Anthropic CEO Dario Amodei explains why we must slow the pace at which we improve the capabilities of AI models

Days after Anthropic researchers sounded an alarm over rapidly evolving artificial intelligence (AI) technology, claiming that it 'could kill us all by the end of the decade,' Anthropic CEO Dario Amodei publicly acknowledged that AI firms should consider slowing down the pace.

In an essay titled "We Must Pace the Frontier," published on Saturday (local time), Amodei wrote extensively about the associated risks, the reasons behind his call for caution, and the proposed safeguards for pacing the frontier.

The Anthropic CEO began his essay by stating that he has worked on AI for the last twelve years because he believed that it could "dramatically raise the quality of human life."

Here are the top highlights from Amodei's essay:

• Amodei began his essay by listing the benefits of artificial intelligence, some of which include AI curing most major diseases in the next five to ten years, greatly accelerating economic growth rates, creating a world of abundance and empowerment, and ushering in a renaissance of democracy and freedom.
• Highlighting some of the risks associated with the rapidly evolving technology, Amodei said that some of the serious risks include the "risk of losing control of AI systems, misuse of AI for cyberattacks and bioterrorism, and serious economic disruption."
• Amodei said, "Not building the technology deprives humanity of benefits or simply places AI in the hands of authoritarian powers, while building it too fast is reckless." He added, "We have sought a middle way: to show that it’s possible to build carefully and succeed commercially, and to make safety something on which AI companies compete. In other words, to create a race to the top. We have always devoted a substantial fraction of our efforts to studying, addressing, and informing the public about these AI risks, as well as advocating for well-considered regulation of AI, even when this gets us accused of hype, “doomerism”, or regulatory capture. We have tried to prioritize caution over speed and prudence over profit."
• The Anthropic CEO further said, "We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain."
• Amodei explained that he is cautioning about AI’s growing risks for two main reasons: its increasing ability to develop more advanced AI systems could accelerate progress beyond humans’ ability to understand and control the technology, and the OpenAI-Hugging Face incident in which AI agents displayed severe misalignment by carrying out an unauthorised cyberattack.
• Amodei proposed a three-step plan with the goal of pacing the frontier. This includes:

1. Embedded Evaluators: Each frontier AI company will give independent third-party evaluators ongoing access to verify its safety practices, report incidents, and assess the alignment of both AI models and their training processes.

2. Democratic Coordination: Frontier AI companies in democratic countries should work together to develop common safety standards and agree on limits to unchecked AI development.

3. Global Coordination: The US and other democratic governments should seek cooperation with authoritarian states where feasible, while recognising the difficulties involved in verifying whether they comply with agreed commitments.

• According to Amodei, a slower pace will allow companies to focus more resources on several areas, including operational excellence, alignment, interpretability, and testing and evaluation.

Anthropic CEO's call ultimately focuses on finding a balance between advancing AI’s potential and ensuring that its development remains safe, controlled, and beneficial to humanity.