'Tentacles squelching wetly': the human subtitle writers under threat from AI (www.theguardian.com)

AI-driven tools are putting pressure on the subtitles-for-the-deaf-and-hard-of-hearing (SDH) profession, but practitioners say automation is neither mature enough nor cheaper in practice. Freelance subtitlers grouped in the non-profit Subtle and industry veterans report that auto-transcription has improved slightly, but AI outputs still require extensive human correction and creative judgement. Viral mistranslations like Stranger Things’ “[Tentacles squelching wetly]” exposed the gaps; Netflix has shown internal subtitling work but won’t detail AI use, while some vendors (e.g., Red Bee Media) have begun experimenting with AI for clients. Meanwhile, humans often end up doing low-paid “quality control” on machine drafts, driving down already thin SDH rates and undermining livelihoods. The significance goes beyond jobs: SDH is an accessibility service that depends on nuanced decisions—what sounds to include, emotional tone, pacing, and cross-scene audio motifs that carry plot meaning. Subtitlers describe a creative workflow (not mere transcription) that translates sensory and emotional cues into concise on-screen text; current AI lacks the contextual reasoning and empathy to replicate that. The net implication for the AI/ML community is clear: improving subtitle automation requires models that handle long-range audiovisual context, pragmatic intent, and diverse user preferences, plus workflows that preserve fair pay and human oversight for accessibility-critical outputs.