Karen Hao, the former senior AI editor at MIT Technology Review and current contributor to Bloomberg Weekend Interview, has spent years embedded in the labs and boardrooms where artificial intelligence gets made. Her latest feature, headlined 'AI Doesn't Have to Be Built This Way,' challenges the assumption that the current trajectory of AI development is either inevitable or optimal.

The Critique: Path Dependence Meets Hype

Hao's argument centers on what she calls the 'path dependency' problem in modern AI—a field that has largely converged on a handful of architectural assumptions (large transformer models, massive datasets, compute-intensive training runs) because that's where the money and talent flowed first. The implication isn't that these approaches don't work; it's that other viable paths may have been foreclosed before anyone seriously explored them. For developers watching from the sidelines, this matters: infrastructure decisions made today will echo for years, and betting exclusively on the current paradigm carries risks that aren't always visible in quarterly earnings calls.

What This Means for Builders

The practical dimension of Hao's argument cuts close to home for anyone building dev tools or infrastructure. If the underlying assumptions about what AI 'needs' are more flexible than the industry suggests, then the entire stack—frameworks, APIs, deployment targets, evaluation metrics—might look different in five years. Hao has been consistent in her view that researchers should spend as much time questioning their priors as chasing benchmarks. Whether or not you agree with her conclusions, the underlying question is worth sitting with: how much of what we consider 'the way' is actually just momentum?

The Broader Conversation

This isn't the first time Hao has pushed back against industry groupthink. Her 2020 investigation into Facebook's AI ethics team (later published as a book) and her ongoing coverage of labor issues in AI training pipelines have established her as someone willing to name uncomfortable dynamics that others prefer to sidestep. The Bloomberg interview format gives her room to go deeper than a traditional news piece, which is probably why it's generating discussion among practitioners even with limited public access to the full text.

Key Takeaways

  • Current AI development paths are not inevitable but path-dependent on early investment decisions
  • Developers should consider whether their infrastructure choices assume too much about future model architectures
  • Hao's work consistently challenges the 'move fast and build anyway' ethos dominating Silicon Valley
  • The interview is part of Bloomberg Weekend Interview's broader effort to feature substantive tech criticism

The Bottom Line

The AI industry has a credibility problem hiding in plain sight: it presents contingent choices as laws of nature. Whether Hao's critique lands or gets absorbed into the same algorithm that buries uncomfortable takes, her willingness to ask 'what if we're wrong?' is exactly the kind of skepticism infrastructure builders need more of—not just from journalists, but from themselves.