MARA
Modeling, Abstraction, and Reasoning Agents: systems that build and use world models through active experimentation and abstract reasoning.
People
Zenna Tavares
Co-founder and Director, Basis Research Institute
Zenna Tavares builds computational and statistical tools for causal reasoning, probabilistic programming, and scientific model discovery. His work asks how machines can reason from observation and interaction in ways that support scientific and societal problem-solving.
About
My research aims to understand human reasoning: how people derive knowledge from observing and interacting with the world. I develop computational and statistical tools for causal reasoning, probabilistic programming, and scientific model discovery.Projects
Current and recent Basis projects.
R-ADA
A rational automated design agent for robotics, combining language models, simulation, probabilistic programming, and Bayesian inference.
Citymaking through participatory modeling
Participatory city models that help residents, community groups, and policymakers reason about uncertain causes and policy consequences.
Recent Publications
Recent papers and preprints.
Benchmarking World-Model Learning
A. Warrier, D. Nguyen, M. Naim, M. Jain, Y. Liang, K. Schroeder, C. Yang, J. B. Tenenbaum, S. J. Vollmer, K. Ellis, Z. Tavares
ExoPredicator: Learning Abstract Models of Dynamic Worlds for Robot Planning
Y. Liang, D. Nguyen, C. Yang, T. Li, J. B. Tenenbaum, C. E. Rasmussen, A. Weller, Z. Tavares, T. Silver, K. Ellis
Assessing adaptive world models in machines with novel games
L. Ying, K. M. Collins, P. Sharma, C. Colas, K. I. Zhao, A. Weller, Z. Tavares, P. Isola, S. J. Gershman, J. Andreas, T. L. Griffiths, F. Chollet, K. R. Allen, J. B. Tenenbaum
NeuroAI for AI Safety
P. Mineault, N. Zanichelli, J. Z. Peng, A. Arkhipov, E. Bingham, J. Jara-Ettinger, E. Mackevicius, A. H. Marblestone, M. Mattar, A. Payne, S. Sanborn, K. Schroeder, Z. Tavares, A. S. Tolias
Combining Induction and Transduction for Abstract Reasoning
W. Li, K. Hu, C. Larsen, Y. Wu, S. Alford, C. Woo, S. M. Dunn, H. Tang, M. Naim, D. Nguyen, W. Zheng, Z. Tavares, Y. Pu, K. Ellis
Articles
Basis essays and updates this person wrote or contributed to.
ExoPredicator: Abstracting Time and State for Robot Planning
We introduce ExoPredicator, a system that learns abstract world models for robot planning. By abstracting state, time, and both endogenous and exogenous causal processes, ExoPredicator enables robots to quickly learn how dynamic environments work and plan efficiently in them.
Building an Unverified Compiler with Agents
Four agents spent 14 days and 93,000 lines of Lean building a verified JS-to-WASM compiler from scratch. The compiler ran; the proofs didn’t close.
AutumnBench: World Model Learning in Humans and AI
We’re releasing a new version of Autumn with human baseline results, AI performance comparisons, and an interactive benchmark for world model discovery. This release includes the MARA protocol and provides a public platform for testing causal reasoning capabilities.
Basis: Designing a New Kind of AI Research Organization
Written in 2022, this founding document defines Basis’ vision and thesis of how to build a new kind of research organization.
Project MARA Preview: Modeling, Abstraction, and Reasoning Agents
Project MARA aims to develop AI systems capable of performing everyday scientific discovery through active experimentation and abstract reasoning. The project will create systems that can discover and apply causal models across diverse domains, from physical robotics to digital interfaces.
NeuroAI for AI Safety
Basis contributed to a new technical roadmap, “NeuroAI for AI Safety,” from Amaranth Foundation. The roadmap aims to make AI systems safer by understanding and implementing the brain’s approach to intelligent behavior.
MetaCOG: Enhancing AI Vision with Human-Inspired Metacognition
In collaboration with Marlene Berke and the Computational Social Cognition Lab at Yale, we’re introducing MetaCOG, a probabilistic model that can learn a metacognitive model of a neural object detector and use it to improve the detector’s accuracy without feedback. This represents a step towards building AI systems that can go beyond representing their inputs and also represent their own thought processes.
Autumn: Causal Discovery Through Program Synthesis
We’re introducing AutumnSynth, an algorithm that synthesizes the source code of simple 2D video games from a small amount of observed video data. This represents a step forward toward systems that can perform causal theory discovery in real-world environments.