OpenAI Employees Advocate for Protections to Address ‘Serious Risks’ of AI

OpenAI Employees Advocate for Protections to Address ‘Serious Risks’ of AI
Image Source: Y CombinatorBy Tue, 04 Jun 2024 17:56:16 GMT

A coalition of current and former employees from leading AI companies, including OpenAI, Google DeepMind, and Anthropic, has united to advocate for greater transparency and protection from retaliation for those who voice concerns about the significant risks associated with artificial intelligence (AI). This group, comprising 13 signatories, highlights the critical need for oversight and the establishment of robust whistleblower protections in the industry.

The public letter, titled "A Right to Warn about Advanced Artificial Intelligence," underscores the potential dangers posed by AI technologies. These include the entrenchment of existing inequalities, the spread of misinformation, and the possible loss of control over autonomous AI systems, which could lead to catastrophic consequences. The signatories, which include prominent figures such as Yoshua Bengio and Geoffrey Hinton, known as the "Godfathers of AI," emphasize that these risks are acknowledged by both AI companies and governments worldwide.

The employees argue that the current environment in AI firms stifles important discussions about these risks due to stringent confidentiality agreements. These agreements prevent employees from sharing their concerns with anyone outside the company, limiting accountability and transparency. Jacob Hilton, a former OpenAI employee, commended the company's recent move to release past employees from non-disparagement agreements but stressed that fears of other forms of retaliation, such as being fired or sued, still persist.

Daniel Kokotajlo, another former OpenAI employee, echoed these sentiments, stating his belief that companies are not adequately prepared for the potential impacts of artificial general intelligence (AGI). He pointed out that the rush to develop AGI without sufficient safeguards or transparency could have dire consequences. Kokotajlo, who resigned from OpenAI partly to avoid signing a non-disparagement agreement, expressed his concerns about the industry's "move fast and break things" mentality.

The letter outlines four key principles that the signatories believe AI companies should adopt to address these issues

  • AI companies should not enforce agreements that prevent employees from criticizing the company regarding risk-related concerns, nor should they retaliate by hindering any vested economic benefits.
  • Companies should establish a verifiably anonymous process for current and former employees to raise risk-related concerns to the company’s board, regulators, or an independent organization with relevant expertise.
  • AI companies should foster a culture that allows employees to openly raise risk-related concerns with the public, the company's board, regulators, or appropriate independent organizations, ensuring that trade secrets and intellectual property are protected.
  • Companies should refrain from retaliating against employees who share risk-related confidential information publicly after exhausting internal processes. Until adequate anonymous reporting mechanisms are in place, employees should retain the freedom to report their concerns to the public.

In response to these calls for action, an OpenAI spokesperson stated that the company is proud of its track record in developing safe and capable AI systems. They emphasized the importance of rigorous debate and ongoing engagement with governments, civil society, and global communities. Despite these assurances, the employees remain steadfast in their demand for more substantial protections and transparency.