OpenAI Forms New Team: Gathering Public Input to Ensure AI Models 'Align with Human Values'
-
On January 17th, OpenAI, a leader in the AI field, announced in its blog that it is forming a new team called 'Collective Alignment'.
This team, primarily composed of researchers and engineers, will focus on designing and implementing processes to collect public input. This will help train and shape the behavior of its AI models, addressing potential biases and other issues.
OpenAI explained in the blog: 'Our goal is to build a system that can collect and "encode" public perspectives on our models' behavior, integrating these valuable insights into our products and services. This way, our future AI models can better "align with human values."'
To achieve this, OpenAI is actively recruiting research engineers from diverse technical backgrounds for this new team. They stated: 'We will collaborate closely with external advisors and other teams, including piloting prototypes integrated into our models. We need experts from various fields to advance this initiative.' The 'Collective Alignment' team is actually an extension of a public project initiated by OpenAI in May last year. The project aims to fund experiments in establishing a 'democratic process' to determine the rules that AI systems should follow. OpenAI explicitly stated that the program is designed to fund individuals, teams, and organizations in developing proof-of-concepts to address questions about AI safeguards and governance.
Research engineer and founding member of OpenAI's new team, Tyna Eloundou, emphasized: 'As we continue to pursue the development of superintelligent models, these models will gradually become an integral part of our society. Therefore, it is crucial to provide people with the opportunity to directly contribute their opinions.'
For example, to ensure that only humans can participate in voting and providing input, OpenAI is also considering collaboration with Worldcoin. Worldcoin is a cryptocurrency project created by OpenAI CEO Sam Altman, which offers an effective method to distinguish between humans and AI bots.
Since the launch of ChatGPT in late 2022, OpenAI has garnered widespread public attention with its generative AI technology. This technology can generate text from simple prompts, making ChatGPT one of the fastest-growing applications in history. However, the rapid development of artificial intelligence technology has also raised concerns. Particularly, its ability to create "deepfake" images and other misinformation has caused unease ahead of the upcoming 2024 U.S. election. Critics point out that AI systems like ChatGPT inherently carry biases due to their input data, and users have already discovered examples of racist or sexist outputs in some AI software.
In its latest blog post, OpenAI reviewed the outcomes of its funded projects, including a video chat interface, a crowdsourced auditing platform for AI models, and "methods for mapping beliefs into dimensions that can be used to fine-tune model behavior." Simultaneously, they publicly released all the code used in the grantees' work along with brief summaries of each proposal.
Although OpenAI attempts to portray this project as unrelated to its commercial interests, this claim appears somewhat unconvincing given CEO Sam Altman's critical stance toward EU AI regulation. Altman, along with OpenAI President Greg Brockman and Chief Scientist Ilya Sutskever, has repeatedly emphasized that AI innovation is progressing so rapidly that existing authorities cannot be expected to adequately control this technology. Therefore, they proposed crowdsourcing as a solution to this challenge.