Former employees who worked at renowned artificial intelligence (AI) developers are urging these pioneering companies to strengthen their whistleblower protections. The aim is to enable them to raise concerns related to risks associated with the advancement of sophisticated AI systems to the public.
A group of 13 former and current employees from OpenAI (ChatGPT), Anthropic (Claude), and DeepMind (Google), alongside prominent figures in the AI field like Yoshua Bengio, Geoffrey Hinton, and Stuart Russell, have launched the 'Right to Warn AI' petition on June 4. The petition is a collective effort to push for a commitment from leading AI companies to allow employees to voice risk-related concerns about AI both internally and externally.
One of the advocates for this cause, William Saunders, a former OpenAI employee, emphasized the necessity for mechanisms to share information about potential risks associated with emerging technologies with independent experts, governmental bodies, and the general public.
According to Saunders, the individuals with the deepest insights into the workings and risks of cutting-edge AI systems often face constraints in sharing their knowledge due to fears of repercussions and overly broad confidentiality agreements.
Right to Warn Principles
The 'Right to Warn AI' petition includes four key propositions directed at AI developers. Firstly, it calls for the elimination of non-disparagement clauses related to risks, ensuring that employees are not silenced by agreements that hinder them from expressing concerns about AI risks or subjecting them to punitive actions.
Secondly, the proposal advocates for the establishment of anonymous reporting channels to encourage individuals to voice apprehensions regarding AI risks. This move aims to nurture a culture where open critiques about such risks are welcomed.
Lastly, the petition demands safeguards for whistleblowers, seeking assurances that companies will not retaliate against employees who divulge information aimed at exposing critical AI risks.
Saunders described these proposed principles as a proactive approach to engage with AI companies in fostering the development of safe and beneficial AI technologies.
Escalating AI Safety Concerns
The petition's emergence coincides with mounting worries about the negligence displayed by AI labs towards the safety of their latest models, particularly in the realm of artificial general intelligence (AGI). The pursuit of AGI entails crafting software with humanlike intelligence and self-learning capabilities.
Daniel Kokotajlo, a former OpenAI employee, cited his loss of faith in the company's responsible actions, particularly concerning AGI development, as a reason for his departure.
Kokotajlo criticized the 'move fast and break things' approach adopted by some entities in the AI sphere, emphasizing its unsuitability for a technology as potent and poorly understood as AGI.
Recent reports, such as the claims made by Helen Toner, a former OpenAI board member, during a Ted AI podcast about Sam Altman's alleged dismissal from OpenAI, have further fueled concerns about transparency and accountability within AI organizations.
