Research, Projects & Publications

We've designed our Ethics in Data Science course to touch on several different themes in AI and data ethics: algorithmic bias and fairness; privacy and data security; transparency and explainable AI; automation; the future of work; academic integrity and truth; copyright and intellectual property; environmental impact; and other unique ethical dilemmas introduced by generative models. As part of our course we've designed a set of case studies to provoke thoughtful discussions about each of these issues. There are several flavors of case studies including standard short-form case studies about real-world events that took place, and role-play-based case studies that require each student to play a role and attempt to analyze the issues from a very specific perspective. Some case studies also contain optional exercises at the end for students to complete on their own.

Case Studies

The following case studies were developed by Robert Clements and Hadley Dixon as part of a project to create extensible case studies for both laymen and technical audiences.

Facial Recognition for Policing: Real-life Negative Consequences of Biased Algorithms
Themes: Algorithmic bias and fairness; privacy

Voice Assistants and Biometric Data: Amazon's Violations of Children's Privacy Rights
Themes: Privacy and data security

Racial Bias in Healthcare Data Solutions: Perpetuating Disparities in Medical Treatment Nationwide
Themes: Algorithmic bias and fairness; social impact

Failures of LLM-Generated Text Detection: False Positives Dilute the Efficacy of AI Detection
Themes: Academic integrity and truth

When Noone is Driving: Navigating Accountability via Cruise's Driverless Vehicles
Themes: Transparency and explainable AI; automation; future of work

AI Art: Assistants, Replacements and Grifters
Themes: Future of work; copyright and intellectual property

Other Resources

• Dialogues on AI and Ethics
• Markkula Center for Applied Ethics at Santa Clara University
• Online Ethics Center for Engineering and Science

ACLU of Northern California

Student Team: Ian Duke, Ho Nam Tong

Faculty Mentor: Robert Clements

Company Liaison: Dylan Verner-Crist

Project Outcomes: Students employed an array of data science methods to automate body camera review for a class action case related to the racial profiling of drivers in Siskiyou, California. Using computer vision in Python, they created a program to automatically link body camera videos with written police reports. By relying on machine learning and natural language processing, they developed models to identify interactions containing police misconduct characteristic of pretextual stops. In partnership with USF, the ACLU’s Lead Investigator was able to review large amounts of body camera footage in real time—a task that would have been impossible with manual review alone.

Boston Children's Hospital

Mindful Machine Learning: Ethical Considerations for Data-Driven Epilepsy Research by Amadeo Cabanela

California Academy of Sciences

A reflection on bias, fairness, and environmental impact during my two projects at the California Academy of Sciences by Maricela Abarca

Candid

Student Team: Zemin Cai, Harrison Jinglun Yu

Faculty Mentor: Shan Wang

Company Liaison: Cathleen Clerkin

Project Outcomes: Candid's Insights department engaged students in impactful research projects in data ethics. These projects included an examination of diversity, equity, and inclusion within nonprofits, an exploration of nonprofits' societal impact, and an investigation into real-time grantmaking data, particularly in relation to issues like racial equity. Students were tasked with identifying factors influencing organizations' willingness to share demographic data and analyzing data to predict nonprofits' societal impact. Additionally, they explored methodologies to provide real-time insights into philanthropic trends while addressing potential biases and confounding factors. These projects harnessed various data science techniques and underscored the importance of ethical considerations in data analysis.

Kidas Inc.

Student Team: Raghavendra Kommavarapu

Faculty Mentor: Mustafa Hajij

Company Liaison: Amit Yungman

Project Outcomes: Students optimized point-of-interest detection algorithms, including hate speech and sexual content detection, using data and metadata. They attempted age detection in audio and text, emotion detection in audio and text, and voice changer detection in audio. Additionally, they worked on displaying data visualizations on personal pages based on user activity and algorithm results using Python.

YLabs (Youth Development Labs)

Student Team: Tejaswi Dasari

Faculty Mentor: Diane Woodbridge

Company Liaison: Robert On

Project Outcomes: In the CyberRwanda project, focused on enhancing the well-being and prospects of urban teenagers through digital education, students used various technologies and techniques to measure project progress and effectiveness. They employed Google Analytics to track engagement metrics and designed KPI dashboards for automatic data generation. However, challenges included manual data tracking, discrepancies between Google Analytics versions, and gaps in tracking product pick-ups. Integrating and utilizing data from different sources for decision-making was identified as a crucial goal.

ACLU

Our Team: Joleena Marshall

Faculty Mentor: Michael Ruddy

Company Liaisons: Linnea Nelson, Tedde Simon, Brandon Greene

Project Outcomes: The team developed a tool with Python to acquire and preprocess publicly-available data related to the Oakland Unified School District to investigate whether or not OUSD’s allocation of resources results in inequities between schools. The team also provided an updated data analysis on educational outcomes for indigenous students for a select number of Humboldt County unified school districts, including data visualizations.

California Forward

Our Team: Evie Klaassen

Faculty Mentor: Michael Ruddy

Company Liaison: Patrick Atwater

Project Outcomes: The team built a tool with Python to determine where high wage jobs are located in California. This tool serves as an extension to current data tools created and maintained by the organization. The team also developed a pipeline to clean and prepare new public data when it is released, and for the tool’s outputs to be regularly updated given any new data.

ACLU Criminal Justice

Our Team: Qianyun Li

Goal: At the ACLU, the student identified potential discrimination in school suspensions by performing feature importance analysis with machine learning models and statistical tests.

ACLU Micromobility

Our Team: Max Shinnerl

Goal: At the ACLU, the student analyzed COVID-19 vaccine equitable distribution data. They developed interactive maps with Leaflet to visualize shortcomings of the distribution algorithm and automated the cleaning of legislative record data. They also developed a pipeline for storing data to enable remote SQL queries using Amazon RDS and S3 from AWS.

Human Rights Data Analysis Group (HRDAG)

Our Team: Bing Wang

Goal: At the Human Rights Data Analysis Group (HRDAG), Bing gleaned critical location of death information from unstructured text fields in Arabic using Google Translate and Python Pandas, adding identifiable records to Syrian conflict data. She wrote R scripts and bash Makefiles to create blocks of similar records on killings in the Sri Lankan conflict to reduce the size of search space in the semi-supervised machine learning record linkage (database deduplication) process.

Toward Realignment: Big Tech, Organized Labor, and the Politics of the Future of Workby, Nantina Vgontzas, Sage Journals

Affective Engagement in #StopAAPIHate on Social Media: The Role of Emotion in Driving Engagement for Counter-hate Content on Twitter, USF CRASE Blog

• “Inside DeepMind's Secret Plot to Break Away From Google”, Business Insider
• “Social Media Content Moderation Is Not Neutral, USF Researcher Says”, SF Public Press
• “How to poison the data that Big Tech uses to surveil you”, Technology Review
• "To Live in Their Utopia: Why Algorithmic Systems Create Absurd Outcomes" by Ali Alkhatib at [CHI 2021] - also available as a video summary on YouTube
• "The politicization of face masks in the American public sphere during the COVID-19 pandemic" by Scoville, C., McCumber, A., Amironesei R., Jeon, J. at the American Sociological Association
• "On the Genealogy of Machine Learning Datasets: A Critical History of ImageNet" by Denton, E., Hanna, A., Amironesei, R., Smart, A., Nicole, H. at Big Data and Society
• "Notes on Problem Formulation" by Amironesei, R., Denton, E., Hanna, A. at IEEE Technology and Society Magazine Journal
• "Algorithmic Conservation in a Changing Climate" by Scoville, C., Chapman, M., Amironesei, R., and Boettiger, C. at Current Opinion in Sustainability Journal
• "'You Can’t Sit With Us': Exclusionary Pedagogy in AI Ethics Education" by Raji, I.D., Scheuerman, M.K., Amironesei, R. at FAccT
• "Genealogy, Archeology, Hermeneutics: Techniques of Interpretation in Machine Learning Datasets," by Amironesei, R., Denton, E., Hanna, A. at [IEEESSIT]
• "Bringing the People Back In: Contesting Benchmark Machine Learning Datasets" by Denton, E., Hanna, A., Amironesei, R., Smart, A., Nicole, H., Scheuerman, M.K. at arXiv

• “The Dark Side of Big Tech’s Funding for AI Research”
• “Google workers reject company's account of AI researcher's exit as anger grows”
• “Is your boss spying on you while you work remotely?”
• “Open Letters by Tech Industry, Google Employees Criticize Google’s Lack of Transparency in AI Research”
• "Bridging Data Science with Ignatian Spirituality"