“We believe that the world should accept that certain bad things will happen, in exchange for the benefits of this technology and for people to have the ability to choose and control,” Altman said in an interview with Decoded of the Politico website, published yesterday (Sunday).
His words were said as part of a discussion about the growing dispute between OpenAI and one of its main competitors, the Anthropic company, regarding the degree of supervision and regulation necessary for advanced artificial intelligence systems. According to Altman, although the positions of the two companies have come closer in recent months on several safety issues, there is still a fundamental gap between them in their worldview. “I think there is a very big gap between us,” he said.
“One lab in San Francisco can’t decide for everyone”
Altman said he understands the view that AI systems may become so powerful and dangerous that they should be limited to a very small number of entities. According to him, there are those who believe that “one lab in San Francisco” should control the technology, make sure nothing bad happens and decide how to distribute its benefits to the public.
However, he rejected this approach and called it “a completely unacceptable compromise”. In his opinion, the attempt to completely prevent the misuse of artificial intelligence may create an equally dangerous concentration of power, in which a limited number of companies or institutions will determine who may use the most powerful systems and for what purposes.
However, Altman stressed that his words do not refer to catastrophic risks. He said that he is not ready to accept scenarios of “truly catastrophic risks”, including a situation of “significant loss of control over AI”. That is, according to him, there is a difference between damages such as fraud, cyber attacks and other malicious uses, and a situation where advanced systems go out of control.
The dispute with Anthropic
The debate is taking place against a background of growing concerns about the ability of advanced models to be used for fraud, cyber attacks, manipulation and the spread of false information. Altman does not deny the risks, but claims that trying to prevent any damage in advance may require such severe restrictions that the public will lose access to the technology and the ability to enjoy its benefits.
In other words, the position of the CEO of OpenAI is that the goal of “zero harm” is unrealistic, and that the main question is not whether AI will cause any harm, but what extent of risk the company is willing to accept in exchange for the benefit it may provide.