Need a research hypothesis? Ask the AI

MIT researchers have created a way to autonomously generate and evaluate promising research hypotheses across fields through human-AI collaboration. In a new paper, they describe how they used this framework to create evidence-based hypotheses that match unmet research needs in the field of biologically inspired materials.

Published today in Advanced materialsThe study was co-authored by Alireza Ghafarollahi, a postdoctoral fellow at the Laboratory of Atomic and Molecular Mechanics (LAMM), and Markus Buehler, the Jerry McAfee Professor of Engineering in MIT’s Departments of Civil and Environmental Engineering and Mechanical Engineering and director of LAMM.

The framework, which the researchers call SciAgents, consists of multiple AI agents, each with specific capabilities and access to data, that leverage “graph reasoning” methods, in which the AI ​​models use a knowledge graph that organizes and defines the relationships between various scientific concepts. The agent-based approach mimics the way biological systems organize themselves into groups of elementary building blocks. Buehler notes that this “divide and conquer” principle is an important paradigm in biology on many levels, from materials to insect swarms to civilizations – all examples where total intelligence is far greater than the sum of abilities individuals.

“Using multiple AI agents, we try to simulate the process by which communities of scientists make discoveries,” says Buehler. “At MIT, we do this by bringing together a group of people from different backgrounds who work together and bump into each other in coffee shops or in MIT’s Infinite Corridor. But it’s very serendipitous and slow. Our quest is to simulate the discovery process by exploring whether AI systems can be creative and make discoveries.

Automate good ideas

As recent developments have demonstrated, large language models (LLMs) have shown an impressive ability to answer questions, summarize information, and perform simple tasks. But they are quite limited when it comes to generating new ideas from scratch. MIT researchers wanted to design a system that allows AI models to perform a more sophisticated, multi-step process that goes beyond recalling information learned during training, to extrapolate and create new knowledge.

The foundation of their approach is an ontological knowledge graph, which organizes and establishes connections between various scientific concepts. To create the graphs, researchers feed a set of scientific articles into a generative AI model. In previous work, Buehler used a mathematical field known as category theory to help the AI ​​model develop abstractions of scientific concepts in the form of graphs, rooted in defining relationships between components, a way that could be analyzed by other models through a process called graphical reasoning. . This focuses AI models on developing a more principled way of understanding concepts; it also allows them to generalize better across domains.

“It’s really important for us to create science-driven AI models because scientific theories are typically rooted in generalizable principles rather than simple knowledge recall,” Buehler says. “By focusing AI models on ‘thinking’ in this way, we can go beyond conventional methods and explore more creative uses of AI.”

For the most recent paper, the researchers used about 1,000 scientific studies on biological materials, but Buehler says the knowledge graphs could be generated using many more or fewer research papers in any field .

Once the graph was established, the researchers developed an AI system for scientific discovery, with several specialized models to play specific roles in the system. Most of the components were built from OpenAI’s ChatGPT-4 series models and used a technique known as in-context learning, in which prompts provide contextual information about the model’s role in the system while allowing it to learn from the data provided.

The individual agents in the framework interact with each other to collectively solve a complex problem that none of them would be able to solve alone. The first task given to them is to generate the research hypothesis. LLM interactions begin after a subgraph has been defined from the knowledge graph, which can happen randomly or by manually entering a pair of keywords discussed in the articles.

In this context, a language model that the researchers named “Ontologue” is responsible for defining the scientific terms in the articles and examining the links between them, thus expanding the knowledge graph. A model named “Scientist 1” then develops a research proposal based on factors such as its ability to discover unexpected properties and novelties. The proposal includes a discussion of potential outcomes, the impact of the research, and a hypothesis about the underlying mechanisms of action. A “Scientist 2” model expands on the idea, suggests specific experimental and simulation approaches, and makes other improvements. Finally, a “Critical” model highlights its strengths and weaknesses and suggests further improvements.

“It’s about building a team of experts who don’t all think the same way,” Buehler says. “They have to think differently and have different abilities. The critical agent is deliberately programmed to criticize others, so not everyone agrees and says it’s a great idea. You have an agent which says: “There is a weakness here, can you do it better?”

Other agents in the system are capable of searching existing literature, which provides the system with a means of not only assessing feasibility, but also creating and evaluating the novelty of each idea.

Make the system stronger

To validate their approach, Buehler and Ghafarollahi built a knowledge graph based on the words “silk” and “energy-hungry.” Using this framework, the “Scientist 1” model proposed integrating silk with dandelion-based pigments to create biomaterials with improved optical and mechanical properties. The model predicted that the material would be significantly stronger than traditional silk materials and require less energy to process.

Scientist 2 then made suggestions, such as using specific molecular dynamics simulation tools to explore how the proposed materials would interact, adding that a good application for the material would be a bio-inspired adhesive. The Critic model then highlighted several strengths of the proposed material and areas for improvement, such as its scalability, long-term stability, and environmental impacts of solvent use. To address these concerns, the reviewer suggested conducting pilot studies to validate processes and performing rigorous analyzes of material durability.

The researchers also conducted further experiments with randomly chosen keywords, which produced various original hypotheses on more efficient biomimetic microfluidic chips, improving the mechanical properties of collagen-based scaffolds and on the interaction between graphene and amyloid fibrils to create bioelectronic devices.

The system was able to come up with these rigorous new ideas based on the knowledge graph path,” explains Ghafarollahi. “In terms of novelty and applicability, the materials seemed robust and new. In future work, we will generate thousands, if not tens of thousands, of new research ideas, and then we can categorize them, try to better understand how these materials are generated and how they could be improved further. »

In the future, researchers hope to integrate new tools for retrieving information and running simulations into their frameworks. They can also easily replace the base models of their frameworks with more advanced models, allowing the system to adapt to the latest innovations in AI.

“Because of the way these agents interact, even a slight improvement to a model has a huge impact on the overall behaviors and outcomes of the system,” says Buehler.

Since publishing a preprint with open source details about their approach, the researchers have been contacted by hundreds of people interested in using these frameworks in various scientific fields and even in areas such as finance and cybersecurity .

“There are a lot of things you can do without having to go to the lab,” Buehler says. “Basically, you want to go to the lab at the very end of the process. The lab is expensive and time consuming. So you want a system that can pursue the best ideas, make the best hypotheses, and predict accurately emerging ideas.