Future House, a nonprofit organization backed by Eric Schmidt, aimed at building “AI scientists” within the next decade, has launched a platform and API with AI-powered tools designed to support scientific research.
Many startups compete to develop AI research tools in the scientific field. The high-tech giants also seem bullish with AI for science. Earlier this year, Google announced its “AI Co-Scientists.” The company said it could help scientists develop hypotheses and experimental research plans.
AI company Openai and the CEOs of humanity claim that AI tools can dramatically accelerate scientific discovery, especially in medicine. However, many researchers do not believe that today’s AI is particularly useful in guiding scientific processes due to its unreliability.
Thursday’s Future House released four AI tools: Crow, Falcon, Owl and Phoenix. Crows can search the scientific literature and answer questions about it. Falcon can perform deeper literature searches, including scientific databases. The owl looks for previous work in a particular subject area. Phoenix also uses tools to help plan chemical experiments.
“Unlike other (AIS), Futurehouse has access to a vast corpus of high-quality open-access papers and specialized scientific tools,” Future House wrote in a blog post. “They have (and) transparent inference and use multi-step processes to consider each source more deeply (…) by linking these (AIs) at scale. Scientists can dramatically accelerate the pace of scientific discovery.”
But, of course, Futurehouse has yet to achieve a scientific breakthrough or have not made any novel discoveries with its AI tools.
Part of the challenges in developing “AI Scientists” is to anticipate an invaluable number of confounding factors. AI may be useful in areas where extensive exploration is required, such as narrowing down a vast list of possibilities. However, it is not clear whether AI will be able to provide the kind of ready-to-use problem-solving that leads to its original breakthrough.
TechCrunch Events
Berkeley, California
|
June 5th
Book now
So far, the results of AI systems designed for science have been almost overwhelming. In 2023, Google said that around 40 new materials were synthesized with the help of one AI called Gnome. However, external analysis revealed that in fact, one of the materials is not NET NEW.
The technical shortcomings of AI and the tendency to hallucinate are wary of scientists to support it for serious work. Even well-designed research can be contaminated by AI fraud. This is a struggle to perform high-precision tasks.
In fact, Futurehouse admits that its AI tools (Phoenix in particular) can make mistakes.
“We are releasing (this) in the spirit of (this) rapid iteration,” the company wrote in a blog post. “Please provide feedback when using it.”