OpenAI seeks public input on how to make its future AI models “align with the values of humanity.”
Today, the AI firm announced that it is launching a new Collective Alignment team of academics and engineers to gather and “encode” public feedback on its models’ actions into OpenAI products and services.
As stated in a blog post, OpenAI plans to collaborate with external advisers and grant teams, including pilots, to include prototypes in model steering. “Recruiting research engineers with diverse technical backgrounds to support our work.”
OpenAI’s public initiative, established last May to provide money to finance experiments in putting up a “democratic process” for defining AI system rules, spawned the Collective Alignment team. At its launch, OpenAI announced it will support individuals, teams, and organizations to build proof-of-concepts to tackle AI governance and guardrail problems.
OpenAI recapped the award winners’ work in its blog post today, ranging from video chat interfaces to crowdsourced AI model audit platforms and “approaches to map beliefs to dimensions that can be used to fine-tune model behavior.” All grantee codes, proposal summaries, and key takeaways were released this morning.
OpenAI has tried to distance the initiative from its economic objectives. Given OpenAI CEO Sam Altman’s critiques of the EU and other restrictions, that’s hard to stomach. Altman, OpenAI president Greg Brockman, and chief scientist Ilya Sutskever have stated that AI innovation is too quick for present institutions to regulate, so we must crowdsource the effort.
Meta and other OpenAI opponents have accused OpenAI of campaigning against open AI R&D to achieve “regulatory capture of the AI industry.” OpenAI rejects this and may cite its funding program and Collective Alignment team as proof of its “openness.”
In any event, OpenAI is under increased scrutiny from lawmakers, including a U.K. review of its connection with closer investor Microsoft. A Dublin-based subsidiary helped the firm decrease its EU data privacy regulatory risk by limiting the bloc’s privacy watchdogs’ unilateral action.
OpenAI declared yesterday that it’s collaborating with groups to minimize the harmful use of its technology to influence elections, likely to appease authorities. The startup’s solutions make AI-produced photos more visible and even detect created material after modification.
