Anthropic CEO Dario Amodei: "For too long the industry lied" about AI risks

Artificial intelligence carries real dangers, asserts Anthropic CEO Dario Amodei, lamenting that the industry has long lied about the risks associated with its development. Former Anthropic researcher Jacob Coxon warned on social media that Anthropic and OpenAI are gambling with humanity's existence through the advancement of self-improving superintelligence, likening it to the plot of "Terminator" films.

Amodei, in response, admitted he had underestimated the rapid progress of AI, acknowledging that the technology's development is following an exponential curve. Rather than advocating for a panic or shutdown, Amodei emphasized the need for a slower pace and more rigorous testing of each AI generation. To achieve this, Anthropic will grant independent evaluators permanent access to their models, comparing this to the role of a food inspector in ensuring compliance with safety protocols.

However, AI experts argue that these measures are insufficient, and Amodei concurs, supporting the idea of federal regulation. Proposed legislation ranges from an outright ban on superintelligence to a mandatory "kill switch" for dangerous AI systems. While Amodei acknowledges the potential benefits of AI, such as medical advancements, he cautions against completely banning the technology or allowing its development in authoritarian countries.

He proposes working together to establish a "speed limit" on AI progress, a notion he admits is challenging due to the competitive advantages gained through rapid development. Ultimately, Amodei believes that if AI is built responsibly, the probability of catastrophic outcomes remains low, but warns that building it incorrectly could result in dire consequences.