.svg)
Basis is a nonprofit applied research organization with a distinctive mission to understand and build intelligence through fundamental mathematical principles while simultaneously advancing society's ability to solve intractable problems. The organization approaches intelligence research by establishing mathematical foundations of reasoning, learning, decision-making, understanding, and explanation, then constructing software that implements these principles at scale.
The organization develops technology capable of reasoning about and operating in environments with boundless complexity and detail. Basis focuses on representing and discovering models of phenomena at unprecedented fidelity and scale, incorporating knowledge from diverse sources including data, experiments, interactions, and human expertise. The research philosophy centers on uncovering fundamental principles of reasoning that confer general problem-solving abilities, rather than creating increasingly complicated technology.
Basis leverages ideas from programming languages, compilers, and databases to uncover general and reusable principles of reasoning, combined with deep learning approaches for scalable approximate algorithms. The organization publishes research at leading venues and develops open-source software as its primary research medium, serving as a bidirectional interface with the broader research community. ChiRho, an experimental language for causal reasoning, exemplifies this approach.
The organization tackles challenge problems that distill general technology from collaboration with domain experts. Current projects include developing rational automated robot design agents that combine language models with probabilistic programming, participatory city modeling platforms that give agency to diverse stakeholders, collaborative intelligent systems research spanning multiple species and scales, intuitive scientific discovery through active program synthesis, and dynamical models for impossible biological measurements.
Basis operates as a 501(c)(3) nonprofit with team members, postdoctoral fellows, research trainees, and advisors from institutions including MIT, Columbia, Cornell, the Broad Institute, and Brown University. The organization maintains an open collaborative model, actively soliciting partnerships with researchers and domain experts from academia, government, social sectors, and industry.
- Building automated robot design systems that combine language models with CAD environments and probabilistic programming
- Developing participatory city modeling platforms for rezoning, street life, and budgeting decisions
- Creating software tools for understanding collaborative behaviors across species and ecosystems
- Constructing active program synthesis systems that build models of real and simulated environments
- Learning dynamical models of cell and tissue multi-omic state by integrating incomplete causal knowledge
- Publishing open-source software for causal reasoning and probabilistic machine learning
- Conducting research on world models, reinforcement learning, and embodied reasoning systems
- Applying compilers and programming language techniques to uncover fundamental reasoning principles
- Developing neurosymbolic methods for program synthesis and automated programming
- Training postdoctoral fellows in AI research focused on understanding and building intelligence
- Collaborating with external organizations on socially-important challenge problems
- Researching collaborative intelligent systems to inform policy decisions and scientific understanding
