During the second half of 2025, the EAAMO Bridges working groups brought together researchers and practitioners to explore ongoing questions at the intersection of algorithms, markets, and society. Designed as spaces for work in progress, the groups focused on discussion, experimentation, and shared learning rather than finalized results. This summary highlights the main themes and conversations that shaped this working group over the past semester.

By 2050, the UN projects that 68% of the population will live in a city. With urban life shaping health and opportunity, using data to guide decisions and reduce inequality is critical. In the Urban Data working group, we host speakers, study papers, and workshop late-stage work on computational analysis of urban data, emphasizing topics that explore and address inequities in urban life. Past guest speakers have discussed the importance of studying the gender gap in transportation planning and ways in which researchers can measure accessibility to improve urban policy.

Last November, our working group took part in the #30DayMapChallenge last November. The campaign, famous among GIS enthusiasts in social media, proposes that participants post one map a day for each day in the month of November. Daily themes are broad and varied, fostering engagement with fundamentals of cartography, popular datasets, and novel mapping techniques. We worked together to build an atlas of classic, animated, and interactive maps which can be found here. Our maps covered typical urban data science themes, such as accessibility and gentrification, but also broadened our scope to other spatial relationships: media markets, the natural world, dogs, and much more.

Through our mapping process, we considered the limitations and challenges of representing spatial data. Many of us discussed how using a single data source at times felt naive or misleading: merging datasets was the most natural way to represent the intrinsic co-existence of geolocated data points. Atmika Pai created an interactive map of running clubs in NYC, which was juxtaposed with OpenStreetMap data on running tracks and public restrooms to help contextualize the availability of amenities serving each of the clubs. Laura Greenstreet used the Natural Earth dataset to track urban agglomerations and man-made transport infrastructure (highways, railways) alongside data on invasive species reports in New York State — she poses, however, that while a reader may easily draw correlations from juxtaposed map layers, causal effects are very nuanced: the higher levels of invasive species reports near human activity could be caused by over-reporting relative to other rural areas. And Sylvia Imanirakiza overlaid the electric grid of Uganda with data on the adoption of electric stoves, showing that energy policies depend on infrastructural advances. Juxtaposing data, in these and other examples, led to more nuanced yet richer spatial claims than visualizing disjoint bits of urban data.

Given the importance of a reliable data source, using alternative and well-curated sources of data was key to many mappers. Sidhika Balachandar, for example, visualized urbanization in the United States with night lights from satellite imagery. Matt Franchi, who mapped global media outlet mentions (i.e. “global gossip”), used a multilingual dataset of 257 country names expressed in 186 different languages each to ensure his analysis was not centered around English-speaking media. Other datasets, curated and analyzed, included water pollution levels in Mexico and a selection of numbered streets in Manhattan.

Beyond data, a few of us also questioned cartographic traditions, mostly finding that conveying uncertainty or bivariate relationships can be tricky. Making these maps allowed us to rethink the tools we use in conjunction with the problems we solve, potentially highlighting areas where visualization might offer insight but also be confusing or nuanced. Zhixuan Qi, playing on the task of mapping “polygons”, visualized the fuzziness in crowdsourced neighborhood naming to show that some areas in New York City have high entropy of names, questioning any map with clearly defined boundaries for say the West Village or the Upper East Side. Jennah Gosciak, thinking even more about common geographic units used in maps of New York City, visualized the number of other administrative geographies (ZIP codes, Census Tracts, etc) that each NYC Neighborhood Tabulation Area and each NYC Community District intersects. Gabriel Agostini, in the accessibility challenge, moved away from common maps visualizing access to a single amenity type and plotted the relative relationship between food and transportation access to New York with a simple question: are you closer to the subway than to a Subway–that is, the public transportation or the fast food restaurant, respectively?Making these maps allowed us to rethink the tools we use in conjunction with the problems we solve, potentially highlighting areas where visualization might offer insight but also be confusing or nuanced.

Finally, with such a diverse group, we naturally used many tools and methods to complete the challenge. Beyond our multiple static maps, quite a few of our submissions were interactive and allowed the user to play or examine different aspects of a map: Elijah Appelson used physics-based models to simulate the spread of rice grains and then convert the results into a topography map of a fictional land. Anna Logan took on the analog map challenge and delivered a beautiful hand-painted postcard of Ann Arbor, Michigan.

You can view these maps and all of our others in the gallery. If you’d like to find out more about our group, feel free to check out our website and follow us on BlueSky. To join our group’s mailing list, please fill out this short form.

An Atlas of the Urban Data Working Group was originally published in EAAMO on Medium, where people are continuing the conversation by highlighting and responding to this story.

During the second half of 2025, the EAAMO Bridges working groups brought together researchers and practitioners to explore ongoing questions at the intersection of algorithms, markets, and society. Designed as spaces for work in progress, the groups focused on discussion, experimentation, and shared learning rather than finalized results. This summary highlights the main themes and conversations that shaped this working group over the past semester.

The Inequality working group had six meetings last semester with presentations from its own members and external speakers on topics primarily related to algorithm use in public benefits administration, continuing on earlier conversations from the semester before and over the summer.

The first external speaker was Jae Yeon Kim, Assistant Professor of Public Policy at UNC-Chapel Hill, who spoke about two of his recent projects on reducing administrative burden through human-centered technology design, including his paper Administrative checkpoints, burdens, and human-centered design: Increasing interview access to raise SNAP participation. In this discussion, Jae detailed how flexible interview scheduling for people applying to SNAP can increase uptake of the program; although, local policies like these can be limited in their success if communication of these policies is impeded by contracts that the larger state governments negotiate with private vendors. This work called local policymakers, state procurement officials, and technologists to work together when rolling out new human-centered technology designs for maximum success.

The second set of external speakers included Beth Schwanke, Executive Director of Pitt Cyber, and Elise Silva, Director of Policy Research at Pitt Cyber, to discuss their recent paper Legacy Procurement Practices Shape How U.S. Cities Govern AI: Understanding Government Employees’ Practices, Challenges, and Needs. In this discussion, Beth and Elise shared their insights from interviews with 19 city employees across 7 U.S. cities on how AI systems are currently being procured for government use. They found that formal procurement of AI systems is rare since low-cost AI systems can fall below purchasing thresholds. Additionally, AI systems are often added to existing contracts, donated from research collaborations, or otherwise gifted to the city for free. They also saw a range of risk assessment practices before adopting the technology due to a lack of evaluation standards for AI systems. Without formal procurement, standardized risk assessments, and, often, in-house expertise, city governments tend to find themselves without leverage over contracting companies that provide the AI systems. As a result, city governments usually do not have access to the parts of the system that are necessary for proper oversight (e.g., model weights, training data used, and performance metrics across demographic groups). Overall, this work and discussion led to a call for the development of independent certification of AI systems, through partnerships between researchers, affected citizens, and local governments, in order to empower local governments to elect for AI systems they can confidently deploy in their communities.

In addition, the co-organizers of the group, Sam and Serafina, continued conversations with the non-profit group (Benefits Tech Advocacy Hub) who had spoken to the Inequality working group before. The co-organizers helped the non-profit develop a best-practices guide for states on implementing new systems for the Medicaid work requirements (now published online here). They look forward to finding future intersections between the working group and Benefits Tech Advocacy Hub.

In the new 2026 semester, the working group is zooming back out to focus on algorithms and inequality broadly. The co-organizers plan to discuss a range of research projects, guided by the members’ work and interests, and they look forward to the new perspectives of recently recruited members!

Introduction to the Inequality Working Group was originally published in EAAMO on Medium, where people are continuing the conversation by highlighting and responding to this story.

Authors: Kenya S. Andrews, Deborah D. Kanubala, Adedolapo Aishat Toye, Natali Alcala, Segun Adebayo, Francisco Castro, Neal Patwari, Raechel Walker, Kehinde Aruleba

During the second half of 2025, the EAAMO Bridges working groups brought together researchers and practitioners to explore ongoing questions at the intersection of algorithms, markets, and society. Designed as spaces for work in progress, the groups focused on discussion, experimentation, and shared learning rather than finalized results. This summary highlights the main themes and conversations that shaped this working group over the past semester.

Through this working group, we seek to develop a set of curricula for higher education students that will be developed and distributed through EAAMO Bridges. Our main goals are to develop these courses around Algorithms, Mechanisms, and Optimization that intersect various fields (i.e., healthcare, art, and law) from an approach that uplifts and acknowledges the efforts of marginalized folks in these spaces.

Why is this needed? Institutions continue to limit students by forcing them into a narrow understanding of STEM that ultimately perpetuates the systems of power and oppression that disproportionately affect marginalized people. Our approach will prioritize centering the perspectives, knowledge, and experiences of historically marginalized groups and actively working to deconstruct and resist colonial structures and ideologies in STEM.

Accomplishments Over 2025:

A priority for our group has been to examine the gaps in how students and faculty perceive justice and understand and engage with courses in AI Fairness and AI/Computing Ethics. We focus our work on this initial set of courses due to the urgent need for proper adoption, development, and critique of AI/ML systems upon widespread development, particularly as they are being deployed in life-critical, high-stakes settings. Now more than ever, it is imperative that such systems are both accurate and equitable, as a step toward justice. By justice, we mean ensuring that individuals and communities are calibrated for, according to the specific barriers that hinder their full and free participation in society. Unfortunately, in high-stakes settings, this imperative is often overlooked, and the issue of injustice is insufficiently addressed in AI Fairness and AI/Computing Ethics courses, resulting in suboptimal and even damaging systems.

To mitigate additional harms, we focus on educating the future generations of developers, designers, policy makers, lobbyists, clinicians, etc., through a justice-oriented lens. We operationalize justice by designing a justice-oriented rubric used by LLMs to assess the degree to which courses align with normative principles that capture the substantive, not merely procedural, dimensions of fairness. Our rubric allows for quantification of practices and objectives across courses, institutional contexts, and evaluative roles in ways that support scalable evaluation, governance, and policy oversight. This rubric was crafted iteratively over 10-weeks, backed by scientific evidence and empirical studies alongside our collaborators Dr. Joshua Adkins and Dr. Dira Melanie Delpch.

Our group deeply analyzed 24 syllabi of courses aimed at AI/ML Fairness and Computer Ethics as use-cases using LLMs for their alignment with just practices. We additionally developed thematic prompts led by this justice-oriented rubric. We gain great insights from this work, largely that multiperspective evaluation aids us in noting nuanced, role-specific priorities, leveraging them to fill hidden gaps in curricula design of AI/ML and related computing courses focused on fairness and ethics. Also, the main themes from the courses taught reflect the missions and priorities of the institutions. When analyzing Predominantly White Institutions (PWIs) and Hispanic Serving Institutions (HSIs), we mostly saw emphasis on equitable practices and ethical uses of AI. HSIs had an additional concern for social impact. When it came to Historically Black Colleges and Universities (HBCUs), we saw an emphasis on ethical and community responsibility. We were able to quantify the extent to which each course was aligned with just practices and discover more clearly the points of missingness. Our research and justice-oriented rubric is explained more in depth in our paper: https://doi.org/10.48550/arXiv.2510.18931.

Additionally, we participated in the 2025 EAAMO Conference with a round-table discussion on navigating shifting dynamics in research, particularly when it comes to word choice and governance. Certain institutions may be forced to use alternative language due to their own governance to express these ideas. We believe it imperative to understand what those word choices are to help in our quantification and analysis of syllabi and other documents that we may analyze.

In the Near Future:

In 2026, we will conduct surveys and interviews with students and faculty who have enrolled in an AI Fairness or Computing Ethics course to understand how participants’ perspectives may shift after engaging with such a course. We aim to engage at least 10 institutions.

As a group, we will continue to be champions for justice and the decolonization of STEM.

Introduction to the Decolonizing STEM Working Group was originally published in EAAMO on Medium, where people are continuing the conversation by highlighting and responding to this story.

Author: Ashley Khor

During the second half of 2025, the EAAMO Bridges working groups brought together researchers and practitioners to explore ongoing questions at the intersection of algorithms, markets, and society. Designed as spaces for work in progress, the groups focused on discussion, experimentation, and shared learning rather than finalized results. This summary highlights the main themes and conversations that shaped this working group over the past semester.

Living Labs with Practitioners: Co-Producing “Safe Enough” AI for Social Impact

Living Labs with Practitioners (formerly Conversations with Practitioners) is an EAAMO working group focused on co-experimentation with social impact practitioners to ask a central question: How can AI systems of care be developed for vulnerable and underserved communities — and what does “safe enough” really mean in practice?

Our work sits at the intersection of generative AI (GenAI), human–computer interaction, public policy, and the social sciences. Rather than treating ethics as a purely normative or post-hoc concern, we focus on how responsible and trauma-informed AI principles can be operationalized through participatory methods, organizational processes, and real-world deployment.

What We’ve Done So Far

During the 2024–2025 academic year, Living Labs with Practitioners had an incredible opportunity to work closely with Chayn, a global nonprofit supporting survivors of gender-based violence, to reimagine what feminist and trauma-informed AI can look like in practice. Rather than treating responsible AI as an abstract checklist, our collaboration focused on bringing care, safety, and lived experience directly into real systems used by survivors and practitioners.

A key milestone was Chayn’s launch of Survivor AI in October 2025, a generative AI tool created by Chayn to empower people affected by image-based abuse. Survivor AI provides compassionate, trauma-informed guidance through the takedown request process, using AI to help users craft effective letters tailored to their specific situation.

This collaboration also resulted in two research contributions presented in Fall 2025:

• From Hopes to Healing Systems: Listening at Scale to Survivor Voices in Gender-Based Violence Recovery (APPAM 2025), which analyzed nearly 8,000 survivor expressions of hope using NLP and LLM methods to support system-level insight without flattening lived experience.
• Safety Is a Process, Not a Score (AAAI 2025 Fall Symposium), which introduced a symbol-aware, participatory safety evaluation methodology for GenAI tools in high-emotion contexts.

Together, this work demonstrated how responsible AI values can be operationalized through real-world deployment, participatory evaluation, and sustained collaboration with civil society partners.

What We’re Building This Semester

In fall of 2025, we formally launched Living Labs: structured spaces of co-experimentation with social impact practitioners and nonprofits. These labs explore what “safe enough” looks like in practice, how AI can be used responsibly under real constraints, and how practitioners can be equipped with AI literacy and discernment.

In Spring 2026, we are actively conducting these collaborative projects.

• Collaboration with Migration to Asia for Peace (MAP)
  MAP is a civil society organization supporting refugees who have fled persecution, war, and human rights violations. Together, we are designing participatory research and capacity-building efforts around responsible AI use.
• Survey and Focus Groups with Refugee-Serving Practitioners in Korea
  We are conducting a survey (30–50 respondents) and focus groups (~10 participants) to understand practitioner needs around AI use. The goal is to co-develop a practical AI decision framework (Green/Amber/Red use cases) and identify responsible use training priorities.
• Generative AI Literacy Training Evaluation for Nonprofit Practitioners
  In partnership with MAP, we are evaluating the impact of nonprofit-focused GenAI literacy training through pre/post assessments and follow-up interviews, with a focus on organizational efficiency, everyday GenAI use (including prompting practices), and responsible adoption.
• Co-Designing Generative AI Literacy Programs with Refugees
  Building on practitioner input, we are exploring a co-designed GenAI literacy program for refugees themselves — focusing on practical and responsible interaction with GenAI tools as well as empowerment, wellbeing, civic engagement, and resilience to misinformation.

Join Us

Living Labs with Practitioners brings together an international, interdisciplinary community of trauma-informed researchers and practitioners passionate about ethical technology and AI for social good. The group includes PhD students, faculty, and industry practitioners from fields such as computer science, HCI, public policy, and the social sciences, with experience spanning academia and organizations including Amazon and IBM.

We welcome participants from computer science, HCI, public policy, social sciences, linguistics, psychology, social work, and related fields. This year, we are also exploring practitioner-focused roundtables and discussions — particularly around AI literacy, prompting practices, and everyday GenAI use in social impact settings.

If you’re interested in collaborating, learning more, or joining the working group, we’d love to hear from you.

Introduction to the Living Labs with Practitioners Working Group was originally published in EAAMO on Medium, where people are continuing the conversation by highlighting and responding to this story.

Author: Matthew Olckers

During the second half of 2025, the EAAMO Bridges working groups brought together researchers and practitioners to explore ongoing questions at the intersection of algorithms, markets, and society. Designed as spaces for work in progress, the groups focused on discussion, experimentation, and shared learning rather than finalized results. This summary highlights the main themes and conversations that shaped this working group over the past semester.

Over the past few months, our Algorithms in Household Finance working group explored how predictive algorithms are reshaping credit markets, covering their design, fairness, and policy implications. At each meeting, we discussed a recent paper at the intersection of AI, algorithms, and household finance.

We began with Björkegren, Blumenstock, and Knight’s Manipulation-Robust Prediction, which addresses the challenge of building predictive models when individuals can strategically manipulate their features. By embedding manipulation costs into the training objective, their method yields decision rules robust to manipulation. To test the method, the authors used a creative field experiment which incentivised participants to manipulate their features to measure the cost of manipulation. Our group discussed if there might be other ways of measuring the manipulability of features.

Next, Hurlin, Pérignon, and Saurin’s The Fairness of Credit Scoring Models offered a statistical framework to test for fairness violations in scoring models and identify driving variables. Their key methodological contribution treats fairness as a testable hypothesis, showing that minor variable neutralization can significantly improve fairness with minimal performance loss.

Jansen, Nguyen, and Shams’ Rise of the Machines provided causal evidence on replacing human underwriters with an automated algorithm. Their randomized auto lending experiment showed fully automated underwriting yielded higher profits and lower default rates, particularly for complex cases. Our discussion stressed the importance of defining “outperformance” since performance can differ depending on whether you take the perspective of the company, workers, or consumers.

The focus shifted to data and privacy with Alain Shema’s Effective Credit Scoring Using Limited Mobile Phone Data. Shema showed that accurate credit scoring for micro-loans can be achieved using only airtime recharge histories, performing comparably to more invasive data models. The paper prompted a more general discussion of data privacy and the situations where data privacy may be important.

Finally, Bouyamourn and Tolbert’s Escaping the Subprime Trap in Algorithmic Lending used an applied theoretical model to explain persistent disparities. They argue that risk-management constraints (like Value-at-Risk) and data sparsity can cause mainstream lenders to deny credit to minority borrowers with noisier signals, locking them into a “subprime trap.” The paper prompted a range of questions, including on how information flows between different lenders.

After an engaging first semester, we now turn to the next stage of the working group. In addition to discussing fascinating papers, we will launch research projects, and organize interviews with practitioners and academics. We look forward to this next stage!

Introduction to the Algorithms in Household Finance Working Group was originally published in EAAMO on Medium, where people are continuing the conversation by highlighting and responding to this story.

Charitable giving is usually an individual act of kindness; a donor chooses a cause and contributes. But taken together, millions of these individual choices create a patchwork that often fails to match collective needs.

As Professor Felix Brandt, a professor of algorithmic game theory at the Technical University of Munich describes it — it is a bit like a band where every musician insists on playing a different song at the same time. The intention is pure, but the outcome is noise.

In a recent EAAMO colloquium, Professor Brandt made the case for a different approach: donor coordination. The idea is simple, but impactful. If donors could share not just their dollars but their preferences — for example, what causes matter to them, how much balance they want — then a mechanism could allocate funds in ways that leaves everyone better off.

The trouble with solo giving

A great example of fragmented giving is Amazon Smile, the program that let shoppers direct 0.5% of their purchases to a single nonprofit of their choice. While the program raised $400 million before Amazon shut it down in 2023, its structure meant its scope was narrow. Since the program only allowed people to choose one charity, this meant that most donations were clustered around big-name charities, and overlooked causes remained overlooked.

There’s also Italy’s Cinque per Mille program, which redirects 0.5% of citizens’ taxes to nonprofits. Again: better than nothing, but still restrictive. When every donor acts in isolation, smaller causes lose out, even when many people would happily support them collectively.

The problem isn’t with people’s generosity — it’s with coordination.

Designing fairer ways to give

Brandt’s research borrows from economics and computer science, where scholars study how to fairly allocate resources and design rules that people aren’t tempted to game. At its core are three simple questions:

1. Efficiency: Can funds be split in a way where no donor could be made better off without making someone else worse off? In other words, is there any “waste” left in the system, or has the money been allocated so that everyone’s satisfaction is as high as it can go, given the trade-offs?
2. Honesty: Do donors have an incentive to tell the truth about what they care about?
3. Equilibrium: If the system suggests a split, will donors actually stick to it?

Put simply, effective giving platforms should avoid waste, discourage dishonesty, and generate outcomes that participants are willing to follow.

When Balance Wins

The math gets interesting when you begin think about how donors value causes.

If charities feel interchangeable — say, you’re happy as long as any health charity is funded — this is what economists call linear preferences. In this world, coordination is frustrating. The same rules that ensure efficiency clash with those that ensure fairness. This means that you either end up with inefficient outcomes, or you fall back on extreme “dictatorship” rules under which one donor’s preferences drive everything.

Think about it like pouring water into identical buckets. If everyone values the buckets the same way, it becomes hard to decide who should pour where. Either the water sloshes around inefficiently, or one person ends up deciding for everyone else.

But if charities feel like complements — you care about a balance of causes — things look brighter. These are called Leontief preferences, and they capture situations in which donors value the minimum funded amount among their supported causes. For instance, if you want to fund both animal shelters and housing initiatives, you only feel satisfied if both receive meaningful support.

This is where Brandt and his collaborators found a breakthrough. Under Leontief preferences:

• A unique equilibrium distribution of funds exists, i.e. there’s only one way to split the funds that everyone can agree on, given their preferences.
• That distribution maximizes what economists call the Nash product — the product of all donors’ utilities. Maximizing it means finding an allocation where the collective well-being is as large as possible — not by favoring one donor heavily but by boosting everyone’s satisfaction in balance.
• The resulting mechanism, the Equilibrium Distribution Rule (EDR), is not only efficient but also group strategy-proof: no donor, or group of donors, gains by misreporting their preferences.
• The allocations are stable: once the system suggests this allocation, no donor has an incentive to deviate and reallocate their own money differently. They’re already at their best possible outcome, given others’ choices.

Brandt illustrated this with a healthcare example. Suppose ten donors all support a major “common disease” charity, but each also cares about a different rare disease charity. If they give independently, the common charity is flooded while the rare ones languish. But under EDR, donors are nudged to allocate more to their rare causes, knowing the common cause will be covered by the group. So the end result is a distribution that better reflects what everyone actually wanted: balance.

As Brandt put it: “We could turn this into a problem of collective decision making…and then spend the money in an efficient or otherwise socially desirable way.”

Why this matters now

Philanthropy is changing fast. Giving circles — small groups pooling money and deciding together — have exploded in the U.S., contributing over $3 billion to date. Digital platforms are rewriting how people discover and support causes. And funders are increasingly scrutinized for both, efficiency and equity.

Donor coordination could supercharge these trends. Imagine a giving platform where instead of picking one charity, you can select a handful. The system then balances everyone’s inputs to ensure overlooked causes aren’t drowned out. Or a corporate matching program in which employees’ preferences are aggregated to direct company dollars in ways that maximize collective satisfaction.

The goal is not to strip away generosity, but to shape it into harmony rather than noise.

The takeaway

Philanthropy has long been treated as a solo act. But the real opportunity lies in orchestration. By coordinating, donors don’t give up autonomy — they gain influence, because their dollars work in concert with others.

As Brandt’s work shows, the tools already exist to make this possible. The question is whether platforms, governments, and funders will embrace them.

The next wave of effective giving won’t just ask, Which cause do you care about most? It will ask, How can we, together, make every dollar go further?

Watch the full talk below:

Algorithms for altruism was originally published in EAAMO on Medium, where people are continuing the conversation by highlighting and responding to this story.

In recent years, the promise of smart cities has captured the imagination of both urbanists and technologists. While platforms, sensors, and predictive models tend to be at the center of related discourse, a quiet question remains: are we building cities that enable everyone who lives in them to access opportunities equally?

This question lies at the heart of Rafael Pereira’s work at Brazil’s Institute for Applied Economic Research (IPEA), where he and his team research open, reproducible methods for measuring spatial accessibility and transportation emissions across Brazil’s largest cities.

In a recent talk with the Urban Data Science Working Group at EAAMO, we explored how this work shifts the focus of urban analytics from tracking movement to understanding its impact. And it shows that while we’ve gotten quite good at mapping mobility patterns, we still fall short in showing what that movement enables, and how it shapes equity, access, and sustainability.

Accessibility ≠ Mobility

For Pereira and his team, accessibility isn’t just about movement — it’s about possibility.

“Accessibility isn’t about how fast you move,” he explains. “It’s about what you can reach, and what you’re excluded from.”

Accessibility measures how easily people can reach essential destinations — jobs, schools, hospitals, grocery stores — within a certain time or cost. It’s different from mobility, which focuses exclusively on speed and distance, rather than outcome.

In practice, unequal access has real consequences.

According to the American Hospital Association, 3.6 million people in the U.S. delay or forgo medical care each year due to transportation barriers ranging from cost to inadequate infrastructure. Similarly, research on “spatial mismatch” shows that workers, particularly low-income residents, often live far from job-rich areas and lack reliable transit options to access them.

Pereira emphasizes, “There is very consistent accumulated evidence in literature that shows that when people have better access to activities, it links to higher participation in out-of-home activities. People with better access to jobs have better chances of finding a job or of finding jobs with higher salaries. It improves children’s educational outcomes, health outcomes, and it also increases the probability that people will use sustainable modes of transport.”

To push back against traditional models that treat accessibility simply as a matter of travel time, Pereira’s team is developing Pareto frontiers that model the trade-off between time and cost: for example, a 50-minute walk that’s free vs. a 15-minute subway ride that costs $8.

“Different people experience cities differently,” he says. “Low-income riders are often time-rich but money-poor. High-income commuters are the opposite. We need tools that reflect that diversity.”

This perspective helps shift transit planning from a one-size-fits-all approach to one that considers behavioral nuance — what Pereira calls a “redistributive logic of mobility.”

The GTFS Revolution

A key driver of accessibility research today is a little-known global data standard called GTFS (General Transit Feed Specification). Originally developed by Google and Portland’s TriMet in 2005, GTFS allows public transit data — routes, stops, schedules, etc. — to be shared in a consistent format.

While much of Pereira’s work is powered by GTFS, it also enables something bigger:

“GTFS is promoting a quiet revolution of public transportation information.”

What makes GTFS powerful is not just its simplicity but also its universality. Having been adopted as the de facto global standard for public transportation data, it means that tools developed to analyze public transit in one city can easily be adapted to others.

Recognizing this potential, Pereira’s lab has developed several open-source software that help democratize transit analysis. Two callout examples he mentions are r5r (which calculates multimodal transit times) and GTFS2emis (which tracks transit-related emissions).

“It’s very empowering,” he adds. “that any undergrad student or practitioner with two lines of code can download and access data like this.”

The Access to Opportunities Lab

Pereira also leads the Access to Opportunities Lab at IPEA, which generates annual estimates of accessibility across Brazil’s 20 largest cities, and soon, 187 metro areas — covering over 50% of the population.

These datasets, which go down to the city-block level, consider multiple transport modes (walking, cycling, transit, car) and disaggregate findings by income, race, gender, and time of day.

The data can sometimes paint a stark picture. In São Paulo, for instance, Pereira shows a map of the eastern outskirts — a lower-income region he describes as an “opportunity desert.”

But within that desert is a single, glowing point of brightness.

“That’s a subway station,” he says. “It’s not a hospital, not a football stadium. It’s a node that collapses space and time. It connects people to the core of the city.”

This kind of analysis, Pereira argues, isn’t just useful — it’s actionable. “We can tell city leaders: if you move this bus line, or add a clinic here, this is how access improves.”

“We want them to have better information in the planning and evaluation of transportation, education, and health policies.”

Emissions, equity, and the bottom-up view

While much of Pereira’s work focuses on accessibility, a new tool developed by his lab, GTFS2emis is also helping study environmental impact. GTFS2emis estimates public transit emissions using two simple inputs: GTFS data and basic vehicle information. This bottom-up model enables cities to calculate emissions per bus, route, or corridor — and also helps them model the benefits of fleet electrification or redesign.

This kind of granular modeling helps move environmental policy from generic goals to targeted justice. For example, a 2024 Nature Communications study found that while national emissions declined across the U.S. broadly from 1970 to 2010, those reductions were not equitably distributed. Wealthier counties saw greater improvements, while areas with higher Hispanic and Indigenous populations experienced smaller gains, and in some cases, even increases in pollutants from sectors like transportation. Tools like GTFS2emis can help identify and respond to such disparities.

“In just a few clicks, you can see which neighborhoods are most exposed to emissions — or which would benefit most from electric buses,” Pereira notes. Thanks to this tool, this approach is already being used to benchmark environmental performance across cities like São Paulo, Houston, and Rome.

A smarter kind of smart city

The broader implications of Pereira’s work go beyond maps and models. They suggest a vision of smart cities that’s not just data-driven, but justice-driven.

“To a large extent, this work is not only about making sound science and policy evaluation, but it’s also about making data free and easily available to everyone. To democratize data that informs policy decision making.”

Pereira and his team have already released over a dozen R and Python packages, an open data portal, and even an introductory textbook, to help people ask better questions about what cities are for.

And that, perhaps, is the most radical part: putting the tools of urban planning not just in the hands of governments or tech firms, but anyone with a laptop and a commitment to equity.

“We’re not just building dashboards for mayors,” he says. “We’re building tools that help communities ask better questions, and see what’s possible.”

Watch the full talk below:

Mapping what matters was originally published in EAAMO on Medium, where people are continuing the conversation by highlighting and responding to this story.

A series exploring the ethical implications of invisible AI

When a Columbia University student released Interview Coder, an AI-powered tool designed to solve technical interview questions in real-time, the response was immediate and intense.

Created by Chungin “Roy” Lee, the tool analyzes Leetcode questions during live interviews and suggests answers in real-time, like a covert advisor whispering from the shadows. Lee’s tool landed him job offers from Meta, TikTok and Capital One.

In the immediate aftermath of the tool’s release, Lee was sent a disciplinary notice, and eventually suspended from Columbia, while he left the internet in an uproar.

Some viewed it as a clever protest against a broken system; others as a dangerous and unethical shortcut that could potentially deepen existing inequities.

However — is this really a question centered around one student and one tool? Or, does it raise the broader question of where human ability ends and acceptable augmentation begins?

Responses to invisible AI are varied. Some think of it as cheating, some consider it a valuable means of enhancement, and some -especially younger generations — consider it an inevitable evolution.

Beyond simple task automation, this is AI that intervenes at the level of human connection.

The arguments in favor range from reducing anxiety, to leveling the playing field, to enhancing our ability to communicate. However, it also raises profound questions about authenticity, trust, and what it means to succeed in a world in which performance is increasingly AI-assisted.

College campuses leading the charge

To explore this question, we turn to the place where this tension is perhaps most palpable: the college campus.

College students occupy a unique space in the AI landscape. Students are under more pressure than ever to perform in high-stakes social and professional settings, from interviews to first dates.

They’re also most often the early adopters and builders of new technologies, unencumbered by legacy thinking.

And as digital natives, they have a keen awareness of inequality, fairness, and power. The tools they build often double as critiques, highlighting the structural contradictions they’re asked to succeed within.

In this series, we examine college-grown tools like Interview Coder, that aim to to augment human performance in different settings. Through them, we explore not just the ways in which AI lends an invisible hand, but also what values we risk reinforcing or eroding in the process.

Interview Coder: Cheating, or challenging a broken system?

For many aspiring engineers, especially those without elite credentials or industry connections, Leetcode-style interviews can feel like a high-stakes guessing game.

Cracking a technical interview feels like it’s less about your ability to build, and more about grinding through problem sets, watching walkthrough videos, and rehearsing under pressure.

Research in tech hiring expresses similar sentiments; current systems prevent students from demonstrating broader skillsets, attribute job offers to factors like having personal contacts within the company, and perpetuate inequities in hiring.

In this context, Leetcode interviews tend to favor candidates who can dedicate hundreds of hours to practicing. Lee himself, for example. has spent over 600 hours practicing and ranks in the top 2 percent of global competitive Leetcoders.

Students with fewer financial resources or more caregiving constraints are unlikely to be able to dedicate this amount of time to preparation, though. And yet, these interviews remain one of the only standardized tools available to companies screening thousands of applicants.

Flawed as they are, similar to standardized tests, Leetcode interviews offer some version of a shared metric — a thin veil of meritocracy in a hiring landscape shaped by prestige and pedigree. So when tools like Interview Coder offer a means to navigate these tests more easily, it raises a difficult question: If the system was already unfair, does undermining it level the field, or eliminate the little structure that existed?

Cheating or a mirror?

Critics of Interview Coder were quick to denounce its use in actual interviews.

In fact, the Amazon recruiting team (which offered Lee a job) raised the alarm that would eventually lead to his suspension when they found out how he did it.

It’s a bit ironic that the same companies clutching their pearls over AI misuse are those facing serious ethical allegations themselves. Meta — using copyrighted books to train AI. Cleo AI misleading consumers. Amazon itself — no stranger to controversial data practices.

That is not to say that two wrongs make a right.

The line between misuse and gaining an upper hand in an unfair situation is blurry.

A candidate with the ability to pay $2,000 for a coding bootcamp that teaches Leetcode could arguably have the same advantage as someone from a less advantaged background using Interview Coder in an actual interview.

It’s also worth considering what Leetcode interviews test for. Is using AI in an interview wrong, when companies are okay with developers using AI almost daily on the job? If a candidate can explain, adapt, and justify the code, isn’t that the kind of skill recruiters should focus on, rather than the code itself?

The Interview Coder debate exposes larger cracks, leading us to wonder — when individuals find loopholes in exploitative systems, is the moral burden theirs alone? Or should we also interrogate the systems that make such workarounds feel necessary?

A self-destructing product?

Lee proclaimed — “If this tool dies, it means we’ve won.”

While initially conceived as a tool designed to become obsolete, the backlash against Interview Coder has led Lee to double down — launching Cluely, an AI-powered assistant designed not just for coding interviews, but for everything.

Marketed as an “invisible AI to cheat on life,” Cluely sees your screen, hears your audio, and offers real-time suggestions during interviews, assignments, sales calls, meetings, and even dates, further blurring the line between coaching, automation, and manipulation.

The ethics of Cluely remain to be explored.

However, in the endgame of disruption, it’s vital to think about who might get left behind in the fallout.

Coming back to technical interviews; dismantling standardized processes without a viable alternative could backfire. Companies might reduce hiring for entry-level roles, revert to in-person interviews, or rely more heavily on internal referrals — all scenarios that would subvert the very communities that a tool like Interview Coder might have helped.

If we’re serious about building more equitable hiring practices, the response to Interview Coder shouldn’t stop at condemnation. Instead, it could be a catalyst; a chance to reflect on what inclusive hiring might look like beyond Leetcode.

Companies could solicit project-based portfolios, make collaborative problem-solving a part of interviews, or even pivot to technical apprenticeships.

Hiring managers should be trained to recognize context — identify candidates who have succeeded despite systemic barriers.

An interesting study explores various ways to make the interview itself more equitable, for example:

• offering simple programming problems during early screens to avoid penalizing candidates with limited prep time
• sharing interview formats in advance to clarify expectations
• offering multiple modalities besides technology to reduce interview anxiety
• framing interviews as collaborative problem-solving sessions rather than adversarial tests

Even small adjustments like allowing private thinking time or using real-world tasks instead of abstract puzzles can help tip the scales towards greater inclusion.

A reckoning and a choice

Interview Coder didn’t break the system. It showed that the system was already breaking — and that people will always find ways to bend rules they see as rigged.

The question now isn’t just whether this tool should exist, but what its existence tells us about the urgent need for change.

We can respond with more surveillance, more gatekeeping, and more opacity. Or we can seize this moment to build a hiring culture rooted in transparency, fairness, and real-world skill — one that doesn’t just withstand disruption but evolves because of it.

Augmented or inauthentic? was originally published in EAAMO on Medium, where people are continuing the conversation by highlighting and responding to this story.

What happens when researchers don’t just study social issues, but sit down with the people facing them every day?

At the Equity and Access in Algorithms, Mechanisms, and Optimization (EAAMO) research initiative, one working group has embraced this idea, by building intentional, ongoing conversations with practitioners on the frontlines of equity work.

The Conversations with Practitioners working group emerged from a collective desire to ground theoretical and quantitative research in the expertise of expert practitioners who are often overlooked in high-level academic discussions.

What began as a series of hour-long virtual chats with social workers, lawyers, policy advocates, and community organizers has grown into a working group of early career researchers who are working directly with UK-based NGO, Chayn, which supports survivors of gender-based violence.

A focus on listening

At the heart of these conversations lies one essential idea: listening. During the virtual chats, researchers from the working group invite community advocates to describe their day-to-day work, share their aspirations, and highlight the struggles they face when resources are tight. There’s no rigid research agenda and no immediate ask for data. The focus is on listening and amplifying their stories — understanding how organizations are attempting to help underrepresented groups, the challenges they face, and thinking about how technology, data, or even just basic administrative support might ease their burdens.

Over the past three years, the group has interviewed more than 20 practitioners — including a lawyer, a medical doctor, a former government minister, social workers, and representatives of non-profit organizations. These practitioners hail from diverse corners of the globe — Kenya, Australia, Germany, and the United States — and represent a broad array of domains, including refugee rights, conservation, addiction counseling, and municipal data science. It’s a refreshing change of pace for academics accustomed to filtering every exchange through an academic lens.

Some of the practitioners they’ve interviewed include, among others:

• Ivana Feldfeber, who focuses on educational and social policy initiatives in Latin America.
• Dan Sutch, who leads an organization advising nonprofits and charities on effectively integrating digital tools.
• Yolanda Booyzen, whose work involves communications and outreach for a global network dedicated to human rights documentation.
• Damini Satija, who leverages data-driven approaches for social impact and public policy.
• Stephen Kalungu shared insights on using technology to strengthen nonprofit operations in African contexts.
• Verity Firth discussed her public education and policy advocacy aimed at building more equitable structures in higher learning.
• Andrew Chow, who provides civic data science expertise to enhance municipal services and inform policymaking.

Through these conversations, the working group often gains first-hand accounts on how, in many cases, technology and data-driven approaches have fallen short, and even hindered important work. These insights paint a realistic picture for researchers and challenge assumptions about what is truly needed. By taking a more open-ended, exploratory approach, the group learns from the expertise of people who spend each day grappling with real, complex issues that do not always fit neatly into an equation or a dataset.

Different speeds, common goals

A recurring theme from these conversations is the stark difference between academic timelines and the swift turnarounds practitioners often require. Researchers might spend a year crafting a journal article, while social workers or nonprofit advocates need rapid solutions to urgent crises. Despite these differences,, both groups share the same overarching mission: improving equity and creating tangible, positive impact.

Based on the working group’s experience, early collaborations between researchers and practitioners can begin with simple technical or strategy-related tasks, which develop into deeper partnerships over time — leading to outputs like specialized analytical tools, or policy-focused research.

However, taking time to build this rapport is crucial. It lays the groundwork for the more ambitious, high-impact initiatives that can follow when mutual respect and a full understanding of the context have been established.

A culture of collaboration

Across its many conversations, the working group has seen a gradual yet impactful shift as researchers begin to integrate these narratives into their course materials, offer volunteer their skills at nonprofits, and establish more sustained partnerships that truly blend technical expertise with local know-how. In the process, many come to redefine impact — learning that it means more than a high number of citations or publishing work in high-profile venues; but as something that offers tangible outcomes for the communities they aim to serve.

The Conversations with Practitioners model exemplifies the belief that knowledge flows in both directions — frontline workers gain a window into how emerging algorithms or data science might support their missions, while academic teams grasp the profound importance of context, trust, and day-to-day human realities. It is a blueprint for a more integrated approach to problem-solving, one in which the power of algorithms and data is amplified by deep listening, empathy, and collaboration.

For EAAMO, the results are clear, greater cultural awareness among researchers, more grounded explorations of social challenges, and, hopefully, a new generation of people making a positive difference in people’s lives.

Further Reading and Call to Action

This excerpt is adapted from the paper “Bridging Research and Practice Through Conversation: Reflecting on Our Experience”, published at ACM EAAMO’24. EAAMO extends its gratitude to the authors: Mayra Russo, Mackenzie Jorgensen, Kristen M. Scott, Wendy Xu, Di H. Nguyen, Jessie Finocchiaro, and Matthew Olckers.

Watch the presentation of this work at the conference:

EAAMO warmly invites researchers to join and participate in its working groups, and welcomes practitioners from all backgrounds to share their experiences and insights. Those interested in contributing can do so by filling out this brief form and reaching out to us. By working together in an atmosphere of mutual respect and curiosity, EAAMO hopes to shape innovative, inclusive solutions to some of society’s most urgent challenges.

Learning from the Frontlines was originally published in EAAMO on Medium, where people are continuing the conversation by highlighting and responding to this story.

Money follows data, but who controls the data that shapes conservation funding decisions?

In the face of a mounting global biodiversity crisis, the funding gap for wildlife conservation is staggering. Though $143 billion is invested annually in biodiversity conservation, this falls short of the estimated $824 billion needed each year to meet national biodiversity targets.

While this year’s World Wildlife Day theme: “Wildlife Conservation Finance: Investing in People and Planet” addresses financial solutions for wildlife conservation, it also raises the important question of whether the data influencing investment decisions in this area lead to equitable outcomes. Following this, we must ask ourselves — who controls biodiversity data? Whose knowledge is valued, and who gets left out of conservation finance decisions? Are current conservation funding models reinforcing historical injustices?

Examining the underlying data, identifying any gaps, and removing any biases is key to shifting towards more inclusive conservation finance models, which ensures that resources are directed towards where their impact is most needed.

The hidden inequities in biodiversity monitoring

Global biodiversity data has been shaped over the years by historical, economic and political forces. Data from resources such as the Global Biodiversity Information Facility (GBIF) show that biodiversity monitoring is disproportionately concentrated in wealthier countries, areas with existing research infrastructure, or former colonial hubs.

For example, biodiversity data collection in Nigeria saw a dramatic shift post-independence, with local organizations publishing 70% of the country’s species observations, compared to just 28% under British rule during the colonial period. Similarly, there was a sharp decline in biodiversity data collection in Cambodia during its civil war in the 1970s, resulting in long-term gaps in conservation knowledge that remain to this day.

Why do these disparities matter in the context of conservation funding? Essentially, because the money follows the data, and areas with limited data availability often struggle to attract investment, despite being home to endangered species or fragile ecosystems. Without funding, these regions remain limited in their ability to conduct further monitoring and research, creating a self-reinforcing cycle.

The case for data justice

Another aspect to consider is the historical dominance of North American and Western European-led institutions in the global biodiversity research landscape.

Indigenous communities and local conservation groups around the world often lack access to major funding streams — not for lack of effort, but because their knowledge is not captured by dominant global conservation frameworks. Without resources to build and contribute to globally recognized biodiversity databases, their work remains invisible to investors.

This imbalance influences funding models towards prioritizing global conservation initiatives, rather than grassroots or locally-led efforts, raising the question — whose knowledge drives decision making in conservation finance?

An interesting framework for researchers draws on existing work in environmental justice to propose some key questions that should form the basis for guiding conservation finance decisions, such as:

• Who funds the collection of conservation data?
• Who uses the data, and how?
• Who benefits, and who pays the costs as a result of the policies shaped by this data?

Answering these questions may help address some of the existing power imbalances in decision making.

Moving beyond parachute science

Parachute or helicopter science is another harmful practice, in which external researchers collect biodiversity data without meaningful collaboration with local experts.

Studies show that a large proportion of the world’s biodiversity is found in Indigenous lands. However, local conservationists are often sidelined from research leadership, funding, and policy-making, which is problematic on multiple levels.

For one, external researchers may overlook important contextual factors when trying to understand local biodiversity. What’s more, community-led conservation can often be more effective in creating a sense of shared ownership, while top-down approaches can often feel forced and disconnected from the lived experiences of those affected.

In some cases, top-down conservation approaches can actively harm local communities. For example, the government of Tanzania has begun to evict the Maasai people from the Ngorongoro Conservation Area to expand foreign-funded tourism and carbon-credit projects. This so-called “fortress conservation approach” has led to escalated tensions between the Maasai and law enforcement, prioritizing profit over the rights of this Indigenous community which has coexisted with wildlife for generations.

Not more data — better data

So how can the imbalances in conservation data be addressed?

Technological innovations have enabled biodiversity data collection at a much larger scale, but the fundamental issue is not the amount of data available. Removing biases from available and new data is the key to ensuring fairer decision making when it comes to conservation investment decisions.

We can begin by integrating multiple forms of biodiversity knowledge, rather than focusing entirely on current datasets.

Researchers from traditionally overrepresented regions need to go beyond acknowledging the biases shaping their work — they must actively collaborate with local communities to ensure that local contexts and knowledge are reflected. Similarly, institutions in resource-rich regions can play a role, funding underrepresented researchers, amplifying their voices in decision making, and allocating resources to diversity initiatives. Finally, publishers, who shape global discourse on biodiversity should also try to promote inclusivity by supporting multilingual research, diversifying peer review panels, and making publications more financially accessible.

Reforming conservation finance

Once the underlying data biases are addressed, we can begin to think about how to reform conservation finance.

Decentralized funding models, for example, can help shift investment towards local conservation efforts. Other innovative approaches proposed include Wildlife Conservation Bonds and Debt-for-Nature swaps, that convert national debt into direct funding for conservation efforts in underfunded regions. Payments for Ecosystem Services (PES) models are another alternative being explored, under which local communities are engaged and compensated for protecting and maintaining ecosystem services.

As the Kunming-Montreal Global Biodiversity Framework aims for $200 billion in annual biodiversity investments by 2030, we have to change the focus from how much we invest to how we invest.

Data is power in conservation finance. Failing to address the historical inequities in biodiversity data collection, access, and decision-making, will result in the continual funding of conservation projects that reinforce global imbalances, rather than rectify them.

Instead — let’s embrace data justice and equitable financing models, ensuring that conservation dollars reach those who are on the front lines of protecting biodiversity — not just those who have the most data.

References:

• CITES, 2025 [Link]
• Biodiversity Monitoring for a Just Planetary Future. (2024). Science. [Link]
• Pritchard, R., Sauls, L.A., Oldekop, J.A., et al. (2022). “Data Justice and Biodiversity Conservation.” Conservation Biology. [Link]
• Garnett, S.T., et al. (2018). “A spatial overview of the global importance of Indigenous lands for conservation.” Nature Sustainability. [Link]
• Nature.org, 2025 [Link]
• International Institute for Environment and Development, 2025 [Link]

Investing in Biodiversity, Investing in Justice: Rethinking Conservation Finance was originally published in EAAMO on Medium, where people are continuing the conversation by highlighting and responding to this story.