We are a nonprofit AI-for-science research lab in San Francisco. We were founded in 2023 by Sam Rodriques and Andrew White. Our mission is to build AI systems that can scale scientific research and accelerate the pace of discovery, so humanity can proceed as quickly as possible to find cures for disease, solutions for climate change, and other species-accelerating technologies.

How long have you been around for?

FutureHouse was founded in 2023.

How large is your team?

We have somewhere between 10 and 15 people, depending on when this was last updated.

What is your relationship to Edison Scientific?

Edison Scientific is a commercial spinout of FutureHouse.

What is an AI Scientist?

AI Scientists take broad scientific questions and autonomously generate hypotheses, design experiments, analyze data, and produce novel discoveries.

FutureHouse has been building toward this since 2023. Here's how we think about it — and why it matters.

The Idea

The term "AI Scientist" has been around longer than most people realize. It first appeared in the 1980s, referring to human scientists who work on artificial intelligence. The meaning flipped around 2008 to mean computer systems that do science. In recent years, the concept has moved from academic speculation to an active and fast-growing field.

Today, "AI Scientist" gets used loosely. Sometimes it means any system that uses AI to automate some piece of the scientific process. Under that broad definition, a chatbot with a literature search tool would qualify.

FutureHouse co-founder Andrew White has proposed something more precise:

An AI Scientist is a system whose input is a general direction of discovery and whose output is experimental results, analysis, and a paper describing a novel discovery — one that could pass peer review in a specific domain.

That's the bar. It's high. No system has fully cleared it yet, including ours. But the pieces are falling into place, and they're doing so faster than most people expected.

Four Layers of Science Automation

Not every AI system that touches science is an AI Scientist. We think about science automation in four layers — a framework FutureHouse developed to distinguish between what exists today and what we're building toward.

Layer 1 is the Human. Contains "The Quest". Examples of a Quest include: figure out how the human brain works; gain the ability to deliver any gene to any cell in the body; and so on.

Layer 2 is the AI Scientist. A continuously updating world model, through hypothesis generation and experimentation.

Layer 4 is the AI Tools. Contains predictive models (e.g. AlphaFold), APIs, laboratory experiments.

Let's explore each layer.

1. AI Tools

At the bottom are AI tools, which are models and interfaces that help make predictions or manipulate the world — protein structure prediction, molecular simulations, lab automation hardware. This is where the field is currently most crowded.

2. AI Assistant

An AI assistant is a language model agent that executes specific tasks. An AI science assistant can perform tasks like searching the literature, analyzing a dataset, or annotating protein functions. They go beyond tools — they follow multi-step instructions and reason about scientific context. FutureHouse built several of these, including PaperQA2, which achieves expert-level performance on scientific literature search, and specialized agents for chemistry, data analysis, and precedent research.

3. AI Scientist

An AI Scientist actually runs a loop: it builds a model of the world, generates hypotheses from that model, tests them, and updates the model based on new data. This is the core of what science is. FutureHouse's Robin demonstrated this loop: it autonomously generated and validated a therapeutic hypothesis for dry age-related macular degeneration in 2.5 months. Edison Scientific’s Kosmos is our current AI Scientist that can conduct complex analyses.

4. Human

Humans provide what we call a "quest," which is a broad scientific direction, not a specific question. Something like: build a complete, mechanistic model for the brain. Or: annotate all the proteins of unknown function in the genome. The AI Scientist takes it from there. We'll know we've successfully built an AI Scientist when a human sets a quest and the system comes back with a discovery supported by experiments, analysis, and a written paper.

01

AI Tools

AI Assistant

AI Scientist

Human

At the bottom are AI tools, which are models and interfaces that help make predictions or manipulate the world — protein structure prediction, molecular simulations, lab automation hardware. This is where the field is currently most crowded.

An AI assistant is a language model agent that executes specific tasks. An AI science assistant can perform tasks like searching the literature, analyzing a dataset, or annotating protein functions. They go beyond tools — they follow multi-step instructions and reason about scientific context. FutureHouse built several of these, including PaperQA2, which achieves expert-level performance on scientific literature search, and specialized agents for chemistry, data analysis, and precedent research.

An AI Scientist actually runs a loop: it builds a model of the world, generates hypotheses from that model, tests them, and updates the model based on new data. This is the core of what science is. FutureHouse's Robin demonstrated this loop: it autonomously generated and validated a therapeutic hypothesis for dry age-related macular degeneration in 2.5 months. Edison Scientific’s Kosmos is our current AI Scientist that can conduct complex analyses.

Humans provide what we call a "quest," which is a broad scientific direction, not a specific question. Something like: build a complete, mechanistic model for the brain. Or: annotate all the proteins of unknown function in the genome. The AI Scientist takes it from there. We'll know we've successfully built an AI Scientist when a human sets a quest and the system comes back with a discovery supported by experiments, analysis, and a written paper.

The Challenge

An AI Scientist doesn't replace human scientists. It multiplies them.

A single researcher working with an AI Scientist can cover ground that would previously require a full team. Our AI Scientist Kosmos can compress 6 months of research into a single day.

For diseases that affect millions but don't attract large research programs. For rare conditions. For questions that fall between disciplines. For the thousands of important biological problems that simply don't have enough people working on them, this changes the equation.

That's where FutureHouse lives.

What we're doing