FutureHouse Unveils AI Tools It Hopes Will Speed Up Scientific Progress

Backed by Eric Schmidt, Nonprofit Launches Platform to Support Researchers

FutureHouse, a nonprofit backed by former Google CEO Eric Schmidt, has taken a bold step toward its mission of building an “AI scientist” within the next ten years. This week, the organization introduced its first major offering: a platform and API equipped with four AI-powered tools designed to support and accelerate scientific research.

A Crowded Race to Automate Research

FutureHouse enters a fast-growing field where numerous startups—and major tech players—are vying to automate aspects of the scientific process using artificial intelligence. Earlier this year, Google introduced its own AI system dubbed the “AI co-scientist,” claiming it could assist researchers in generating hypotheses and structuring experiments.

Prominent voices in the AI world are optimistic. The CEOs of OpenAI and Anthropic have both suggested that AI could significantly speed up discovery in fields like medicine. However, skepticism remains. Many in the scientific community argue that today’s AI is still too unreliable to guide core scientific processes.

Meet the Tools: Crow, Falcon, Owl, and Phoenix

The newly released FutureHouse tools include:

• Crow: Searches and summarizes scientific literature in response to user questions.

• Falcon: Conducts deeper research, pulling from specialized scientific databases.

• Owl: Helps identify previous work in a chosen area of study.

• Phoenix: Assists in designing chemistry experiments using domain-specific tools.

According to FutureHouse’s official blog post, “Unlike other [AIs], FutureHouse’s have access to a vast corpus of high-quality open-access papers and specialized scientific tools.” The company also emphasizes a multi-stage reasoning process that allows these tools to assess sources in greater depth. “By chaining these [AI]s together, at scale, scientists can greatly accelerate the pace of scientific discovery,” they claim.

Promises vs. Proof

Despite the ambitious rollout, FutureHouse has yet to produce a scientific breakthrough or uncover a novel finding using its tools.

One of the most difficult aspects of building a functional AI scientist lies in addressing countless unpredictable variables. While AI may be well-suited for narrowing down options or exploring broad datasets, its effectiveness in solving complex, novel problems remains uncertain.

A notable example comes from Google’s AI project GNoME. In 2023, the company reported that the AI had helped synthesize approximately 40 new materials. However, an external review found that none of the materials were genuinely new discoveries.

Reliability Still a Stumbling Block

AI’s known limitations—particularly its tendency to produce inaccurate or fabricated information—continue to raise concerns among researchers. Even well-structured experiments could be compromised by errors in AI-generated data or methodology, particularly in fields that demand high precision.

FutureHouse acknowledges these concerns, especially regarding its chemistry-planning tool Phoenix. “We are releasing [this] now in the spirit of rapid iteration,” the team writes. “Please provide feedback as you use it.”

Author: Din Kumar