Harvard University and the Former Minister of Public Education in Mexico (August 2015 to December 2017). Aurelio Nuño received a degree in political science and administration at the Universidad Iberoamericana, and he later earned a master's degree at University of Oxford (UK). He articulated one of the most comprehensive and ambitious education reforms in many decades in Mexico. The reform consisted of five structural changes: 1) a new national curriculum for the Twenty-First Century skills; 2) a new merit-based professional system for training, hiring and promoting teachers; 3) a new model of school organization; 4) a national strategy for inclusion and equity; and 5) a new model of governance within the education system. Currently, he is a visiting scholar at the Harvard Graduate School of Education and he is writing a book on the politics of education.

Public health in India is undergoing a digital transformation. Community health workers, newly equipped with smartphones, are now reporting beneficiary-level data with village level granularity, in real-time. This talk will explore how Khushi Baby, the technical support partner to the Department of Health in Rajasthan, India, and collaborators, are using this new big data, from over 70K community health workers, to drive insights at the beneficiary, health-worker, and community level.

Economists typically prescribe prices for guiding the allocation of scarce resources, arguing that the implicit selectivity of the price system helps allocate resources to those who value them the most. However, in many contexts, considerations such as fairness, equity, or consumption externalities provide arguments against using prices—and indeed, many ethicists and policy-makers opt for schemes that allocate resources free of charge to certain selected groups. The price system, they argue, directs resources to those who are able to pay the most, which may not match up with true needs or moral desert. The question of whether to use prices or priorities played out in the context of allocating vaccines during the Covid-19 pandemic. Although prices could help identify individuals with the highest private values for vaccines, most countries opted for a priority system with rationing. This paper derives the optimal scheme from economic primitives. The key insight is that while social considerations may indeed limit the role that prices play in the optimal mechanism, they are typically not sufficient to rule out prices completely. As a result, a priority system with rationing may coexist with a pricing scheme; such a hybrid mechanism allows the designer to leverage observable information while simultaneously screening for unobservable characteristics.

Sex workers sit at the intersection of multiple marginalized identities and make up a sizable workforce: the UN estimates that at least 42 million sex workers are conducting business across the globe. Sex workers face a unique and significant set of digital, social, political, legal, and safety risks; yet their digital experiences have received little study in the CS and HCI literature. In this talk we will review findings from a 2-year long study examining how sex workers who work in countries where sex work is legal (Germany, Switzerland, the UK) use technology to conduct business and how they have developed digital strategies for staying safe online and offline. We will then describe how these findings can inform broader conversations around internet governance, digital discrimination, and safety protections for other marginalized and vulnerable users whose experiences bisect the digital and physical.

Ed "Kingfish" Lada, Jr., President and CEO for Goodwill Keystone Area and Goodwill Keystone Area Foundation in Central and Southeastern PA, discusses how A.I. (in particular computer vision machine learning), has been explored within his non-profit organizations to modernize operations while also creating pathways to the future of workforce development. Goodwill is a $7b federated non-profits in North America that are the original social enterprise. In this session, Ed will share his journey doing R&D, Goodwill's path forward, and why it is imperative that the technology is viewed as an augmentation of the human experience, not the replacement of the human experience.

JustFix's work building anti-eviction web apps early in the pandemic led us to change our philosophy on how tech can best serve tenants. We experienced the mismatch between legislative and software dev cycles (not the way you'd think.), the limits of organizing within the system, and the power of open data to put external pressure on bad landlords and brokers. In this talk, we'll preview some of JustFix's new tool ideas that fit this "type three" paradigm of working outside existing systems to push for housing justice. The discussion will be moderated by Daniel Waldinger in bridging social, economic, and policy perspectives on public housing allocation.

The A.I. Industry has powered a futuristic reality of self-driving cars and voice assistants to help us with almost any need. However, the A.I. Industry has also created systematic challenges. For instance, while it has led to platforms where workers label data to improve machine learning algorithms, my research has uncovered that these workers earn less than minimum wage. We are also seeing the surge of A.I. algorithms that privilege certain populations and racially exclude others. If we were able to fix these challenges we could create greater societal justice and enable A.I. that better addresses people's needs, especially groups we have traditionally excluded. In this talk, I will discuss some of these urgent global problems that my research has uncovered from the A.I. Industry. I will present how we can start to address these problems through my proposed "A.I. For Good" framework. My framework uses value sensitive design to understand people's values and rectify harm. I will present case-studies where I use this framework to design A.I. systems that improve the labor conditions of the workers operating behind the scenes in our A.I. industry. I conclude by presenting a research agenda for studying the impact of A.I. in society; and researching effective socio-technical solutions in favor of workers.

Most of the big internet companies, such as Facebook, Google, or Twitter, generate their revenue from targeted advertising. To offer advertisers with advanced targeting capabilities, these companies collect large amounts of user data to build elaborate profiles. Based on these profiles an advertiser can then choose to target only, say, female Facebook users living in Norte de Santander who are aged 18-24, who used to live in Venezuela, and who have access to an iOS device. To help advertisers in planning their advertising campaigns and the related budget needs, the advertising platforms provide so-called audience estimates on how many of their users match the provided targeting criteria. In the example above, Facebook estimates that there are 1,800 matching monthly active users (as of Mid-December 2021). In this talk, I'll describe how, in close collaboration with different UN agencies, we're tapping into these audience estimates to (i) monitor international migration, (ii) track digital gender gaps, and (iii) map wealth inequalities. We consistently find that, despite fake profiles, and noise in the inference algorithms, data derived from the advertising platforms can provide valuable information that is complementary to other data sources. So data collected for the explicit purpose of selling advertisements and profit maximization can be repurposed for social good. At the same time, our work shows the risk of identifying vulnerable groups, rather than individuals, which is often not adequately considered in discussions focused on individual privacy. Furthermore, it raises questions on what the relationship between internet giants and statistical offices should be.

A great number of academic and industry initiatives seek to mitigate what are increasingly understood as ethical concerns in the development, implementation, and evaluation of algorithms in social contexts. Often, when confronted with questions of ethics within these socio-technical systems, engineers ,data scientists, and others turn to the field of Philosophy for help in trying to decide the balance between the right thing to do and what we can do. In this talk, Dr. Rodriguez presents her arguments for why most strands of moral philosophy are ill equipped to help us deal with the realities of algorithmic bias. Her arguments use evidence gleaned from years of professional experience in human services, her study of moral philosophy, as well as arguments put forth by a school of Black Philosophers. The aim of the talk is to introduce the audience to more contextual forms of philosophizing, ones grounded in the realities of marginalized existence.

As practising Data Science researchers and practitioners, the COVID-19 pandemic has highlighed both the need for data driven decision making and the reality of what it really takes to get to that point. It is not only about throwing data + model at a problem. It is about understanding the environment that one is in and then strategising on what might best work for that environment. In this talk I look back at some of the work we have done within responding to different challenges within both Data Science and Natural Language Processing. I place at the center people and how they are the important piece in our practice.

We study the effects of a reform that increased the stakes of the national exam at the end of high school for university enrolment in Spain. We find a negative effect of the reform on female test scores, driven by a worse performance in the exams for which the stakes increase the most, and driven by students expected to be top performers. The effect on test scores translates into significant changes in students' allocation to college: female enrolment in the most selective degrees declines, together with expected earnings.

We will discuss the following questions. What is democratic backsliding? What makes a democracy stable? Which processes potentially lead to instability of a democratic system? How can complexity science help us understand and mitigate democratic backsliding? The talk is based on this paper and this paper.

Colleges and universities are among those that have been most affected by the COVID-19 pandemic. Even with a potential end to the ongoing global restrictions in sight, the manner in which these institutions operate is likely to be permanently altered as a result of the significant impact of the virus. Most notably, a more traditional, siloed approach to research will play a key role in effectively meeting the needs of local communities. Additionally, the existing teaching-learning paradigm will need to be reimagined, keeping in mind the transformation of the educational landscape. This session will address key global trends in the education sector, as well as propose some ideas on how universities can adopt a more proactive approach to cope with these changes.

The Census is doing something new and interesting this year: implementing a differentially private algorithm called TopDown, so named because it works its way down the levels of a geographic hierarchy. The Census is already facing lots of scrutiny of its ability to provide a reliable count for marginalized communities under unprecedented political pressure from the White House and in pandemic conditions. Will this privacy protection mechanism result in erasure and undercounting for the same hard-to-count groups? What are the impacts on redistricting and the enforcement of the Voting Rights Act? I'll describe joint work with Aloni Cohen, JN Matthews, Bhushan Suwal, and Peter Wayner, incorporating reconstruction data from Mark Hansen and Denis Kazakov. No prior knowledge of U.S. politics required.

Bistra Dilkina is a Gabilan Associate Professor of Computer Science at the University of Southern California. She is also Co-Director of the Center for AI in Society (CAIS). Before that, Dilkina was an Assistant Professor in the College of Computing at the Georgia Institute of Technology and a co-director of the Data Science for Social Good Atlanta summer program. She received her PhD from Cornell University in 2012, and was a Post-Doctoral Associate at the Institute for Computational Sustainability until 2013. Dilkina is one of the junior faculty leaders in the young field of Computational Sustainability, and has co-organized workshops, tutorials, special tracks at AAAI and doctoral consortium on Computational Sustainability. Her work spans discrete optimization, network design, stochastic optimization, and machine learning.

This will be a (mostly) non-technical talk describing my experience working on this project. The paper is motivated by healthcare exchanges, where a designer can't directly set prices, but can regulate which providers enter the market. The two extremes to have in mind are typical employer exchanges in the US (which typically limit entry to few providers) or government exchanges (which typically do not). The paper provides a mathematical model to reason about tradeoffs between these two paradigms. I will describe the model and results from this paper, but mostly focus on my experience learning about healthcare through the MD4SG working group, from the perspective of someone with a CS theory background. The referenced work is joint with Meryem Essaidi (Princeton) and Kira Goldner (Columbia).

Operations research methods have been used to identify and evaluate solutions to the reconfiguration of public school attendance area boundaries for over fifty years. The talk will explore connections between evolving issues in public education and advances in optimization, computing and geographic information systems, beginning with early work motivated by Supreme Court decisions to desegregate schools. We will also discuss how the limitations of early models and solution approaches hindered their applicability. The years since have seen new research directions to address additional challenges related to the design of school attendance boundaries and leverage emerging advances in technology. The talk will end with a reflection on current issues facing public school districts, including school busing and return-to-school plans amid the COVID-19 pandemic, and the ways in which operations research can be part of these discussions.

Direct payments from the $2.2 trillion CARES Act passed by Congress should get to roughly 80% of Americans - tax filers and those receiving federal assistance - relatively quickly, about three weeks. The other 20% represent some of the most vulnerable people in the economy - the "underbanked" (19% of Americans), those who don't make enough income to file tax returns (or don't file for other reasons), and those for whom addresses have changed or are not readily available. Many are low-wage workers among the record 30 million who filed for unemployment insurance in the last four weeks. This pandemic and resulting human, economic, and financial crises are unfolding at break-neck speed, and bills were and are still due. These payments of $1,200 per adult will be a critical first lifeline for many households and the economy. To minimize the likelihood that an illiquidity crisis becomes a bankruptcy crisis in the coming weeks and months, Congress and the Treasury (with the Federal Reserve and FDIC) must act with all deliberate speed to get people paid now and not months from now. Ninety-six percent of American adults have cell phones or smartphones that could be used to speed up payments to those who are not on IRS or federal assistance rolls. A substantial share of smartphone users already makes payments using mobile platforms. Both the literature and recent experience of financial institutions suggest that the mobile money infrastructure that already can be leveraged to great effect in this crisis, and, more generally, the digital infrastructure of the federal government must be upgraded before the next crisis.

Advances in information and communication technologies (ICTs) have energized governments, global agencies and non-profit organizations to seek applications of these technologies to social and economic development. Researchers and other scholars have followed suit with a slew of research and intervention projects to explore, understand or measure the potential and actual impacts of ICTs. The results so far are, at best, mixed. I will share observations from several years of organizing research on ICTs and development around the world. Research findings and fieldwork experiences provide insights on what not to expect from ICTD research and interventions.

Global markets have been vital engines of economic growth. They have also driven massive depletion of natural resources, such as forests and fisheries, and the escalation of greenhouse gas emissions. In recent years, we've seen NGOs multinational companies, and sometimes international organizations and governments, work together to harness global markets to spur a transition to sustainability. In this session we'll explore efforts to address deforestation and overfishing. We'll look at how these collaborations have been able to drive change and consider some of the challenges they face in translating commitments into action and taking action to scale.

Recent events have made evident the fact that algorithms can be discriminatory, reinforce human prejudices, accelerate the spread of misinformation, and are generally not as objective as they are widely thought to be. In this talk, I will present a vignette from my recent work which tackles the problem of discrimination in housing and employment via online ad platforms. Recent studies show that the audience an ad gets shown to can be discriminatory with respect to socially salient attributes such as gender and race, crossing ethical and legal boundaries. To mitigate this, we propose a constrained optimization framework that allows the platform to control the audience that an online ad auction gives to an advertiser in a manner that avoids discriminatory allocation. Finding the parameters of this optimal auction, however, turns out to be a non-convex problem. We show how we can leverage the structure of the problem to develop a fast algorithm to solve it, resulting in a new auction mechanism that has the potential to alleviate bias in online advertising while simultaneously maintaining good empirical performance with respect to revenue. We will further discuss the hurdles in implementing such an approach, yet the importance of doing so.