AI scientists combine theory and data to discover equations

AI News

In 1918, American chemist Irving Langmuir published a paper investigating the behavior of gas molecules adhering to solid surfaces. Based on careful experimentation and his theory that solids provide discrete places for gas molecules to pack, he developed a series of equations that describe how well gas adheres when pressure is applied. derived the formula for

Now, nearly 100 years later, an “AI scientist” developed by IBM Research, Samsung AI, and researchers at the University of Maryland Baltimore County (UMBC) has recreated an important part of Langmuir’s Nobel Prize-winning work. . This system, an artificial intelligence (AI) acting as a scientist, also rediscovered Kepler’s third law of planetary motion. This law calculated the time it takes for one cosmic object to orbit another, producing a good approximation of Einstein’s relativistic time. – Law of expansion. It shows that fast-moving objects slow down.

This research was supported by the Defense Advanced Research Projects Agency (DARPA). A paper describing the results will be published April 12 in Nature Communications.

machine learning tools that infer

The new AI scientist, dubbed “AI-Descartes” by researchers, joins AI Feynman and other recently developed computing tools aimed at speeding up scientific discovery. At the core of these systems is a concept called symbolic regression that finds an equation that fits the data. Given basic operators such as addition, multiplication, and division, the system can generate hundreds to millions of candidate equations to find the one that most accurately describes the relationships in your data.

Cartesian AI has several advantages over other systems, but its most distinguishing feature is its ability to reason logically, said lead author of the paper, a research scientist at Samsung AI in Cambridge, UK. Christina Cornelio said. When there are multiple candidate equations that fit the data well, the system identifies which equation best fits the underlying scientific theory. This reasoning ability is also what distinguishes the system from “generative AI” programs such as ChatGPT. ChatGPT’s large language model limits your logical skills and can even mess up basic math.

“In our work, we combine the first-principles approach that scientists have used for centuries to derive new formulas from existing background theories, and the more common data-driven approach in the machine learning era. We’re merging,” says Cornelio. “This combination allows us to leverage both approaches to create more accurate and meaningful models for a wide range of applications.”

AI – the name Descartes is in favor of 17thThe 19th-century mathematician and philosopher Rene Descartes argued that the natural world could be described by some fundamental laws of physics, and that logical deduction played an important role in scientific discovery.

Suitable for real-world data

This system works especially well with noisy real-world data. This can break traditional symbolic regression programs that can miss real signals in an attempt to find a formula that captures all the false zigzags and zigzags in the data. It also handles small data sets well and can find reliable equations given just 10 data points.

One of the factors slowing the adoption of tools for frontier science like AI Descartes is the need to identify and code background theories related to open scientific questions. The team is working on a new dataset containing both real-world measurement data and relevant background theory to refine and test the system on new terrain.

I also hope to eventually train a computer to read scientific papers and construct background theories on its own.

“This work required human experts to write down in formal, machine-readable terms what the axioms of the background theory were. The system just doesn’t work, says co-author Tyler Josephson, an assistant professor of chemistry, biochemistry and environmental engineering at UMBC: “In the future, we’ll automate this part of the work as well so that more science and engineering can do more.” We want to be able to explore the field,” he says.

This goal motivated Josephson’s research on AI tools to advance chemical engineering.

Ultimately, the team hopes AI-Descartes, like a real person, will inspire productive new approaches to science. “One of the most exciting aspects of our work is the potential to significantly advance scientific research,” he says Cornelio.

/Release. This material from the original organization/author may be of a point-in-time nature and has been edited for clarity, style, and length. and do not take a stand. All views, positions and conclusions expressed herein are solely those of the author.

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *