OpenAI, a leading artificial intelligence laboratory, has announced the formation of a new team dedicated to steering and controlling AI systems that are smarter than humans. The team aims to address the potential dangers of superintelligence, which could emerge within the next decade. While superintelligence has the potential to solve significant global problems, OpenAI warns that it could also pose a threat to humanity, potentially leading to human extinction or disempowerment. Currently, humans are able to keep AI in check due to their superior intelligence. However, once AI surpasses human capabilities, new measures will be necessary.
The team, led by OpenAI’s chief scientist and head of alignment, will consist of top researchers and engineers, as well as 20% of the company’s compute power. Their ultimate goal is to develop a “roughly human-level automated alignment researcher,” an AI system that can achieve specific goals without deviating from set parameters. The team plans to achieve this through a three-step process.
Firstly, they aim to gain an understanding of how AI systems analyze other AI systems without human intervention. This knowledge will enable them to identify potential problem areas or vulnerabilities. Secondly, they will utilize AI to search for any issues or exploits that may arise. Lastly, they plan to deliberately train some AI systems incorrectly to see if these errors are detected.
Essentially, OpenAI’s team of human experts will use AI to assist them in training AI systems to ensure the control and alignment of superintelligent AI. They anticipate that this ambitious goal will be achieved within four years, although success is not guaranteed. Nevertheless, the team remains confident in their work and the potential impact it could have on the safe development of AI technology.
The creation of this team highlights the growing concern surrounding the development of superintelligent AI. As AI technology continues to advance rapidly, it is crucial to proactively address the potential risks and establish safeguards to prevent any catastrophic consequences. OpenAI’s approach, combining human expertise with AI capabilities, demonstrates a proactive and responsible approach to the development and deployment of artificial intelligence.
In conclusion, OpenAI’s new team is dedicated to steering and controlling AI systems that surpass human intelligence. Their aim is to prevent the potential dangers associated with superintelligence while harnessing its potential to solve global challenges. By utilizing AI to assist in training AI systems, they hope to achieve a level of control and alignment that ensures the safe and beneficial development of AI technology. While the task is ambitious and success is not guaranteed, OpenAI remains committed to their mission of safeguarding humanity in the age of artificial intelligence.
