**AI-Descartes, a new AI scientist, successfully reproduced a Nobel Prize-winning work using logical reasoning and symbolic regression to find exact equations. The system is effective against real-world data and small data sets, with the future goal being the automation of building scientific theories**

AI-Descartes, a scientist-like artificial intelligence developed by researchers at IBM Research, Samsung AI and the University of Maryland, Baltimore County, reproduced key parts of the Nobel Prize winners' work, including Langmuir's equations of gas behavior and Kepler's third law of planetary motion.

Supported by the Defense Advanced Research Projects Agency (DARPA), the AI system uses symbolic regression to find equations that fit the data, and its most unique feature is its ability to think logically. This allows AI-Descartes to determine which equations best fit a scientific theory. The system is particularly effective with noisy data (data that has a lot of disturbances) in the real world and small datasets with too few samples. explain. The team is working on creating new data sets and training computers that can read scientific articles and build background theories to refine and expand the system's capabilities.

The system demonstrated its capabilities on Kepler's third law of planetary motion, Einstein's law of relativistic time dilation, and Langmuir's gas adsorption equation.

****************

In 1918, the American chemist Irving Langmuir published an article examining the behavior of gas molecules bound to a solid surface. Guided by the results of careful experiments, as well as his theory that solids offer discrete sites for gas molecules to fill, he worked out a series of equations that describe how much gas will bind, given the pressure.

Now, nearly a century later, an "AI scientist" developed by researchers at IBM Research Samsung IA and the University of Maryland, Baltimore County (UMBC) has recreated a key part of Langmuir's Nobel Prize-winning work. The system - artificial intelligence (AI) functioning as a scientist - also rediscovered Kepler's third law of planetary motion, which can calculate the time it takes for one space object to orbit another space object given the distance separating them, and created a good approximation of Einstein's law of relativistic time dilation, which shows that time slows down for fast moving objects.

The research was supported by the Defense Advanced Research Projects Agency (DARPA). An article describing the results was published (April 12) in the journal Nature Communications.

**A machine learning tool that reasons**

The new artificial intelligence scientist - dubbed "AI-Descartes" by researchers - joins AI Feynman and other recently developed computing tools aimed at accelerating scientific discovery. At the core of these systems is a concept called symbolic regression, which finds equations to fit the data. Given basic arithmetic operations, such as addition, multiplication, and division, the systems can generate hundreds to millions of possible equations, searching for those that most accurately describe the relationships in the data.

"AI-Descartes offers several advantages over other systems, but its most unique feature is its ability to make sense," says Christina Cornelio, a research scientist at Samsung AI in Cambridge, England, who is the paper's lead author. If there are several candidate equations that fit the data well, the system identifies which equations best fit the underlying scientific theories as well. The ability to think logically also differentiates the system from "generative artificial intelligence" programs like ChatGPT, whose large language model has limited logical skills that can sometimes disrupt basic math calculations.

"In our work, we merge a first-principles approach, which scientists have used for centuries to derive new formulas from existing background theories, with a data-driven approach that is more common in the era of machine learning," says Corneliu. "This combination allows us to take advantage of both approaches and create more accurate and meaningful models for a wide variety of applications."

The name AI-Descartes is a nod to the 17th-century mathematician and philosopher René Descartes, who argued that the natural world can be described by a few basic physical laws and that logical deduction plays a key role in scientific discovery.

**Fits real-world data**

According to the researchers, the system works particularly well on real-world noisy data (data that is sensitive to perturbations in the sampling process, for example), which can trip up traditional symbolic regression programs that might ignore the real signal in an effort to find formulas that capture every stray zig and zag in the data. It also handles small data sets well, and even finds reliable equations when fed with no more than ten data points.

One factor that may slow the adoption of tools like AI-Descartes for real-world scientific discoveries is the need to identify and codify the background theories associated with open scientific questions. The team is working on creating new datasets containing both real measurement data and accompanying background theory to refine their system and test it in a new field. They also want to eventually train computers to read scientific papers and build the background theory themselves.

"In this work, we needed human experts to write, in formal, computer-readable terms, what the axioms of the background theory are, and if the human missed something or got one of them wrong - the system wouldn't work," says one of the study's authors, Tyler Josephson, professor of chemical engineering , Biochemical and Environmental at UMBC. "In the future," he says, "we would like to automate this part of the work as well, so that we can explore many more fields of science and engineering." This goal drives Josephson's research into artificial intelligence tools to advance chemical engineering.

Ultimately, the team hopes that their artificial intelligence, like the real person, might inspire a new and productive approach to science. "One of the most exciting aspects of our work is the potential for significant progress in scientific research," says Corneliu.

**More of the topic in Hayadan:**

## 4 תגובות

This article is here to help AI

The lie spread by mathematics,

Regarding the existence of a single ratio number suitable for all circles

and its value is close to 3.14

A circle is a literary name, which comes instead of the geometric name - a closed circular line.

The simple and understandable form of a closed circular line is round, but the sophisticated form of a closed circular line is the curvature of a closed circular line.

There are an infinite number of closed circular lines, from zero length to infinite length, and each line of these infinite lines has a curved shape unique only to it.

The shorter a closed circular line, the greater its shape of curvature.

The longer a closed circular line is, the smaller its shape of curvature.

To express in a precise quantitative way a form of curvature, we will use the ratio number

The length of a closed (partial) circular line is the length of the straight diameter line

There are an infinite number of such ratios, and they are in a narrow range, between 3.14 and 3.16

Mathematics spread the lie of a single ratio number corresponding to all closed circular lines, and its value is close to 3.14

This lie has been around for 2000 years, and it's time for those responsible for math studies to get rid of it.

The extension appears in the book

Asbar's magic journey on the wings of natural knowledge

A. Asbar

I have a task for artificial intelligence - to discover the concept of variable pie

A. Asbar

With the help of AI they will recreate what exists. How will they continue to renew?? The human mind will not be replaced so quickly.

In Hebrew 'Descartes', after the famous philosopher ('I think means I exist').