Research labs in large companies are pushing the horizons of what artificial intelligence can do in areas such as image recognition, natural language processing and more. And some large firms are allocating more and more capital in the race to create these opportunities.
For example, meta-platforms Inc.
says that by mid-summer its AI Research SuperCluster system will host 16,000 Nvidia Corp.
a graphics processor, a massive engine that Meta claims will be the fastest of its kind in the world. It is also a DeepMind Technologies lab owned by the parent company Google Alphabet Inc.,
announced the development of a new language model, commenting: “The level of reading comprehension in high school is approaching human performance.”
Although these campaigns go differently, both have the potential to drive tomorrow’s success in discovering drugs, new materials, climate change tools, more thorough analysis of military drone personnel and more.
And they are hardly the only ones: Microsoft Corp.
International Business Machines Corp.
and others are also participating in the AI Marathon.
Consumers and investors who are more focused on stock market turmoil may not pay attention to projects not directly related to activities or quarterly results. But research and development often releases products that go beyond the lab’s original goals. For example, the Parc Lab of Xerox Holdings Corporation was a pioneer in the field of personal computers. Researchers from Bell Labs, at the time part of what is now AT&T Inc.,
developed a transistor and made a prototype of an early mobile phone.
The money spent is huge. Alphabet’s R&D spending rose to $ 31.562 billion last year from $ 27.573 billion in 2020; Last year, Meta spent $ 24.655 billion on research and development, up from $ 18.447 billion last year, company documents show. Alphabet said R&D spending in 2021 was 12.3% of revenue, and Meta said it was 21% of its sales.
In contrast, the S&P 500 spent an average of 2.82% of its revenue on research and development, according to Morningstar Inc., which calculated the weighted average revenue for the components of the index in each of the last fiscal years.
The parent of Facebook Meta ranks in the AI race dedicated to data scaling and computing power. Deep learning is a form of AI designed to mimic aspects of human neurons. It began in 2010 when the ImageNet project proved that it was possible to use two GPUs developed by Nvidia to teach the larger AI model to recognize labeled images, said Gyeongghong Cho, co-founder of the startup Prescient Design and an associate professor in computer science. and Data Science at New York University.
Meta has said its AI Research SuperCluster contains 6,080 Nvidia GPUs, and that number will grow to 16,000. said Shubho Sengupta, a software engineer involved in the project.
The supercomputer will help Meta researchers build models of artificial intelligence that can work in hundreds of languages, analyze text, images and video together, and develop augmented reality tools and rich multidimensional experiences in the metaworld, the company said. The technology will also help him more easily identify harmful content, Meta said.
London-based Alphabet’s DeepMind uses a technique called reinforced learning. With this approach, the algorithm is learned through trial and error.
Last month, DeepMind announced a new artificial intelligence system called AlphaCode, which competes with software developers. In December, DeepMind announced the development of a new language model that it said was reducing the scale without compromising performance. This complements DeepMind’s assertion that it has made a leap in understanding protein structures, which is crucial to drug discovery.
“This demonstrates that creative approaches to artificial intelligence can lead to radical performance improvements without the need to make more serious crunches of numbers,” said Schumann Gosemayumder, head of AI at F5. Inc.
and a Google veteran.
It is too early to predict what the commercial application of this study will be.
Dr. Cho of New York University, whose startup Prescient was sold to Genentech Inc., a subsidiary of Roche, says large corporate labs represent much of the research in AI.
Such work will not lead to corporate results in the short term. But in the longer run it may be.
“Such investments from leading technology companies, which are largely tied to AI technologies, will naturally encourage and compel other companies, both technological and non-technological, such as pharmaceuticals,” Dr. Cho said.
Write Stephen Rosenbush at email@example.com
Copyright © 2022 Dow Jones & Company, Inc. All rights reserved. 87990cbe856818d5eddac44c7b1cdeb8