In a bold statement, OpenAI, the creator of ChatGPT, has expressed concerns about the potential dangers of artificial intelligence (AI) surpassing human intelligence and even leading to the extinction of the human race. To address these risks, OpenAI is forming a team dedicated to mitigating the impact of superintelligence.
OpenAI acknowledges that controlling AI systems smarter than humans necessitates scientific and technical breakthroughs. To tackle this challenge within the next four years, the company is allocating 20% of its compute resources to this endeavor.
While OpenAI believes that superintelligence could be the most revolutionary technology ever invented and could potentially solve global issues, it also recognizes the inherent dangers associated with its immense power. The company warns that if left uncontrolled, superintelligent AI could disempower humanity or even lead to human extinction. Although the arrival of such AI may seem distant, OpenAI suggests it could become a reality sometime in this decade.
OpenAI’s newly established team, known as the superalignment team, aims to develop a human-level automated alignment researcher. Alignment research focuses on ensuring AI systems align with human values and intentions. The team plans to train AI systems using human feedback, subsequently training AI systems capable of evaluating other AI systems. Ultimately, they aim to construct an AI that can perform alignment research more efficiently and effectively than humans.
OpenAI acknowledges the ambitious nature of its goal to resolve the technical challenges associated with superintelligence within four years. However, the company remains optimistic. To achieve this objective, OpenAI is seeking to recruit researchers and engineers to join the team.
OpenAI co-founder Ilya Sutskever and Jan Leike, the co-head of the new team, emphasize that solving the problem of superintelligence alignment fundamentally lies within the domain of machine learning. They stress the importance of involving machine learning experts, regardless of whether they are currently working on alignment, in finding a solution. OpenAI also emphasizes its commitment to sharing the outcomes of this effort widely and emphasizes the significance of contributing to the alignment and safety of non-OpenAI models.
The concerns expressed by OpenAI echo those of other experts, including Geoffrey Hinton, commonly referred to as one of the ‘Godfathers of AI.’ Hinton recently parted ways with Google, warning about the increasing dangers posed by more powerful AI systems. Additionally, OpenAI’s CEO, Sam Altman, has previously voiced concerns about the potential extinction of the human race due to AI, drawing parallels to nuclear war and pandemics. A survey shows that more than two-thirds of Americans worry about AI’s impact on civilization. Warren Buffett has even compared the creation of AI to the development of the atomic bomb.
However, not everyone shares these concerns. Prof Yann LeCun, the chief scientist at Meta and one of the three AI Godfathers, dismisses warnings about AI posing threats to humanity as ridiculous.
As OpenAI takes a proactive approach to address the potential risks associated with superintelligent AI, the company recognizes the necessity of collaborative efforts and the involvement of experts from various fields. While the timeline for solving the challenges remains ambitious, OpenAI remains hopeful and committed to sharing their progress and contributing to the wider community’s efforts in ensuring the alignment and safety of AI systems.
