A key debate in the world of artificial intelligence is whether large language models should be proprietary—-controlled by their builders—or open source, with contributions from the broader community. Some industry leaders, including OpenAI,
Microsoft
and Google, think it makes more sense to keep models under tighter control. But others, including
IBM
and
Meta,
think an open approach makes more sense.
IBM and Meta have teamed up with a group of more than 50 other companies, research groups and academic institutions to launch the AI Alliance, “a community” focused on advancing “open, safe, responsible AI.” Other large companies in the group include
Intel,
Advanced Micro Devices,
Oracle,
ServiceNow,
Sony,
SoftBank
and
Dell Technologie
s. The group also includes start-ups such as HuggingFace, a marketplace for AI models; Stability AI, which builds open source models; and the AI chip maker Cerebras. Academic members include Yale, UC Berkeley, Cornell, Dartmouth, the University of Illinois and the University of Tokyo, among others. NASA, the National Science Foundation and CERN are members, too.
In announcing the AI Alliance, the group asserted that “open and transparent innovation is essential to empower a broad spectrum of AI researchers, builders and adopters with the information and tools needed to harness these achievements in ways that prioritize safety, diversity and economic opportunity and benefits to all.”
The AI Alliance said it plans to develop “benchmarks and evaluation standards, tools and other resources” to help develop AI systems. Among other things, the group says it wants to help create multilingual models that can “help address society-wide challenges in climate, education and beyond.”
There’s logic in having IBM and Meta as leaders of the group. Meta has made its Llama 2 large language model open source, free to use for both research and commercial purposes. IBM has a long history in open source software; the company owns Red Hat, which provides services for Linux system users.
Dario Gil, director research at IBM, notes that the initial group of AI Alliance members together invest more than $80 billion a year in R&D, with more than 1 million combined staff members. “What’s at stake is nothing less than the future of AI,” he says. “We have tapped into something that is viral and organic…the cumulative brain power around this is unrecognized and underreported.”
Gil was careful to say that he isn’t opposed to the creation of proprietary software, and he thinks there will be a place for both approaches. But as with Linux, he says, IBM sees power in an open source model. “A health AI industry must have a vibrant, open ecosystem.”
Meta likewise has been an early proponent of the open source approach to AI. “We believe it’s better when AI is developed openly—more people can access the benefits, build innovative products and work on safety,” Nick Clegg, Meta’s president for global affairs, said in a statement. “The AI Alliance brings together researchers, developers and companies to share tools and knowledge that can help us all make progress whether models are shared openly or not.”
The open versus close debate around AI has echoes to the same question that evolved decades ago around computer operating systems, with Linux as an open source operating system alternative to proprietary options like Windows and Unix. IBM is playing a similar role here as it did in the formation of the Linux Foundation, a group of Linux developers, more than two decades ago.
Gil spoke to Barron’s from the IBM Quantum Summit, the company’s annual event focused on quantum computing. As Gil has noted to Barron’s in the past, IBM sees the future of computing as including three elements—high precision classical computing; neural-based processing and AI; and qubit-based quantum computing. “Don’t think of them independently,” he says. The future of computing will involve the convergence of all three.
“Some people say this is another step in Moore’s Law,” Gil told Barron’s last year for a story about quantum computing, “but it is more fundamental than that. It is not that often that we get to redefine the nature of information.”
Write to Eric J. Savitz at [email protected]
Read the full article here
