Perplexity Co-Founder Says AI Safety Justifies Power Concentration

In a recent essay, Andy Konwinski, co-founder of Perplexity AI and Databricks, critiques the narrative surrounding AI safety, suggesting it is being misused to centralize power. His essay comes in the wake of Open Frontier, a meeting of around 100 AI researchers held on June 30, where issues of AI monopoly were discussed.

Konwinski points to the controversy surrounding Anthropic's AI model, Claude Fable 5, which, upon its launch, included a provision that would degrade responses for those perceived as competitors. This decision was retracted within 48 hours following public outrage. "The problem isn’t that Anthropic made a bad decision; the problem is that they assumed the decision was theirs to make," he wrote. This statement reflects a broader concern regarding who governs AI research and development.

He argues that centralizing AI capabilities does not inherently mitigate risks associated with the technology but rather introduces new challenges, potentially replicating past mistakes seen in other industries. This perspective mirrors views expressed by UC Berkeley's Jennifer Chayes, who criticized the current safety messaging from major AI companies as a fear campaign aimed at stifling innovation.

Konwinski's vision advocates for a more decentralized approach, proposing the establishment of a research commons for AI that enables broader access and innovation. He notes that AI is foundational infrastructure comparable to railroads and electricity, emphasizing that the future organization of society will hinge on who controls AI access.

His critique serves as a timely reminder of the implications of power dynamics in the rapidly evolving AI landscape, urging stakeholders to reconsider how access and control over AI technologies are regulated.