Luke Hewitt

I build computational tools to study what changes people’s beliefs, with application to AI safety, social science methodology, effective advocacy and public health communication.

Research | Photos | Twitter | LinkedIn | Google Scholar

Currently:

• AI safety consulting, UK AI Security Institute (evaluating LLM persuasion capabilities)
• Senior Research Fellow, Stanford PACS (using LLMs to simulate human experiments)
• Fellow, Future of Life Foundation (AI for Human Reasoning)

Previously:

• AI safety consulting, OpenAI (evaluating GPT-4o persuasion capabilities)
• Co-founder & CTO, Rhetorical Labs
• Member, South Park Commons
• Research data scientist, Swayable (persuasion measurement, expt. design & analysis)
• Co-PI, SSRC Mercury Project team on community-crafted messaging
• PhD in AI / Cognitive Science, MIT (advisor: Josh Tenenbaum)
• MEng in Mathematical Computation, UCL

↓

Research

→ The levers of political persuasion with conversational AI Hackenburg et al. (in review)

→ Encouraging vaccination using the creativity and wisdom of crowds Tappin et al. (in review)

→ Outcome-based Reinforcement Learning to Predict the Future Turtel et al. (working paper)

→ Large language models are more persuasive than incentivized human persuaders Schoenegger et al. (in review)

→ The impact of AI message-testing on public discourse Hewitt (IASEAI 2025)

→ Quantifying the returns to persuasive message-targeting using a large archive of campaigns’ own experiments* - Tappin, Hewitt, Coppock (APSA 2024)

→ How will advanced AI systems impact democracy? Summerfield et al. (in review)

→ Predicting results of social science experiments using large language models Hewitt*, Ashokkumar* et al. (in review)

→ GPT-4o System Card: Persuasion OpenAI (2024)

→ How experiments help campaigns persuade voters: evidence from a large archive of campaigns’ own experiments Hewitt et al. (APSR, 2024)

→ Using survey experiment pre-testing to support future pandemic response Tappin and Hewitt (PNAS Nexus, 2024)

→ Listening with generative models Cusimano et al. (Cognition, 2024)

→ Quantifying the persuasive returns to political microtargeting Tappin et al. (PNAS, 2023)

→ Emotion prediction as computation over a generative Theory of Mind Houlihan et al. (Phil. Trans. A, 2023)

→ DreamCoder: growing generalizable, interpretable knowledge with wake-sleep bayesian program learning Ellis et al. (Phil. Trans. A, 2023)

→ Rank-heterogeneous effects of political messages: Evidence from randomized survey experiments testing 59 video treatments Hewitt et al. (working paper)

→ Hybrid memoised wake-sleep: Approximate inference at the discrete-continuous interface Le et al. (ICLR, 2022)

→ DreamCoder: Bootstrapping Inductive Program Synthesis with Wake-Sleep Library Learning Ellis et al. (PLDI, 2021)

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

→ Learning to learn generative programs with memoised wake-sleep Hewitt et al. (UAI, 2020)

→ Inferring structured visual concepts from minimal data Qian et al. (CogSci, 2019)

→ Learning to infer program sketches Nye et al. (ICML, 2019)

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

→ Auditory scene analysis as Bayesian inference in sound source models Cusimano et al. (CogSci, 2017)