Many researchers have taken a comprehensive view of scientific progress over the last 50 years and come to the same disturbing conclusion: scientific productivity is declining. It takes more time, more money, and searching for large teams that came once faster and cheaper. Although a variety of explanations are given for recession, one is, as research becomes more complex and special, scientists should spend more time to review publications, design refined experiments and analyze data.
Now, philanthropically funded research lab is trying to accelerate scientific research with the AI platform designed to automate several important stages on the route towards scientific progress. The platform is made up of a series of special AI agents for tasks, including information recovery, information synthesis, chemical synthesis design and data analysis.
Futurehouse founders Sam Rodrick PhD ’19 and Andrew White believe that by providing each scientist access to their AI agents, they can break through the greatest hurdles in science and help solve some of the most pressure problems of humanity.
“Natural linguistics are the real language of science,” Rodric says. “Others are creating a foundation model for biology, where machine learning models speak DNA or protein language, and it is powerful. But discoveries are not represented in DNA or protein. The only way we know how to represent discoveries, hypothesis and cause, is with natural language.”
Finding big problems
For his PhD research at MIT, Rodrick demanded to understand the internal functioning of the brain in the laboratory of Professor Ed Boyden.
“The whole idea behind the Futurehouse was inspired by the belief that I had all the information that we had all the information during MIT that we needed to know how the brain works, we do not know it because someone does not have time to read all literature,” Rodrix says. “Even if they can read all this, they would not be able to collect it in a broader principle. It was a fundamental piece of future puzzle.”
Rodric wrote about the need for new types of big research cooperation as the last chapter of his PhD thesis in 2019, and although he spent some time in running a laboratory at Francis Crickery Institute in London after graduation, he found that he found herself gravity towards comprehensive problems in science that he could not take any single lab.
“I was interested in how to automate or score science and what new organizational structures or technologies unlock high scientific productivity,” says Rodric.
When the Chat-GPT 3.5 was released in November 2022, Rodrix saw a way towards the more powerful model that could cause scientific insight on its own. Around that time, he also met Andrew White, a computational chemist at the University of Rochester, who was early accessible to reach the chat-GPT 4. White had created the first major language agent for science, and researchers joined the forces to start a futuristichouse.
The founders began creating separate AI equipment for tasks such as literature discoveries, data analysis and hypothesis generation. They began with data collection, eventually releasing Pepperu in September 2024, which recommends the best AI agent in the world to reconstruct information in rodiacal scientific literature and briefly. At the same time, he released, someone has, a tool that allows scientists to determine whether someone has done specific experiments or discovered specific hypotheses.
“We were just asking, ‘What kind of questions do we ask as scientists all the time?” Rhodrix miss.
When Futurehouse officially launched its platform on 1 May this year, it rebuilt some of its equipment. Paper QA is now a crow, and is anyone now called an owl. Falcon is an agent capable of compiling and reviewing more sources than Crows. Another new agent, Phoenix, the plan of researchers can use special equipment to help using chemistry experiments. And Finch is an agent designed to automate data -driven discovery in biology.
On 20 May, the company demonstrated a multi-agent scientific search workflow to automate the major stages of the scientific process and identified a new medical candidate for dry age-related macular degeneration (DAMD), which is a major cause of irreversible blindness worldwide. In June, Futurehouse released Ether 0, which was for the 24B Open-Weets Reasoning Model Chemistry.
“You really have to think of these agents as part of a large system,” Rodric. “Soon, literature search agents will be integrated with data analysis agents, hypothesis generation agents, an experiment plan agent, and they will all be engineers to work together.”
Agent for all
Today, one can use agents of Futurehouse on Fatchion.futurehouse.org. The company’s platform created enthusiasm in the industry, and stories about scientists using agents to accelerate research.
One of the scientists of the Futurehouse used agents to identify a gene that may be associated with polycystic ovary syndrome and can come up with a new treatment hypothesis for the disease. Another researcher from the Lawrence Berkeley National Laboratory used Crowe, who was able to make AI capable of discovering a pubmed research database for information related to Alzheimer’s disease.
Scientists from another research institute have used agents to arrange a systematic review of relevant genes for Parkinson’s disease, with agents of phythouse better than general agents.
Rodrick says that scientists who think of fewer agents like Google scholars and get out of the stage more like a smart assistant scientist.
Rodrick explains, “People who are finding speculation get more benefits from chat-GPT O3 deep research, while those who are actually looking for loyal literature reviews exit our agents,” Rodric.
Rodric also think that the futuristichouse will soon reach a point where its agents can use raw data from research papers to tested the ability to copy its results and verify the conclusions.
For a long time, Rodrick says that the Futurehouse is working on embedding its agents with taset knowledge, so that the Futurehouse is able to do more sophisticated analysis, while the agents are also giving the ability to use computational tools to detect hypotheses.
Rodrix says, “There has been a lot of progress around the foundation model around the science and language models for protein and DNA, that now we need to provide access to our agents and all other equipment that usually use to do science,” Rodrix says. “The construction of infrastructure is going to be important to allow agents to use more specific equipment for science.”
