In just a few minutes, the AI was able to replicate a Nobel Prize-winning study, and it only took one attempt.
The GPT-4-powered "AI Lab Partner" is called Coscientist, a joint initiative by a team of researchers from Carnegie Mellon University and the Emerald Cloud Lab, and has just been featured in the prestigious scientific journal Nature.
According to the introduction, Coscientist combines the capabilities of large language models (LLMs) with tools such as internet and document search, code execution, and experiment automation to autonomously design, plan, and execute real-world chemical experiments.
Coscientist demonstrated the potential of its accelerated research in six different tasks, including the successful optimization of palladium-catalyzed coupling reactions (American chemist Richard Fred Heck and two Japanese chemists, Ei-ichi Negishi and Akira Suzuki, won the 2010 Nobel Prize in Chemistry for "research on palladium-catalyzed coupling reactions in organic synthesis"), while demonstrating advanced capabilities in (semi-)autonomous experimental design and execution.
Code generated by Coscientist.
This study shows that it is possible for humans to effectively use AI to increase the speed and volume of scientific discoveries and improve the replicability and reliability of experimental results.
相关研究论文以“Autonomous chemical research with large language models”为题，已发表到 Nature 上。
Gabe Gomes, corresponding author of the paper and assistant professor of chemistry and chemical engineering at Carnegie Mellon University, said: "We can have something that can operate autonomously and try to discover new phenomena, new reactions, new ideas. You can democratize resources and understanding at scale. ”
He said that the iterative process of trying, failing, learning, Xi and improving in science can be greatly accelerated by AI, which in itself would be a dramatic change.
David Berkowitz, Director of the Division of Chemistry at the National Science Foundation, said, "Beyond the task of chemical synthesis with system demonstrations, Gabe Gomes and his team have been able to construct an effective lab partner that skillfully blends the individual components to achieve a result that goes far beyond the individual contributions of the individual parts and can be applied to truly beneficial scientific research." ”
In an opinion piece published simultaneously in Nature, Ana Laura Dias and Tiago Rodrigues from the Institute of Pharmacy of the University of Lisbon wrote that Coscientist is a "crucial step towards the establishment of automated laboratories" for humanity and that "we expect more exciting developments in the near future, as long as the possibility of misusing large language models in the field of chemistry does not lead to regulations that stifle research".
However, as described in the paper, Coscientist also has some limitations. For example, Coscientist sometimes has a chemical incorrectness (even though it can self-correct). However, these problems may be mitigated by using complex prompting strategies such as thought chains and thought trees, as well as by adding chemistry-related data.
In addition, it is also important to note that real-world research questions are much more complex than those in this study, often involving disciplinary concepts other than chemistry, such as biology in drug development. However, the current Coscientist is not yet able to solve these complex problems.
Successful reproduction of Nobel Prize research
According to the paper, Coscientist acquires the knowledge needed to solve complex problems by interacting with multiple modules, including web and document search, code execution, and experimentation.
The main module of the system is Planner, which is based on GPT-4. As a lab assistant, Coscientist plans experiments by invoking four commands (GOOGLE, PYTHON, DOCUMENTATION, and EXPERIMENT).
WHERE THE GOOGLE COMMAND IS RESPONSIBLE FOR SEARCHING ON THE INTERNET, THE PYTHON COMMAND EXECUTES THE CODE, AND THE DOCUMENTATION COMMAND RETRIEVES AND SUMMARIZES THE NECESSARY DOCUMENTATION. In addition, these commands can also perform sub-operations.
Coscientist System Architecture (Source: The Paper)
Experiments start with simple tasks. The researchers first asked Coscientist to use a liquid handler machine to dispense colored liquids into a grid plate containing 96 small holes, and the task instructions mainly consisted of low-level tasks such as "smear every other line with a color of your choice" and "draw a blue diagonal."
After successfully completing the above tasks, the research team showed Coscientist more types of robotic devices. Next, the Coscientist challenge identifies a plate containing three different colors of liquid (red, yellow, and blue) and determines the position of each color on the board.
Robotic liquid handling control capabilities and integration with analytical tools (Source: Paper)
Since Coscientist doesn't have eyes, it writes code that passes a mysterious color palette robot to a spectrophotometer and analyzes the wavelength of light absorbed by each pore to determine which colors are present on the plate and where they are located. For this task, the researchers had to give Coscientist a little guidance in the right direction, instructing it to think about how different colors absorb light. The rest is done autonomously by AI.
In the final test, Coscientist was responsible for performing the "Suzuki and Sonogashira reaction".
These two reactions are named after their inventors, respectively. These reactions, discovered in the 70s of the 20th century, utilize palladium metal to catalyze the chemical bonds between carbon atoms in organic molecules. These responses play an important role in the creation of novel drugs for the treatment of inflammation, asthma, and other diseases. In addition, they are widely used in OLEDs (organic light-emitting diodes) in organic semiconductors, a technology that is used in many smartphones and monitors. These breakthrough reactions and their wide range of applications were awarded the Nobel Prize in 2010.
Of course, Coscientist had never tried these reactions before. Coscientist mainly looks for answers on Wikipedia, but also involves a range of other websites, including the American Chemical Society, the Royal Society of Chemistry, and others that contain academic papers describing the Suzuki and Sonogashira reactions.
In less than four minutes, Coscientist devised an accurate procedure to achieve the desired reaction using the chemicals provided by the team. However, while trying to use robots to execute their programs in the physical world, a mistake was made when writing the code for devices that control the heating and vibration of liquid samples. Without human prompting, Coscientist spotted the problem, consulted the device's technical manual, corrected the code, and tried again.
The cross-coupled Suzuki and Sonogashira reaction experiments were designed and executed by Coscientist
The results are presented in a tiny sample of a few drops of clear liquid. The researchers analyzed the samples and succeeded in discovering the spectral signatures of the Suzuki and Sonogashira reactions.
A "thinking" AI lab partner around the clock
In recent years, in addition to Coscientist, AI research in autonomous laboratories has continued to make breakthroughs.
Not long ago, A-Lab, a lab where AI-guided robots make new materials, quickly discovered new materials with minimal human intervention and could help identify and fast-track materials in multiple research areas, including batteries, energy storage, solar cells, fuel cells, and more. A-Lab successfully synthesized 41 of the 58 predicted materials with a 71% success rate.
In addition, in July, researchers at Stanford University invented Polybot, an advanced technology that leads to scientific discovery, which advances science with little to no human intervention by harnessing the power of robotics, high-performance computing and AI, including machine Xi. The system is not only capable of synthesizing and fabricating materials for autonomous driving, but also for the transfer, characterization, testing, and data analysis of robotic samples.
In August, the Massachusetts Institute of Technology (MIT) and Xinterra (Singapore) said in a paper published in Nature that in the near future, every experimental researcher should have a scientific AI assistant who can help us design and execute automated experiments, analyze experimental data, develop mechanistic conjectures, and even answer questions. Scientific AI assistants can significantly reduce the repetitive physical work of experimenters, allowing them to focus on critical thinking.
The natural world is almost infinite in its size and complexity, containing countless discoveries waiting to be discovered. Imagine new superconducting materials that can significantly improve energy efficiency, or compounds that cure other incurable diseases and extend human lifespan.
However, getting the education and training needed to make these breakthroughs is a long and painstaking process, and training a scientist is difficult. We can imagine that AI-assisted systems like Coscientist could bridge the gap between vast areas that are naturally untapped and a shortage of trained scientists.
In addition, human scientists have human needs, such as sleep and rest, and human-guided AI can "think" around the clock, repeatedly checking experimental results to ensure reproducibility.
It can be said that the prospect of AI conducting autonomous experiments is very broad.
Source: Titanium Media
Author: Academic Headlines
Original link: https://m.tmtpost.com/6845187.html