An alliance led by IBM and Meta in collaboration with 45 other organizations is being launched today to promote adoption of open approaches to both the science and software being relied on to build and deploy artificial intelligence (AI) models.
Members of the AI Alliance include AMD, Anyscale, CERN, Cleveland Clinic, Cornell, Dell, EPFL, ETH, Hugging Face, Imperial College London, Intel, INSAIT, NASA, NSF, Oracle, Red Hat, Roadzen, RPI, ServiceNow, Sony Group, Stability AI, University of California Berkeley, University of Illinois, University of Tokyo and Yale University.
The overall goal is to provide the technical grounding needed to provide a better understanding into how AI will actually impact societal issues both in the near and long term, says Anthony Annunziata, director for AI open innovation at IBM. “There’s a lot of noise today,” he says.
As part of that effort, the AI Alliance will foster development of open multilingual, multi-modal models that can help address challenges involving, for example, climate and human health care. Those efforts will start with training resources and benchmarks, tools and other resources to enable the responsible development and use of AI systems at global scale, including the creation of a catalog of tools for vetting safety, security and trust.
The AI Alliance will also seek to identify pragmatic approaches to applying guardrails to AI by, for example, working with governments around the world to best determine policies that have a strong technical foundation, he adds.
Finally, the AI Alliance will also work toward defining best practices for building and deploying AI models across teams made up of data scientists, developers, software and data engineers and cybersecurity professionals in addition to helping to build a hardware accelerator ecosystem for optimizing AI model performance.
In general, the AI Alliance is committed to promoting the adoption of open source software as an alternative to proprietary black box AI models that organizations lack any visibility into how, for example, ethics were baked into the AI model, he adds. It’s especially challenging for universities that lack massive amounts of access to capital to participate and review proprietary AI research and development projects, says Annunziata.
It’s not clear to what degree open approaches to building and deploying AI models might supplant proprietary approaches but if IT history is any guide, the pace at which innovation occurs in an open model will eventually surpass a closed proprietary platform. There may be times when a provider of a platform is able to leap ahead of rivals, but over time it’s been shown that open source software projects not only quickly catch up but are also, thanks to the contributions of so many developers, more likely to surpass the pace at which proprietary platforms are updated.
Regardless of approach, the overall pace at which AI models are being advanced continues to accelerate as more compute resources are made available. The parameters in terms of the sizes of the data sets that can be used to train AI models only increase with each successive wave of new processors, so that capabilities which were once considered to be the realm of science fiction are already becoming a reality faster than society as a whole is prepared to absorb.
