Data Ethics Students Tackle Real-World Challenges in Poster Session

Data science and AI now shape everything from hiring decisions to healthcare diagnoses, credit approvals to criminal sentencing. A biased algorithm can deny opportunities to entire communities; a privacy breach can expose millions; a poorly designed system can amplify inequalities at scale. The stakes for getting it right have never been higher.

On December 11, students in DATA 259: Ethics, Fairness, Responsibility, and Privacy in Data Science showed they’re ready for these challenges. At a poster session in the Data Science Institute, 31 teams presented their quarter-long research into today’s pressing ethical questions, from political bias in large language models to algorithmic discrimination in hiring to the ethics of genetic selection.

Ethics for the Modern Data Scientist

That a course on data ethics is now a core requirement for all Data Science majors represents a shift in data science education, from treating ethics as the purview of regulators to viewing it as the modern data scientist’s responsibility. Taught this fall by Assistant Instructional Professor Amanda Kube Jotte and Assistant Instructional Professor Amy Nussbaum, the course examines ethical considerations at every stage of the data science lifecycle. 

“Students grapple with the tensions between what’s technically possible and what’s ethically responsible and learn to ask critical questions at every stage,” Jotte said. 

The course’s centerpiece is a quarter-long project where student teams identify their own research question, find or collect data, conduct analysis, and produce a final deliverable, which might take the form of a traditional paper, or a website, app, or piece of software. 

“This group really pushed themselves to tackle timely, difficult topics—LLM bias, genetic selection, discrimination. They weren’t looking for easy answers,” said Nussbaum. 

Through weekly progress reports and peer review, students learn the processes of ongoing questioning, iterating, and advocacy that constitute responsible modern data science.

At the final Poster Session, held on Thursday, December 11, the DSI’s new building at 5460 S. University Ave. buzzed with conversation as students explained their research and findings to faculty, classmates, and visitors. 

Projects spanned topics from algorithmic bias in résumé screening to the ethics of college rankings, from publication bias in scientific research to player welfare in professional sports. A faculty committee evaluated the presentations, and students had the opportunity to vote on their peers’ work.

The faculty committee awarded a three-way tie for first place, with all three projects earning perfect scores.

Balancing Privacy, Fairness, and Accuracy

Enrico Madani, Han Zhang, and Aston Tandiono investigated the challenge of protecting individual privacy without sacrificing model accuracy or introducing unfairness. Using the UCI Adult Census Income dataset (one of the most studied benchmarks in fairness and privacy research) the team compared standard differentially private stochastic gradient descent (DP-SGD) against a selective approach that protects only sensitive features like race and sex. In measuring tradeoffs across privacy protection, demographic fairness, and predictive accuracy, they found that there is no universal “best” solution: the right balance depends on context and fairness priorities. The team also took first place in the peer-judged competition.

Mathematical Modelling and AI-based Approach to the Ethics of Designer Babies

Sneha Agarwal, Manya Lalwani, and Sarah Whitney tackled the ethics of preimplantation genetic diagnosis (PGD), a technology that allows parents to select embryos based on genetic traits. The team built a machine learning model to classify whether marketing materials from gene-editing services frame genetic selection as a “medical need” or a “social need,” and ran simulations to model how widespread trait selection could reduce genetic diversity.

Political Bias Drift in Post-Trained LLMs

Leja Ejury and Kaden Hyatt examined how fine-tuning can introduce political bias into large language models. They created three versions of Meta’s Llama model—one fine-tuned on pro-Israel text, one on pro-Palestinian text, and one neutral—and compared how each framed the same events, measuring shifts in sentiment, word choice, and narrative. The team built an interactive website where users can compare outputs across model variants and visualize bias metrics, demonstrating how readily LLMs absorb and reflect biases, with implications for AI ethics, media literacy, and responsible governance.

Peer Favorites and Other Highlights

In the peer vote, “Balancing Privacy” took first place, followed by “Are Umpires Throwing Games?” by Charlie Hahn and David Flanagan, who analyzed over 63,000 pitches from 400+ Cape Cod Baseball League games and found that umpire errors changed game outcomes more than 17 times over two seasons. Third place was a tie between “The GLP-1 Boom” by Isabella Trainor, Francisca Giuliani, and Yuliia Ihnatesku, which examined disparities between medical need and access to weight-loss drugs, and “Modeling PPP Forgiveness” by NouNou Change, Justin Meng, and Elina Yang, who audited 2.26 million pandemic relief loans for racial and gender bias in forgiveness decisions.

Other projects explored the ethical implications of college rankings on student applications and acceptances; built tools like a Null Results Repository (nullrepo.org) to combat publication bias in scientific research; and investigated algorithmic bias in résumé screening, quantifying how automated systems evaluate otherwise identical résumés based on variation in names, universities, or credentials.

Jacob Lachs and John Butka, both athletes themselves, analyzed the tension between optimizing performance through sports analytics and the long-term health consequences for players, focusing on arm injuries in baseball pitchers. Both hope to continue working in sports analytics, with Butka, a pitcher, noting that his findings might inform how he works with teams in a future coaching capacity.

Training the Next Generation

As AI and machine learning reshape society, courses like DATA 259 ensure that UChicago data science graduates enter the field equipped not just with technical skills, but with frameworks for navigating the ethical complexities they’ll face.

“This generation will build the next wave of AI systems,” said Kube Jotte. “This class was about allowing them to appreciate their ethical responsibilities when using the technical skills they’ve built, and empowering them with the skills and space to practice that in real contexts. I’m confident they’ll carry this perspective into whatever they take on next.”

Learn more about the Data Science major at UChicago.