Research Link:
https://www.researchgate.net/publication/391988344_Behavioral_Distortions_Information_Frictions_and_the_Macroeconomics_of_AI_Adoption_A_FirmManager_Choice_Perspective_within_a_Substitution_Framework
Github Link to Simulation:
https://github.com/glombardo/Research/blob/main/AI_TFP_Gains_Guido_Lombardo.ipynb
Goal
This project builds a simple behavioral model on top of a macroeconomic framework to explain what happens to total factor productivity (TFP) when managers gain (or lose) by handing tasks to AI instead of humans. My goal is to encourage the integration of simulation methods, behavioral economics, and macroeconomic theory.
I Set Up the Model so Two traits drive the results:
1) Overconfidence (θ): how strongly a manager believes their judgement beats the “true’’ odds.
2) Incomplete Information (IA): how noisy their view of each task’s complexity is.
Using simulated task-level productivities, this I use this custom method to exploree three practical cases:
1) AI and workers are equally good.
2) AI clearly beats workers.
3) Workers clearly beat AI.
From the simulated patterns I propose four testable hypotheses:
1) Amplifier: confidence boosts performance when signals are clear.
2) Misfire: confidence hurts when signals are fuzzy and AI is worse.
3) Early-Dip Trap: small confidence can lower output before larger confidence corrects it.
4) Hedge-Then-Fade: cautious, poorly informed managers look fine at first but fall behind once their confidence grows.
I conclude with the following research agenda:
1) lab and field tests to measure θ and IA
2) richer models where AI and labour can complement each other
3) policy “guardrails’’ such as quick audits or human-in-the-loop rules to keep the upside of confidence while containing its downside.
