Anthropic has brought in Accenture as its first embedded evaluator to put CEO Dario Amodei’s plan to slow AI development into practice. The arrangement allows Anthropic to keep looking for other evaluators as it moves forward, with the company expecting to announce more partners in the coming weeks.
On Sept. 12, Amodei published a three-step proposal aimed at slowing AI development and giving time for safeguards to be established, a move made amid warnings of catastrophic peril from swift, unchecked advancement. OpenAI CEO Sam Altman and SpaceX CEO Elon Musk both responded positively to the proposal, while Nvidia CEO Jensen Huang did not, saying such regulation was not necessary.
In his proposal, Amodei wrote that “AI has been advancing drastically faster, driven primarily by AI’s growing ability to build the next generation of AI. This dynamic is called recursive self-improvement,” and “left unchecked, it could outrun our ability to understand and control these systems.”.
The proposal’s first step
Amodei’s proposal begins with independent evaluators who hold employee-like access, and Anthropic had already committed to that step on its own, he wrote. The company announced on Friday that it would work with Accenture and its AI subsidiary Faculty to evaluate AI models, assess system alignment and test safety features.
The specific arrangements for this project are still being settled, since embedded evaluation remains an emerging area of work. Both Anthropic and Accenture have each committed to investing at least $1 billion in the project over the next five years, as stated in the announcement.
Funding the effort
Anthropic has pointed out that there is no existing system for funding independent evaluation. In its view, long-term support ought to come from pooled or government sources. Because of how urgent the work is, however, Anthropic will pay for Accenture’s work directly.
“Embedded evaluation is an emerging area, and we look forward to partnering with Anthropic to help accelerate the development of embedded evaluators, which we see as an important part of the safety landscape going forward,” said Julie Sweet, chair and CEO of Accenture.
What the partnership covers
Accenture’s Faculty is expert in testing and evaluating models for some of the world’s leading AI labs and building complex AI systems that are safe and ethical by design, the company said on Friday.
| Role | Company | Investment |
|---|---|---|
| Embedded evaluator | Accenture | $1 billion over five years |
| Partnering with Faculty | Anthropic | $1 billion over five years |
Anthropic’s systems are built so that independent evaluators can examine their workings, and the partnership exists to ensure that such access remains available to them.
Anthropic is placing its wager on Accenture’s experience in assessing and examining models for major AI labs, which it thinks can be adapted into a built-in evaluation system. The company believes this approach will enable it to grasp and manage its systems.
The commitment to making the work happen now is behind Anthropic’s decision to fund Accenture directly, a move that shows how urgent the company considers the project.
Anthropic is not the only entity allowed to join this arrangement, so the company may eventually bring additional evaluators into the program.
Source material: “Anthropic tabs Accenture as embedded evaluator to help with AI slowdown proposal,” Cointelegraph.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.