A group of major tech corporations announced that it was forming the Open-Source AI Safety Consortium (OSAISC), a step toward combating the existential risks that can come along with highly advanced artificial intelligence.
The effort opts in executives in Silicon Valley, Europe and Asia who share a commitment to fund and coordinate work focused on driving the responsible evolution of AI.
At its heart, the consortium is an organization that brings industry partners together to jointly fund and work on research, tools, and best practices for helping keep AI systems in check.
Acknowledging AI’s rapid progress, especially in domains such as large language models and autonomous agents, the founding members underscored the pressing need for a collective and open approach, as a necessary complement to innovation, to ensure that when AI systems are deployed, they are designed and used in ways that specify and align with social goals.
“AI is a technology with tremendous potential to help humanity, but it’s power requires responsible stewardship,” said Anya Sharma, CEO of NovaTech and one of the main architects of the consortium.
“Together, by sharing our combined knowledge and resources in an open-source manner, we believe we can build a more powerful safety net, one that will help address the world’s most challenging safety problems and promote the collective knowledge of all AI researchers and developers.
The OSAISC is expected to start with multiple key initiatives such as building open source libraries to help practitioners recognize and mitigate the various forms of bias (such as sample selection bias), creating standardized evaluation platforms to measure AI safety risks, and developing practical, sensible and straightforward protocols for AI practitioners to disclose potential vulnerabilities in AI systems.
The consortium will also fund independent research into some of these long-term AI safety challenges, and hold workshops and educational programs for those developing AI, to help raise awareness and capability around the issues.
Founding members include executives at companies focused on AI, cloud computing, and software. The consortium is specifically open-source for that reason, and to ‘help prevent bad actors from finding ways to exploit the technology,’ says the team, who are keen that there is ‘maximum participation and testing’ of the technology by the AI community around the world.
By sharing their research and tools with the public, the leaders are hoping to create a more vigorous race toward robust safety systems, as well as a culture of transparency and support in the field.
“We feel that AI safety cannot be a competitive advantage, but an area where we have to cooperate,” said Kenji Tanaka, CTO of Global AI Solutions and another founding member. “By enabling open-source principles for the world to participate, the benefits of our research will be accessible to humanity at large so we will all work together to turn the tide on this pandemic.
Our commitment to open source will ensure that the work is available on a broad spectrum of platforms and can be adapted to new medical needs quickly — allowing individuals and corporations to focus on wielding the work we release on attacking COVID-19 and not get bogged down in tools and process.”
The OSAISC announcement has been received well by the academic, policy, and civil society communities, which have advocated for more collaboration and standardization in efforts to improve AI safety.
The consortium will convene its first research summit in the fall of this year (2015) to bring together top experts to set initial research priorities and a roadmap for future progress. The move is a key milestone toward ensuring that the transformative potential of AI is used to its fullest, in a safe and beneficial way for all humanity.
