Marie-Lou Laprise

Marie-Lou Laprise

Lawyer and computational social scientist

PhD candidate, Princeton University

About me

I am a PhD candidate at Princeton University, where I am also completing a graduate certificate at the Center for Statistics and Machine Learning. I study the impact of machine learning on social decision-making, with a focus on markets and regulation. My work at the interface of AI, finance, law, and sociology asks how data-driven technologies are reconfiguring flows and distributions of information and wealth, in financial markets and beyond. I also investigate complex reasoning in large language models and evaluate whether and how they can be successful at legal reasoning.

I am a co-PI at Princeton’s Debt Collection Lab, where our team studies debt collection lawsuits and inequality in the United States and help make data about private debt collectors accessible to the greater public.

In the private sector, I practiced as a commercial trial lawyer in downtown Montreal (Canada) for five years. I handled high-stakes litigation involving shareholder rights, complex tax planning, and trademark infringement, I advised business owners on their litigation risks, and I participated in a variety of corporate transactions.

Before coming to Princeton, I was at the Faculty of Law of the University of Montreal, where I studied judicial discretion and decision-making by judges and courts, and its relation to socioeconomic history. My thesis has won the 2022 excellence prize of the Quebec Association of Law Professors.

Interests
  • Law & AI, corporate finance, financial regulation
  • Decision-making, collective intelligence, human-AI interaction
  • Neural networks, large language models, dynamical systems and complexity
Education
  • Ph.D. in progress, 2020-...

    Princeton University, Department of Sociology and Center for Statistics and Machine Learning

  • LL.M. (comparative law, with thesis), 2021

    University of Montreal

  • LL.B. (civil law), 2013

    University of Sherbrooke

Research

My work at the interface of finance, law, sociology, and computer science and engineering explores how machine learning and computation technologies transform discretionary decision-making under uncertainty, reshape collective intelligence, and in so doing, change the dynamics of the circulation and accumulation of wealth and information in society. I also study what this means for our laws, and what kinds of regulatory frameworks we need to efficiently regulate AI and the data industry.

My dissertation investigates recent changes in the financial data and analytics industry and human-AI interaction in financial markets, and their impact on investment decisions and asset management practices, with an eye to regulatory concerns emerging from algorithmic trading, the changing face of stock exchange groups, and the rise of alternative data. Throughout my computational work, I employ local language models for help with data cleaning, record matching and text analysis.

In a separate line of work, I study complex reasoning in large language models, whether and how they can learn abstract rules in a way that allows them to generalize algorithmically and compositionally, and what this means for applications of AI in the legal field.

Other work in progress:

  • Divestment without Decarbonization: Private Equity and the Organizational Ecology of Dirty Work (co-authored with Adam Goldstein). We hypothesize that intensified pressures to adopt more sustainable business practices are prompting publicly-traded firms to shed environmental liabilities by divesting carbon-emitting assets. However, rather than reducing net emissions, carbon-intensive plants are relocating to more opaque privately-owned and private equity firms, which face fewer pressures to decarbonize. The result is a dynamic of divestment without decarbonization. We test this by constructing a unique linked dataset covering the period from 2010-2021 for the universe of more than 8000 facilities that reported emissions data to the EPA’s Greenhouse Gas Reporting Program, with time-varying data on the ownership of each facility’s parent firm. We presented this project at SASE 2023 in Rio.

  • The Real Effect of Analyst Forecasts: Managing Earnings and Uncertainty in Non-Financial Firms. In this project, I investigate earnings management practices and share repurchase operations by U.S. publicly traded firms over the past 20 years. Using multiple matched corporate datasets, I explore the dynamics of managers’ earnings guidance and sell-side analysts’ earnings forecasts, and the relational construction of earnings surprises. I replicate and extend previously published results in financial econometrics by showing that, contrary to what has been suggested, earnings surprises cannot be used as a causal instrument in most cases, given their embeddedness within managerial decision-making processes. More broadly, I discuss when and why endogeneity problems may result from using as an instrument or treatment a variable that encodes prediction practices within the field studied.