Comments about the article in Nature: Exclusive: Start-up FutureHouse debuts powerful AI 'reasoning model' for science

Following is a discussion about this article in Nature Vol 593 20 May 2024, by Petr W. Hatfield e.a.
To study the full text select this link: https://www.nature.com/articles/d41586-025-01753-1 In the last paragraph I explain my own opinion.

Reflection

Introduction

As artificial intelligence (AI) tools shake up the scientific workflow, Sam Rodriques dreams of a more systemic transformation. His start-up company, FutureHouse in San Francisco, California, aims to build an 'AI scientist' that can command the entire research pipeline, from hypothesis generation to paper production.
To higher a Scientist which can do that for you, is a very valuable asset.
Today, his team took a step in that direction, releasing what it calls the first true 'reasoning model' specifically trained for scientific tasks.
To unravel 'reasoning' in science, specific chemistry, by human is not simple. Reasoning including the common parts in chemical formulas.
The model, called ether0, is a large language model (LLM) that's purpose-built for chemistry, which it learnt simply by taking a test of around 500,000 questions. Following instructions in plain English, ether0 can spit out formulae for drug-like molecules that satisfy a range of criteria.
Who has discovered these criteria? Are they available in text books?
Ether0, which is open source and publicly available from today, joins a host of other efforts aimed at automating the scientific process, including at Google and the Japanese company Sakana AI. But unlike previous specialist models, ether0 tracks its 'train of thought' in natural language, providing a window into AI's 'black box' and allowing it to answer questions that typically require complex reasoning. Although some general-purpose reasoning models, such as OpenAI o1, have shown improvement on standardized science tests, they have struggled to generate deep insights without targeted training.
Researchers have expressed a mixture of excitement and concern about FutureHouse's advance. “I think it's very cool what they pulled off," says Kevin Jablonka, a digital chemist at the University of Jena in Germany. When playing around with a preview version of ether0, Jablonka found that the model could draw meaningful inferences about chemical properties that it wasn't trained on. “That's impressive and [something] models before couldn't do," he says.

1. AI for science

The model is the latest release from FutureHouse, which was launched in 2023 as a non-profit organization backed by former Google chief executive Eric Schmidt with the mission of expediting the scientific process with AI. In the past year, the company has released an advanced scientific literature reviewer and a platform of AI agents — LLM-based tools designed for specific tasks. These agents draw from the scientific literature and deploy molecular design tools to analyse data and answer detailed questions on drug design. In May, the team announced that it had used these models to propose a new treatment for dry age-related macular degeneration, a major cause of blindness. “Agents are going to be really useful for finding all this stuff in literature that's just kind of staring at us in plain sight," says Andrew White, a chemical engineer at FutureHouse who is on sabbatical from the University of Rochester in New York. But like most LLMs, the agents are fundamentally limited by the amount of chemistry information available on the Internet. “There's right now little real-world impact of those models in labs," says Jablonka, who led a review published last month on the chemistry prowess of LLMs. To achieve something closer to true understanding, computer scientists have turned to 'reasoning models', such as the Chinese model DeepSeek-R1. These AIs are prompted to converse with themselves and show the work that led to their answers. Studies suggest that this internal dialogue seems to improve their accuracy on complex questions, leading Rodriques to suspect that they might be useful for generating new research ideas.
Previously, when leading reasoning models have taken on science problems, they've focused mainly on passing standardized exams with basic textbook understanding, Jablonka says. “There hasn't been one model that's reasoning in any useful way in chemistry so far." FutureHouse set out to change that.

2. Thinking time

FutureHouse researchers took a relatively small LLM from the French start-up firm Mistral AI, which is roughly 25 times smaller than DeepSeek-R1 — compact enough to run on a laptop. Rather than train the model on chemistry textbooks and papers, they found that they could let it learn from taking tests. To do this, White compiled lab-generated chemistry results from 45 scholarly papers, tracking properties such as molecular solubility and smell, and converted them into 577,790 verifiable questions. Can AI review the scientific literature — and figure out what it all means? The researchers taught the base model to 'think aloud' by asking it to read incorrect solutions and reasoning chains generated by DeepSeek-R1. Seven versions of the model each tried to solve a specific subset of the chemistry questions, receiving reinforcement rewards for correct answers. The researchers then merged the reasoning chains from those specialist models into one generalist model. After running through the question set once more, they were left with ether0. The team assessed ether0's performance using a further set of questions, some of which were unrelated to those in the training set. Almost across the board, ether0 outperformed frontier models such as OpenAI's GPT-4.1 and DeepSeek-R1. For some problem types, the model more than doubled the accuracy of its competitors. And it did so for a bargain: training a similar state-of-the-art non-reasoning model to achieve comparable accuracy on reaction predictions used 50 times more data. But because ether0 can generate solutions only in the form of molecular formulae and reactions, it's hard to cross-check its performance against other models and humans on independent benchmarks, Jablonka says. Nevertheless, Jablonka found that the model could correctly reason about molecular structures that it had not been trained on — for instance, by changing a molecule's formula to fit a particular nuclear magnetic resonance spectrum. “I didn't expect this," he says. A view of the equipment being navigated by articulated mechancial arms within an AI-run robotic laboratory at the Insilico Medicine research facility in Suzhou, China. Advances in AI could pave the way for robotic laboratories that completely automate parts of the scientific process. Credit: Qilai Shen/Bloomberg via Getty The biggest opportunity presented by these reasoning models, Rodriques says, is that “you get to see what they are thinking throughout the entire process". His team found that if it allowed the models to reason for longer, the responses became more accurate but less legible — mixing in different languages and inventing new words. The team decided to prioritize interpretability over accuracy by limiting the reasoning time. Subbarao Kambhampati, a computer scientist at Arizona State University in Tempe, objects to this approach. He and his colleagues have found that the improvement in accuracy that LLMs get from reasoning has little to do with whether their reasoning was correct. Presenting an unverified chain-of-thought in readable English gives the false impression that the model understands and checks each step, Kambhampati argues. “You're playing to the cognitive flaws of humans," he says. “That seems to me to be the wrong way to engender trust." Nevertheless, Kambhampati welcomes the effort to bring reasoning models to chemistry. “My feeling is that these models will generate very good scientific calculators — and that is nothing to be sneezed at," he says.

3. Fighting futures

Rodriques dreams well beyond calculators. He thinks that embedding reasoning abilities into specialized agents will enable his team to automate the scientific method end-to-end — and not just for chemistry. Although FutureHouse plans to focus on chemistry, which it says is better-suited to reasoning in natural language, the team anticipates that other groups will generalize the approach across the natural sciences. How AI-powered science search engines can speed up your research Rodriques predicts that within two years, most good scientific hypotheses “may be generated by us or systems like the ones we're building". In keeping with Schmidt's vision for AI science, Rodriques foresees advances in robotics that will automate bench work, leaving human scientists with the responsibility of “resource allocation" — choosing which research questions to pursue. Some push back against this future. “Is that what people who are training to be scientists today want to be?" asks Lisa Messeri, an anthropologist at Yale University in New Haven, Connecticut. Last year, she and M. J. Crockett, a psychologist at Princeton University in New Jersey, argued3 that an over-reliance on AI tools could create “scientific monocultures", in which certain questions and approaches dominate and the knowledge produced becomes less diverse and robust. Even Jablonka, who is also developing AI scientists for chemistry, cautions that “there's a lot of danger in just throwing everything into being driven by LLMs" run by big tech companies. Rodriques acknowledges these concerns, but says “pumping the brakes is not the right solution". “AI is going to dramatically accelerate the process of scientific discovery," he adds, “but like any other tool, it will be flawed."

1. Chalenges

Reflection 1

Reflection 2

If you want to give a comment you can use the following form Comment form

Created: 20 December 2024

Back to my home page Index
Back to Nature comments Nature Index