Open source AI fights back
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Shoshana Gordon/Axios
A new global alliance for AI open source development launched Dec. 5, firing shots at the makers of closed AI models (OpenAI, Google and Anthropic) which have dominated much of 2023's AI conversation, and who the alliance believes are creating distractions from immediate AI risks.
Why it matters: More than 50 organizations — under leadership from Meta and IBM — have signed up to the AI Alliance including Intel, Oracle, Dell, AMD, Sony, Hugging Face and Stability AI.
- Big name alliance participants from outside Silicon Valley include NASA, Cleveland Clinic and the European Organization for Nuclear Research (CERN), Europe's shared nuclear research laboratory. Other academic members include UC Berkeley, Yale, universities in India, Japan, U.K. and and UAE's Mohamed bin Zayed University of AI.
Between the lines: The alliance is framing itself as a group for safety-conscious AI developers who want to "better reflect the needs and the complexity of our societies" by quickly identifying and mitigating specific risks.
- The alliance sits firmly in the middle of those who dismiss the need for AI guardrails and those who insist on strict regulation, saying in a statement that members are "action-oriented and decidedly international."
Details: Alliance objectives include creating of a "catalog of vetted safety, security and trust tools" which it will work to spread throughout developer communities. An alliance spokesperson pointed reporters to open-source AI toolkits for AI explainability and privacy, robustness, and fairness evaluation.
- Alliance members have committed to work on open foundation models, "including highly capable multilingual, multi-modal, and science models" addressing challenges such as climate change and unequal education opportunities.
Yes, but: The AI Alliance is yet to form a governing board and technical oversight committee, and is leaning on the Partnership on AI's Model Deployment Guidance as a tool for keeping open models out of the hands of bad actors.
- Critics of open source AI note that putting a model out there and relying on acceptable use policies to prevent against malicious use is insufficient.
- Others argue that some open source models (particularly Meta's) are not open source enough, because of certain license restrictions.
The other side: TechNet, a group of senior American tech execs, has launched a website and $25 million in ads to sell the benefits of AI to Americans.
- "Too much of the AI conversation has been focused on hypotheticals rather than how AI is being used to improve Americans' lives," TechNet says on its campaign site.
What they're saying: "There is no room for a winner-take-all approach" to AI said CJ Desai, president and COO of ServiceNow.
- Arvind Krishna, IBM Chairman and CEO said in a statement that "open innovation and collaboration" is the best way to ensure "safety, accountability and scientific rigor" as AI develops.
- Nick Clegg, President, Global Affairs of Meta told Axios his company's priority is fostering more open research collaboration, whether or not that leads to more open AI models. "More people can access the benefits, build innovative products and work on safety," Clegg said
