Google DeepMind AI Reveals Potential for Thousands of New Materials

16 August 2022, North Rhine-Westphalia, Cologne: The lettering and logo of Google pictured on a glass pane in the press center of Koelnmesse. (dpa)
16 August 2022, North Rhine-Westphalia, Cologne: The lettering and logo of Google pictured on a glass pane in the press center of Koelnmesse. (dpa)
TT

Google DeepMind AI Reveals Potential for Thousands of New Materials

16 August 2022, North Rhine-Westphalia, Cologne: The lettering and logo of Google pictured on a glass pane in the press center of Koelnmesse. (dpa)
16 August 2022, North Rhine-Westphalia, Cologne: The lettering and logo of Google pictured on a glass pane in the press center of Koelnmesse. (dpa)

Google DeepMind has used artificial intelligence (AI) to predict the structure of more than 2 million new materials, a breakthrough it said could soon be used to improve real-world technologies.

In a research paper published in science journal Nature on Wednesday, the Alphabet-owned AI firm said almost 400,000 of its hypothetical material designs could soon be produced in lab conditions.

Potential applications for the research include the production of better-performing batteries, solar panels and computer chips.

The discovery and synthesis of new materials can be a costly and time-consuming process. For example, it took around two decades of research before lithium-ion batteries – today used to power everything from phones and laptops to electric vehicles – were made commercially available.

“We're hoping that big improvements in experimentation, autonomous synthesis, and machine learning models will significantly shorten that 10 to 20-year timeline to something that's much more manageable,” said Ekin Dogus Cubuk, a research scientist at DeepMind.

DeepMind’s AI was trained on data from the Materials Project, an international research group founded at the Lawrence Berkeley National Laboratory in 2011, made up of existing research of around 50,000 already-known materials.

The company said it would now share its data with the research community, in the hopes of accelerating further breakthroughs in material discovery.

"Industry tends to be a little risk-averse when it comes to cost increases, and new materials typically take a bit of time before they become cost-effective," said Kristin Persson, director of the Materials Project.

"If we can shrink that even a bit more, it would be considered a real breakthrough."

Having used AI to predict the stability of these new materials, DeepMind said it would now turn its focus to predicting how easily they can be synthesized in the lab.



SDAIA, KAUST Launch MiniGPT-Med Model to Help Doctors Diagnose Medical Radiology through AI

SDAIA, KAUST Launch MiniGPT-Med Model to Help Doctors Diagnose Medical Radiology through AI
TT

SDAIA, KAUST Launch MiniGPT-Med Model to Help Doctors Diagnose Medical Radiology through AI

SDAIA, KAUST Launch MiniGPT-Med Model to Help Doctors Diagnose Medical Radiology through AI

The Center of Excellence for Data Science and Artificial Intelligence at the Saudi Data and Artificial Intelligence Authority (SDAIA) and King Abdullah University of Science and Technology (KAUST) have introduced the MiniGPT-Med model.

The large multi-modal language model is designed to help doctors quickly and accurately diagnose medical radiology using artificial intelligence techniques.

Dr. Ahmed Alsinan, the Artificial Intelligence Advisor at the National Center for Artificial Intelligence and head of the scientific team at SDAIA, explained that the MiniGPT-Med model is capable of performing various tasks such as generating medical reports, answering medical visual questions, describing diseases, locating diseases, identifying diseases, and documenting medical descriptions based on entered medical images.

The model was trained on different medical images, including X-rays, CT scans, and MRIs.

The MiniGPT-Med model, derived from large-scale language models, is specifically tailored for medical applications and demonstrates significant versatility across different imaging methods, including X-rays, CT scans, and MRI. This enhances its utility in medical diagnosis.

Dr. Alsinan highlighted that the MiniGPT-Med model was developed collaboratively by artificial intelligence specialists from SDAIA and KAUST.

The model exhibits advanced performance in generating medical reports, achieving 19% higher efficiency than previous models. It serves as a general interface for radiology diagnosis, enhancing diagnostic efficiency across various medical imaging applications.