Scientific research in fields such as chemistry, biology and artificial intelligence has long been based on human experts to explore knowledge, generate ideas, design experiences and refine results. However, as problems become more complex and highly intensity of data, discovery slows down. While AI tools, such as language models and robotics, can manage specific tasks, such as literature research or code analysis, they rarely include the entire research cycle. Filling the gap between the generation of ideas and experimental validation remains a key challenge. For AI to advance independently, it must propose hypotheses, design and execute experiences, analyze the results and refine approaches in an iterative loop. Without this integration, AI risks producing disconnected ideas that depend on human supervision for validation.
Before the introduction of a unified system, the researchers relied on separate tools for each stage of the process. Great languages could help find relevant scientific articles, but they have not directly fueled the design of experiences or the analysis of the results. Robotics can help automate physical experiences and coding of libraries like Pytorch can help create models; However, these tools operate independently of each other. There was no unique system capable of managing the entire process, training ideas to verify them through experiences. This led to bottlenecks, where the researchers had to link the points manually, slowing down progress and leaving room for errors or missed opportunities. The need for an integrated system that could manage the entire research cycle has become clear.
Researchers from the Novelseek team from the Artificial Shanghai Intelligence Laboratory have developed NovelistAn AI system designed to execute the entire scientific discovery process independently. Novelseek includes four main modules that work in tandem: a system that generates and refines research ideas, a feedback loop where human experts can interact with and refine these ideas, a method to translate ideas into code and experimentation plans, and a process of carrying out several cycles of experiences. What makes novelseek stands out is its versatility; It operates in 12 scientific research tasks, in particular by predicting chemical reaction yields, including molecular dynamics, providing for chronological series data and handling functions such as 2D Semantic segmentation and classification of 3D objects. The team designed Novelseek to minimize human involvement, accelerate discoveries and provide consistent and high quality results.
The system behind Novelseek involves several specialized agents, each has focused on a specific part of the research workflow. The “survey agent” helps the system understand the problem by seeking scientific articles and identifying relevant information according to keywords and task definitions. He adapts his research strategy by first studying articles, then going deeper by analyzing full text documents for detailed information. This ensures that the system captures both general trends and specific technical knowledge. The “Review Code” agent examines the existing code bases, whether downloaded or from public standards like Github, to understand how current methods work and identify the areas to be improved. He verifies how the code is structured, search for errors and creates summaries that help the system to rely on previous work. The “Ideas Innovation Agent” generates creative research ideas, pushing the system to explore different approaches and refine them by comparing them to related studies and previous results. The system even includes a “planning and execution agent” which transforms ideas into detailed experiences, manages errors during the test process and ensures smooth execution of search plans in several steps.
Novelseek has given impressive results on various tasks. In the prediction of yield in chemical reaction, novelseek improved the performance of a basic line of 24.2% (with a variation of ± 4.2) to 34.8% (with a much smaller variation of ± 1.1) in just 12 hours, the progress that human researchers generally need months to reach. In the prediction of the activity of the amplifier, a key task in biology, novelseek increased the pearson correlation coefficient from 0.65 to 0.79 in 4 hours. For 2D semantic segmentation, a task used in computer vision, precision increased from 78.8% to 81.0% in just 30 hours. These performance increases, carried out in a generally necessary fraction of time, highlight the effectiveness of the system. Novelseek has also successfully managed large complex code bases with several files, demonstrating its ability to manage research tasks at the project level, not only in small isolated tests. The team has created the open code, allowing others to use, test and contribute to its improvement.
Several key dishes of research on novelseek include:
- Novelseek supports 12 research tasks, in particular the prediction of the chemical reaction, molecular dynamics and classification of 3D objects.
- The accuracy of the reaction performance forecast increased from 24.2% to 34.8% in 12 hours.
- The prediction performance of the activity of the activity increased from 0.65 to 0.79 in 4 hours.
- The 2D semantic segmentation precision increased from 78.8% to 81.0% in 30 hours.
- Novelseek includes agents for the search for literature, code analysis, generation of ideas and execution of experiences.
- The system is open-source, allowing reproducibility and collaboration between scientific fields.
In conclusion, Novelseek shows how the combination of AI tools in a single system can speed up scientific discovery and reduce its dependence on human effort. It connects the key steps, generating ideas, transforming them into methods and tests them through experiences, into a single rationalized process. What has taken months or years that has taken months or years can now be done in days or even in hours. By connecting each step of research on a continuous loop, novelseek helps teams pass difficult ideas to real world results more quickly. This system highlights the power of AI not only to help, but also to stimulate scientific research in a way that could reshape the way the discoveries are made in many areas.
Discover the Paper And GitHub page . All the merit of this research goes to researchers in this project. Also, don't hesitate to follow us Twitter And don't forget to join our 95K + ML Subdreddit and subscribe to Our newsletter.
Nikhil is an intern consultant at Marktechpost. It pursues a double degree integrated into materials at the Indian Kharagpur Institute of Technology. Nikhil is an IA / ML enthusiast who is still looking for applications in fields like biomaterials and biomedical sciences. With a strong experience in material science, he explores new progress and creates opportunities to contribute.
