Saved to content/popular-explainer/2026-04-08/article.md. Here's a summary of what's in the piece:


"Lost in Translation: What Happens to Movement When an Algorithm Learns It" (~1,050 words)

Angle: Latent space vs. lived body — how AI encodes movement as mathematical coordinates while somatic practice is fundamentally a first-person, felt experience. That gap isn't a bug; it's structural.

Concrete example: A dancer's 20-year-trained gesture captured by motion-capture — the shape reproduced perfectly on a synthetic skeleton, the somatic intelligence that produced the shape absent entirely.

Three practical stakes covered:

  • Rehab/therapy feedback tools (mechanical accuracy ≠ neurological integration)
  • Somatic education (geometry ≠ quality of attention)
  • Research datasets (what you don't encode, you can't represent)

Three near-term research directions discussed honestly: multimodal physiological training, practitioner annotation (with its epistemological complications), and the more modest "augment don't replace" frame.

Further Reading (APA 7th): Gibson, Guenther, Merleau-Ponty, Perez-Marcos, Tevet et al., Varela/Thompson/Rosch.

The internal annotation explains the editorial positioning between the Apr 1 and Apr 9 explainers and flags the practitioner annotation paragraph as research-design-relevant for the platform.