Shivs Telegram Media: A Plan for Maintaining Control over Super-Intelligent AI
2 min read
OpenAI, a leading artificial intelligence research organization, is facing criticism for veering off its original mission of building AI for the benefit of humanity. The company’s focus on commercial ambitions has overshadowed its commitment to safeguarding against potential risks associated with superintelligent AI.
To address these concerns, OpenAI has established a research group named Superalignment. The group’s primary objective is to develop methods to control highly capable and potentially dangerous AI models, as Artificial General Intelligence (AGI) approaches at a rapid pace.
As part of its commitment to the Superalignment project, OpenAI plans to allocate a significant portion, one-fifth, of its computing power. This move reflects the organization’s dedication to explore innovative solutions and preventive measures for potential risks associated with AGI.
In a recent research paper, OpenAI detailed experiments focused on enabling an inferior AI model to guide the behavior of a more intelligent one without compromising its capabilities. The researchers investigated the use of supervision to fine-tune AI systems. However, as AI continues to advance, automating this process becomes crucial due to the growing difficulty in obtaining meaningful human feedback.
One of the control experiments conducted by OpenAI revealed that teaching a more recent system using an inferior one diminished its capabilities. To address this setback, the researchers experimented with training progressively larger models and implemented an algorithmic tweak. This adjustment enabled the stronger model to follow the guidance of the weaker one without experiencing a significant loss in performance.
While OpenAI acknowledges that these methods are not infallible, it views them as an initial step towards further research on the subject. Experts in the field appreciate OpenAI’s proactive approach in addressing the challenges associated with controlling superhuman AI. However, they emphasize that it will require years of dedicated effort to effectively tackle this complex problem.
OpenAI’s commitment to the Superalignment group and its ongoing research efforts demonstrate the organization’s dedication to ensuring the safe development and deployment of AGI. As the race towards superintelligent AI continues, OpenAI remains focused on aligning AI’s capabilities with the benefit of humanity.
“Devoted food specialist. General alcohol fanatic. Amateur explorer. Infuriatingly humble social media scholar. Analyst.”