New Paper: The Cost of Generative AI Collaboration Strategies to the Psychological Experience of Work
New Paper: A Simple Explanation Reconciles “Algorithm Aversion” and “Algorithm Appreciation”: Hypotheticals vs. Judgments
New Paper: A Simple Explanation Reconciles “Algorithm Aversion” and “Algorithm Appreciation”: Hypotheticals vs. Judgments
Why do people make inaccurate judgments?
When are people willing to use algorithms to improve their accuracy?
What is the optimal way to collaborate with GenAI?
|
Jennifer M. Logg, Ph.D., is an Assistant Professor of Management at Georgetown University's McDonough School of Business. Prior to joining Georgetown, she was a Post-Doctoral Fellow at Harvard University at Harvard Business School and Harvard Kennedy School. She received her Ph.D. from the University of California, Berkeley’s Haas School of Business.
Accuracy & Algorithms Logg's research examines why people fail to view themselves and their work realistically. It focuses on how individuals can assess themselves and the world more accurately by using advice and feedback from algorithms. She studies perceptions of algorithms in her primary line of research, Theory of Machine. It tests how people respond to algorithmic advice. Broadly, this work examines how people expect algorithmic and human judgment to differ. The Psychology of GenAI GenAI needs Psychology. Logg's recent research takes a cost-benefit perspective to GenAI by testing how different GenAI Collaboration Strategies impact both productivity and the psychological experience of work. The F.B.I., U.S. Senate, Air Force, and Navy have invited her to speak with decision-makers on the topic of algorithms and predictive accuracy . During her Ph.D., she was a collaborator on the Good Judgment Project, funded by IARPA, Intelligence of Advanced Research Projects Activity, the US intelligence community’s equivalent of DARPA. She served on the faculty committee that created the Masters of Science in Business Analytics at Georgetown University's McDonough School of Business. She was a member of the "Theory of AI Practice" working group, funded by the Rockefeller Foundation through Stanford University's Center for Advanced Study in the Behavioral Sciences. Currently, she is a Faculty Fellow at Georgetown University's AI, Analytics, and the Future of Work Initiative. |
New Working Papers
The Cost of Generative AI Collaboration Strategies to the Psychological Experience of Work
Generative AI (GenAI) is transforming the way people work by increasing productivity and changing workflows. Yet, little is known about the optimal way to use it in order to balance both productivity and the psychological experience of work (i.e., control, responsibility, and creativity). We examined different strategies, which we call GenAI collaboration strategies, and their associated productivity benefits and psychological costs. In two pre-registered experiments (N = 1,299), we examined two common uses of GenAI, using it to draft versus edit. First, using GenAI, relative to not using it, improved productivity but degraded control, responsibility, and creativity. Second, we identified the optimal GenAI collaboration strategy: using GenAI to edit. Productivity was similar between collaboration strategies, but using GenAI to draft further hurt people's psychological experience of work. Yet, the majority of people chose the suboptimal strategy, including a sample of teachers creating lesson plans for a class. GenAI is a powerful tool to increase productivity, but it is not without costs. Thus, it is not enough for organizations to provide employees with GenAI; how people use GenAI has implications for their psychology.
A Simple Explanation Reconciles “Algorithm Aversion” and “Algorithm Appreciation”: Hypotheticals vs. Judgments
(Top 10 SSRN Download List in 3 Topic Categories within 1 month)
We propose a simple explanation to reconcile research documenting algorithm aversion with research documenting algorithm appreciation: elicitation methods. We compare self-reports and judgments. When making actual judgments, people consistently utilize algorithmic advice more than human advice. In contrast, hypotheticals produce unstable preferences; people sometimes report indifference and sometimes report preferring human judgment. Moreover, people fail to correctly anticipate behavior, utilizing algorithmic advice more than they anticipate. A slight change in the framing of a hypothetical task additionally moderates algorithm aversion. Stated preferences about algorithms are less stable than actual judgments, suggesting that algorithm aversion may be less stable than previous research leads us to believe.
Risk Creep: Learning Under High Uncertainty is Associated with a Creeping Tolerance for Risk
(Top 10 SSRN Download List in 5 Topic Categories within 2 months)
According to The Federal Reserve, uncertainty has reached its “highest levels in decades” based on economic, political, and geopolitical data (Londono, Ma, & Wilson, 2025). But we know little about how people learn when uncertainty stems from both environmental and social uncertainty, what we call high uncertainty. Can people naturally learn under high uncertainty with only limited and inconsistent information to inform behavior? In order to test the boundary conditions of decision making under uncertainty, we tracked perceptions and behavior at the height of uncertainty during the COVID-19 pandemic. We measured whether risk perceptions of people’s own behavior and observations of others’ behavior in the past week were associated with current behavior, which we call minimal experiential learning and minimal signal learning. Tracking the co-evolution of multiple learning mechanisms over time, we find that, ruling out habituation, both were associated with a gradual increase in precarious activity that put people at risk of negative consequences. We call this risk creep and suspect that it is a common by-product of learning under high uncertainty.
Hybrid Intelligence: A Paradigm for More Responsible Practice
(Stanford's Center for Advanced Study in the Behavioral Sciences White Paper)
We propose an alternate approach to mainstream AI practice that broadens the focus beyond algorithms viewed in isolation to processes of human-algorithm collaboration. The envisioned practice would harness human and machine complementarities to develop systems of human-machine hybrid intelligence.
Research Awards & Rankings
Algorithm Appreciation
- 2,700+ Citations
- Ranked #1 in 2021 as "Most Cited Organizational Behavior and Human Decision Processes Articles Since 2018"
- Top 10% of authors on SSRN (by new downloads) 2017-Present
Is Overconfidence a Motivated Bias?
- Early Career Award as judged by the Journal of Experimental Psychology's editors, from five sections (2019)
Teaching Awards
Top 50 Undergrad B-School Profs: Poets & Quants (2021)
Georgetown Career Champion: student nominated (2022)
The Cost of Generative AI Collaboration Strategies to the Psychological Experience of Work
Generative AI (GenAI) is transforming the way people work by increasing productivity and changing workflows. Yet, little is known about the optimal way to use it in order to balance both productivity and the psychological experience of work (i.e., control, responsibility, and creativity). We examined different strategies, which we call GenAI collaboration strategies, and their associated productivity benefits and psychological costs. In two pre-registered experiments (N = 1,299), we examined two common uses of GenAI, using it to draft versus edit. First, using GenAI, relative to not using it, improved productivity but degraded control, responsibility, and creativity. Second, we identified the optimal GenAI collaboration strategy: using GenAI to edit. Productivity was similar between collaboration strategies, but using GenAI to draft further hurt people's psychological experience of work. Yet, the majority of people chose the suboptimal strategy, including a sample of teachers creating lesson plans for a class. GenAI is a powerful tool to increase productivity, but it is not without costs. Thus, it is not enough for organizations to provide employees with GenAI; how people use GenAI has implications for their psychology.
A Simple Explanation Reconciles “Algorithm Aversion” and “Algorithm Appreciation”: Hypotheticals vs. Judgments
(Top 10 SSRN Download List in 3 Topic Categories within 1 month)
We propose a simple explanation to reconcile research documenting algorithm aversion with research documenting algorithm appreciation: elicitation methods. We compare self-reports and judgments. When making actual judgments, people consistently utilize algorithmic advice more than human advice. In contrast, hypotheticals produce unstable preferences; people sometimes report indifference and sometimes report preferring human judgment. Moreover, people fail to correctly anticipate behavior, utilizing algorithmic advice more than they anticipate. A slight change in the framing of a hypothetical task additionally moderates algorithm aversion. Stated preferences about algorithms are less stable than actual judgments, suggesting that algorithm aversion may be less stable than previous research leads us to believe.
Risk Creep: Learning Under High Uncertainty is Associated with a Creeping Tolerance for Risk
(Top 10 SSRN Download List in 5 Topic Categories within 2 months)
According to The Federal Reserve, uncertainty has reached its “highest levels in decades” based on economic, political, and geopolitical data (Londono, Ma, & Wilson, 2025). But we know little about how people learn when uncertainty stems from both environmental and social uncertainty, what we call high uncertainty. Can people naturally learn under high uncertainty with only limited and inconsistent information to inform behavior? In order to test the boundary conditions of decision making under uncertainty, we tracked perceptions and behavior at the height of uncertainty during the COVID-19 pandemic. We measured whether risk perceptions of people’s own behavior and observations of others’ behavior in the past week were associated with current behavior, which we call minimal experiential learning and minimal signal learning. Tracking the co-evolution of multiple learning mechanisms over time, we find that, ruling out habituation, both were associated with a gradual increase in precarious activity that put people at risk of negative consequences. We call this risk creep and suspect that it is a common by-product of learning under high uncertainty.
Hybrid Intelligence: A Paradigm for More Responsible Practice
(Stanford's Center for Advanced Study in the Behavioral Sciences White Paper)
We propose an alternate approach to mainstream AI practice that broadens the focus beyond algorithms viewed in isolation to processes of human-algorithm collaboration. The envisioned practice would harness human and machine complementarities to develop systems of human-machine hybrid intelligence.
Research Awards & Rankings
Algorithm Appreciation
- 2,700+ Citations
- Ranked #1 in 2021 as "Most Cited Organizational Behavior and Human Decision Processes Articles Since 2018"
- Top 10% of authors on SSRN (by new downloads) 2017-Present
Is Overconfidence a Motivated Bias?
- Early Career Award as judged by the Journal of Experimental Psychology's editors, from five sections (2019)
Teaching Awards
Top 50 Undergrad B-School Profs: Poets & Quants (2021)
Georgetown Career Champion: student nominated (2022)
|
|
|
|
Aug. 8, 2019
Algorithms as Magnifying Glasses: Using Algorithms to Understand the Biases in Your Organization Harvard Business Review |
|
Oct. 26, 2018
Stop Naming Your Algorithms: Do People Trust Algorithms More Than Companies Realize? Harvard Business Review |
|
|
May 10, 2019
Harnessing the Wisdom of Crowds: How Asking Multiple People for Advice Can Backfire Harvard Business Review |
|
Nov. 17, 2020
A Social Perspective of a Cognitive Bias: Overconfidence is Contagious Harvard Business Review |