FutureHouse releases AI tools it claims can accelerate science

Future houseA non-profit organization supported by Eric Schmidt who wants to build an “AI scientist” in the following decade has launched its first major product: a platform and API with AI-driven tools that have been designed to support scientific work.

Many, many startups race to develop AI research tools for the scientific domain, and some have huge amounts of VC financing behind them. Tech Giants also resemble AI for science – Google revealed earlier this year “AI co-scientist‘That the company said they could help scientists make hypotheses and experimental research plans.

The CEOs of AI companies Openai and Anthropic have claimed That AI tools can greatly accelerate the scientific discovery of science, especially in medicine. But nowadays, many researchers do not consider AI as particularly useful in guiding the scientific process, largely because of its unreliability.

FutureHouse released four AI tools on Thursday: Crow, Falcon, OWL and Phoenix. CROW can search for scientific literature and answer questions about it; Falcon can carry out deeper literature studies, including scientific databases; Owl is previously looking for work in a certain field; And Phoenix uses tools to plan chemical experiments.

‘In contrast to others [AIs]Futurehouses have access to a huge corpus of high-quality open-access articles and specialized scientific tools, ”the non-profit organization written In a blog post. “She [also] have transparent reasoning and use a multi-phase process to consider every source more depth […] By turning it [AI]S Together on scale, scientists can greatly speed up the pace of scientific discovery. “

Futurehouse still has to achieve a scientific breakthrough or make a new discovery with its AI tools.

Part of the challenge when developing an “AI scientist” is anticipating an unprecedented number of confusing factors. AI can come in handy in areas where broad exploration is needed, such as reducing a huge list of possibilities, but it is less clear whether it can do the type of out-of-the-box problem solving that leads to bona fide breakthrough.

Results of AI systems designed for science have so far been largely underwhelming. In 2023, Google said that around 40 new materials had been synthesized With the help of one of his AIS, called Gnome. But an external analysis It was not even found that none of those materials was new.

The technical shortcomings and risks of AI, such as the tendency to hallucinate, are also wary of scientists to endorse it for serious work. Even well -designed studies can ultimately be affected by misbehaving AI, which is struggling with performing very accurate activities.

Futurehouse indeed acknowledges that his AI tools – especially Phoenix – can make mistakes.

“We are free [this] Now in the spirit of fast iteration, “the company said in his blog post.” Please give feedback while you use it. “