Karen Hao questions who controls AI’s goals and public narrative

A critical look at the AI industry

In this “Most Replayed Moment” from The Diary Of A CEO, journalist Karen Hao discusses the ideas and incentives guiding major artificial-intelligence companies. The conversation ranges from disputed theories of intelligence to corporate influence over research and journalism. It represents Hao’s account and opinions within the podcast, rather than independently corroborated reporting.

AGI ambitions rest on a contested theory of intelligence

Hao says some leading AI researchers regard the human brain as a vast statistical model. Within that framework, making artificial models progressively larger could eventually produce systems equal or superior to human intelligence. She stresses, however, that this is a scientific hypothesis—not an established fact—and that critics consider the comparison reductive.

That distinction matters, Hao argues, because companies are organising their strategies around the hypothesis. She says the drive to scale models entails collecting more data, constructing more data centres and using extensive human labour. Rather than accepting human duplication as technology’s natural goal, she asks why AI development should not focus more narrowly on tools intended to improve outcomes such as drug discovery or healthcare.

The ‘empire’ metaphor challenges control over resources

Hao describes major AI companies as “empires,” a metaphor she uses to connect their scale and ambitions with claims over data, intellectual property, land and labour. In her telling, these businesses also seek to shape knowledge by funding much of the research into AI’s capabilities and limitations. Her resulting allegation is that industry priorities can determine which questions receive support and that unwelcome findings may face resistance.

As an example, Hao recounts the case of researchers Timnit Gebru and Margaret Mitchell at Google, alleging that conflict over a critical paper preceded their departures. The episode does not include Google’s response. Hao also discusses what she characterises as scrutiny and intimidation directed at an OpenAI critic, but these claims are presented solely through her account in the conversation.

China features in a narrative of urgency

Hao argues that AI companies justify concentrated control by presenting a stark choice: a supposedly responsible organisation must develop powerful technology first, or a dangerous rival will. She says China is now frequently cast in that rival role, while Google occupied it in OpenAI’s earlier rhetoric. In her view, optimistic promises of abundance and warnings of catastrophic harm work together to support an approach with limited public participation.

Access can shape technology coverage

The discussion also turns to journalism. Hao says technology companies can reward favourable access—or withhold it from reporters who scrutinise them or platform critics. She recounts her own deteriorating relationship with OpenAI, saying the company ultimately declined interviews for her book and did not answer extensive requests for comment. The host describes experiencing similar prolonged promises of access from an unnamed AI figure’s team.

Hao says losing direct access initially made her question whether critical reporting had harmed her career. She ultimately concluded that working without the industry’s “front door” strengthened her independence, pointing to more than 300 interviews completed for her reporting. For her, the wider concern is that access management can influence which views reach the public and how AI companies are perceived.