The CEOs of leading AI companies OpenAI and Anthropic have recently called for a deliberate slowdown in the development of artificial intelligence technologies. Anthropic’s CEO, Dario Amodei, published a detailed essay urging AI firms to collaborate in pacing the advancement of AI to ensure safety measures keep up with rapidly evolving capabilities. OpenAI’s CEO, Sam Altman, supports this approach and has also decided against taking the company public this year, citing concerns about AI safety as a reason to avoid an ill-advised IPO at this time.
Expert Gary McGrath, CEO of the Berryville Institute of Machine Learning, provides insight into these concerns, emphasizing that current AI models, while sophisticated, lack human-like intent or understanding. He suggests that referring to AI as “intelligent” can be misleading and proposes viewing these systems as “alien intelligence” — powerful tools created and directed by humans rather than autonomous entities with their own goals. McGrath stresses that any misuse or dangerous outcomes ultimately stem from human actions, not the AI itself.
McGrath also critiques the tendency within the industry to anthropomorphize AI models, attributing intentions or consciousness to them, which can cloud judgment and lead to misunderstandings about the technology’s true nature. He points out that while AI systems can perform complex tasks, they do so under human instruction and are not yet autonomous or self-directed. This perspective challenges some of the alarmist narratives and highlights the importance of clear, accurate communication about AI capabilities and limitations.
Regarding calls for regulation and transparency, McGrath advocates for greater openness from AI companies about their models’ architectures, training data, and testing processes. He argues that independent external experts could better assess risks if given access to this information, rather than relying solely on internal evaluations or self-policing by the companies themselves. While acknowledging that government regulation may struggle to keep pace with rapid AI development, he suggests that enforcing existing computer security laws and demanding transparency could be effective interim measures.
Finally, McGrath cautions against framing AI development as a geopolitical arms race akin to nuclear proliferation, particularly concerning competition with China. Instead, he encourages international cooperation and a balanced view that recognizes both the risks and the significant benefits AI can bring to humanity. Ultimately, he calls for treating AI as powerful tools requiring responsible stewardship and transparency, rather than as autonomous agents, to ensure safe and beneficial progress in the field.
Useful Links
