Sam Altman argues for accepting certain risks of AI for its benefits, highlighting OpenAI's stance on regulation and safety measures.
Sam Altman, the CEO of OpenAI, emphasizes that society must come to terms with certain drawbacks of artificial intelligence in exchange for its benefits and accessibility. In a recent interview with Politico, he stated, "We believe the world must accept some bad things for the benefits of this technology and the freedom it gives people."
Altman highlighted issues such as significant breaches, misuse of AI, and fraud as potential downsides of the technology. He noted that completely mitigating these risks would necessitate stringent regulations, a compromise he is not prepared to make. According to Altman, the overall good that can be achieved through AI will outweigh the negative aspects.
However, he did stress that there are limits to what is acceptable, particularly regarding catastrophic outcomes like losing control over AI systems.
Contrasting Views on AI Regulation
The interview also explored the differing approaches of OpenAI and Anthropic concerning AI safety and governance. Altman pointed out a "fundamental difference in perspective" between the two organizations. OpenAI supports broader access to advanced AI systems with fewer regulations, while Anthropic advocates for a more cautious approach.
He acknowledged the arguments of those in favor of strict controls, who believe that the technology's potential dangers warrant oversight by a single lab in San Francisco. Altman, however, deemed this viewpoint as an unacceptable compromise, arguing that the pursuit of zero harm should not lead to centralized control of AI technology.
Recent Developments in AI Safety
Despite their differences, OpenAI and Anthropic have begun to align on certain safety measures. On September 12, Anthropic CEO Dario Amodei called for a slowdown in AI model advancements to ensure that safety protocols keep pace with technological progress. He proposed that independent experts should have continual access to AI systems for behavior evaluation and compliance verification.
On the same day, Altman voiced support for regulating the pace of advanced model development and acknowledged the necessity of independent assessments. OpenAI has also ramped up its advocacy for mandatory regulations, endorsing federal safety requirements tied to the capabilities of neural networks.
Before national regulations are enacted, OpenAI is backing four bills in California focused on independent AI assessments, auditor standards, youth protection, and defenses against biological threats. The company has identified serious threat categories, including offensive cyber capabilities and harmful manipulations, and has outlined specific procedures for risk assessment and mitigation.
During a United Nations Security Council meeting on September 23, Altman reiterated the importance of keeping powerful AI systems under human oversight and fostering international cooperation in this field.
Notably, a U.S. Senate subcommittee has initiated an investigation into OpenAI's actions following a July incident involving Hugging Face.