Luke Hewitt

Artificial Intelligence researcher and computational social scientist

I’m co-founder/director of Rhetorical Impact Lab, a research org which uses RCT experiments and machine learning to help public communication campaigns improve the impact of their messaging.
I’m a Senior Research Fellow at the Stanford Polarization and Social Change lab, where I study the capacity of Large Language Models to predict treatment effects in social/behavioral sciences.
I’m co-PI for the SSRC Mercury Project team on Combatting health misinformation with community-crafted messaging.

Previously:

My PhD was advised by Josh Tenenbaum (MIT), primarily developing scalable Bayesian methods for explainable AI. For my thesis I also conducted the largest RCT meta-analysis of political advertisements to date, working with David Broockman, Alex Coppock and Ben Tappin.
I worked on research methods at Swayable, designing their 2020 national polling methodology which successfully predicted Biden’s vote share to within 0.5pp (compared to the 538 bias of 2pp).

↓

Academic research by topic

‣

Political science

Political campaign advertising • How experiments help campaigns persuade voters: evidence from a large archive of campaigns’ own experiments (Hewitt et al. 2023)

Targeted messaging • Quantifying the persuasive returns to political microtargeting (Tappin et al. 2022) • Rank-heterogeneous effects of political messages: Evidence from randomized survey experiments testing 59 video treatments (Hewitt et al. 2022)

Persistence • Estimating the Persistence of Party Cue Influence in a Panel Survey Experiment (Tappin et al. 2021)

‣

Machine Learning

Structured generative models • Hybrid memoised wake-sleep: Approximate inference at the discrete-continuous interface (Le et al. 2022) • Learning to learn generative programs with memoised wake-sleep (Hewitt et al. 2020)

Program synthesis • DreamCoder: Bootstrapping Inductive Program Synthesis with Wake-Sleep Library Learning (Ellis et al. 2021) • Learning to infer program sketches (Nye et al. 2019)

Deep generative models • The Variational Homoencoder: Learning to learn high capacity generative models from few examples (Hewitt et al. 2018)

‣

Cognitive science

Emotion • Emotion prediction as computation over a generative Theory of Mind (Houlihan et al. 2023)

Perception • Bayesian auditory scene synthesis explains human perception of illusions and everyday sounds (Cusimano et al. 2023) • Auditory scene analysis as Bayesian inference in sound source models (Cusimano et al. 2017)

Concept learning • Inferring structured visual concepts from minimal data (Qian et al. 2019)

[CV]