From election forecasts to Wall Street trading to sports analytics, prediction has actually gotten worse when computers generate it. The "Uses and Abuses of Prediction Seminar Series" highlights how prediction shapes our outlooks and decisions—and why it falters.
This event is co-sponsored by the Departments of English, Economics, Political Science, and Statistics and the Institute for Cyber Law, Policy, and Security.
Spring 2023
"Textual Speculations: How Generative AI Predicts the Next Word"
Annette Vee
April 19 at 1 PM ET
Via Zoom: Watch Here
In-person: Posvar hall 3911
The current discourse around generative AI is steeped in speculation: how effective can large language models get? How will they affect employment and education? And are they leading to artificial general intelligence (AGI)? But beyond the discourse, the models themselves are built on speculation: drawing from a giant dataset of natural language in text, they predict the next word in a sequence. Earlier approaches to natural language generation (such as Markov models) also predicted the next word, but recent large language models (LLMs) combine more complicated algorithms, concepts of attention, and larger datasets to conceal their predictive nature and produce far more coherent and plausible natural language. Yet AI writing detectors operate on this idea that AI writing is more predictable than that of humans: humans tend to write with greater "burstiness" and "perplexity." With the contrast between human and AI writing as a framing device, this talk traces the ways that prediction has operated in generative AI and other historical attempts to automate writing. Attendees of the talk will come away with an understanding of: current Large Language Models driving generative AI writing and how they differ from earlier models; how AI models do and don't replicate human writing; and the practical effects of generative AI in writing and pedagogy.
Annette Vee is Associate Professor of English and Director of the Composition Program at University of Pittsburgh, where she teaches writing and digital composition. She is the author of Coding Literacy (MIT Press, 2017) and has published on computer programming, digital literacy, blockchain technologies, intellectual property, and AI-based text generators.
"Automating Early Warning: The Possibilities and Limits of Predicting Conflict"
Christopher Rauh and Hannes Mueller - March 28 at 1 PM ET
Watch the Discussion Here
About the program
The presentation will showcase the approach to conflict forecasting used for conflictforecast.org that leverages the power of machine learning and natural language processing. By analyzing patterns in newspaper text, algorithms can identify indicators of potential conflict and develop early warning systems for policymakers and other stakeholders. The presentation will highlight the key features and trade-offs of this approach, including its scalability and accuracy. We will also discuss some of the key challenges and limitations of conflict forecasting in general, and our approach in particular. Finally, we will illustrate how predictions can be used to support decision making when considering when and where to prevent conflict or to intervene. The presentation will feature case studies and real-world examples to illustrate the potential of this approach.
About the Presenters
Christopher Rauh is a Professor at the University of Cambridge, Research Professor at PRIO, Fellow of Trinity College Cambridge, and a Research Affiliate at CEPR and HCEO. His fields are Labor Economics and Political Economy. He is a co-founder of conflictforecast.org, a website providing monthly predictions about conflict risk. He has published in top Economics and Political Science journals, such as American Political Science Review, Journal of European Economic Association, and Journal of Public Economics, and has led to projects with the German Foreign Office and the Foreign, Commonwealth & Development Office. His work has been featured widely across the media including the Economist, The Guardian, Washington Post, the BBC, FAZ, and Der Spiegel, and Bloomberg.
Hannes Mueller is a tenured researcher at the Institute for Economic Analysis (IAE/CSIC) and an Associated Research Professor at the Barcelona School of Economics (BSE). He is affiliated to the CEPR Development Economics program since 2015 and a Research Fellow since 2022. He publishes in leading journals in science, economics and political science such as the American Economic Review (AER), the American Political Science Review (APSR), the Journal of the European Economic Association (JEEA) and the Proceedings of the National Academy of Sciences (PNAS). In the last five years Hannes has specialized in the use of supervised and unsupervised machine learning methods in applications in economic and political science. He directs the Masters in Data Science for Decision Making at the BSE and numerous projects that introduce heterogenous data like text or images into social science research. One of the projects is the development of the conflict forecast webpage conflictforecast.org. This work has become a key resource for governments and international organizations engaged in conflict prevention and has led to collaborations and research contracts with the Spanish central bank (BdE), the German foreign office, the UK Foreign, Commonwealth & Development Office, the IMF, several UN organizations, the World Bank and numerous NGOs.
Predicting Well-Being in the Real-World and Real-Time: Possibilities and Challenges
Saida Heshmati
January 31, 1 PM ET
A key part of grasping a fuller understanding of human flourishing for creating a culture of health involves considering well-being as a continual process of healthy functioning that unfolds in context and over time, rather than a static endpoint of wellness. Human flourishing in the real-world and in real-time is often characterized by person-specific nuances that are most often clouded by aggregate-level assessments, overcasting the researcher’s view of the underlying contextual and cultural causalities. Using a dynamical systems approach, I will demonstrate Ecological Momentary Assessment and other field-based designs, ecologically-valid measurement tools, and the analysis of intensive longitudinal data to uncover the complexities of individualized social and behavioral dynamics that shape health and well-being.
About the speaker: Dr. Saida Heshmati is Assistant Professor of psychology at Claremont Graduate University. Her research lies in the understanding of how optimal development unfolds over time in diverse samples through dynamical systems perspectives. Using her expertise in human development and state-of-the-art analytical methods, she examines large datasets related to individual and group characteristics that influence psychological well-being as part of positive development. Through her work, she aims to bring together a suite of measurement tools and research designs in the service of developing idiographic, culturally-informed, and context-sensitive approaches to understanding optimal development in youth, in particular those who are marginalized. Dr. Heshmati has a multicultural background which has informed her scientific research; she is an Iranian-American scholar and an immigrant who has lived in five different countries and travelled to more than 20 countries, and still counting.
Fall 2022
"Algorithms in Criminal Justice"
Megan Stevenson
December 8, 1 PM ET
Watch the discussion here.
Megan Stevenson is an economist, criminal justice scholar, associate law professor, and professor of Economics at the University of Virginia. She conducts empirical research in various areas of criminal justice reform, including bail, algorithmic risk assessment, misdemeanors and juvenile justice. She publishes in both law reviews and economic journals, including the Stanford Law Review, the Washington University Law Review, the Minnesota Law Review, the Boston College Law Review, the Boston University Law Review, the Review of Economics and Statistics, and the Journal of Law, Economics, & Organization.
"The Art and Science of Election Polling"
Elliott Morris and Michael Colaresi
October 18, 1 PM ET
Watch the discussion here.
G. Elliott Morris is a staff data journalist and US correspondent for The Economist. He writes about American politics, public opinion polling, demographics, and elections. He is responsible for many of the paper’s election forecasting models, including the 2020 US presidential election forecast and polling models for several European countries. He writes for The Economist's weekly “Checks and Balance” newsletter on US politics. He is proficient in machine learning models, Bayesian statistics, and the various tools in the standard social science toolkit.
Michael Colaresi is the William S. Dietrich II Chair of Political Science and the research and academic director of Pitt Cyber, as well as the director of the Pitt Disinformation Lab. His work leverages the accelerating availability of computational tools, including machine learning and Bayesian approaches, along with unstructured information, such as from digitized text, to build and improve models of information technology in democracies, national security secrecy and oversight, international and intrastate violence, and changes in human rights over time. He also develops computational and visual tools that enable domain specialists to work alongside computer scientists to improve specific applications. In 2022-23, he is a fellow of the Stability and Change program at the Center for Advanced Studies in Oslo, Norway. He was the co-editor of the journal International Interactions from 2014-2019 and was co-recipient of the Best Visualization Award from the Journal of Peace Research in 2017 and the Gosnell Prize for Excellence in Political Methodology from the Methodology section of the American Political Science Association in 2006. His book Democracy Declassified was shortlisted for the 2015 Conflict Research Society Book Prize. He has been PI or co-PI on four NSF grants and is a research affiliate for the ERC-funded Violence Early Warning Project at the University of Uppsala and the Peace Research Institute Oslo. At the University of Pittsburgh he co-founded the new major in Computational Social Science and in his previous position at Michigan State University, he founded and directed the Social Science Data Analytics initiative.
“Expertise and Bad Predictions: How Can We Do Better?”
Gayle Rogers, Illah Nourbakhsh, and Jennifer Keating
September 13
Watch the discussion here.
Gayle Rogers is an Andrew W. Mellon professor and chair of English at the Dietrich School Special Liaison for Outreach and Development. He is also an affiliated faculty with the Global Studies Center, Center for Latin American Studies, European Studies Center, and Cultural Studies program. He works primarily on the topics of risk and prediction, the history of ideas, global modernisms, translation theory, comparative literature, critical history, and the intersections of literature, economics, and risk theory.