■ HIGH RISK ■ Media & Communication
For routine conversations, AI interpretation is already good enough and effectively free — that market is gone or going. Court, medical, and diplomatic interpreting survive on accuracy stakes and accountability, but they're a smaller island than the profession's current size.
“Real-time translation earbuds don't charge $150 an hour.”
Our AI replacement risk score — how we score jobs
Interpreters convert speech between languages live — simultaneously from a conference booth, consecutively across a deposition table, or over video for a hospital that needs Tagalog at 3 a.m. The work is cognitively brutal: holding a speaker's meaning in working memory while rendering the previous sentence, managing dialect, idiom, and emotion in real time. Conference interpreters work in pairs and swap every half hour because the cognitive load is unsustainable. Much of the actual market, though, is humbler: business meetings, immigration interviews, customer calls, and community interpreting billed hourly through agencies.
Speech-to-speech AI has improved at a pace that should alarm anyone billing by the hour. Real-time translation is now built into earbuds, phones, and video-conference platforms; for a routine business call between patient parties, machine output is genuinely usable and costs nothing. That routine tier — tourist help, basic business meetings, low-stakes customer interactions — is evaporating first, and it's the volume tier that feeds agencies and part-time interpreters. Remote video interpreting platforms, which already commoditized the work, are the natural delivery channel for replacing it entirely.
The resistant core is where errors have consequences and accountability must attach to a person. Courts require certified interpreters because a mistranslated answer can shift a verdict; medical interpreting errors kill people, and hospitals carry the liability; diplomacy and high-stakes negotiation involve nuance, register, and discretion no model is trusted with. AI also still fumbles code-switching, heavy accents, culturally loaded idiom, and the interpreter's quiet judgment about when a literal rendering would mislead. Our risk score is high because the profession's revenue pyramid has a wide, automatable base — even though its certified apex may actually gain value as the generalist tier disappears.
Automatability: our editorial assessment of current and near-term AI capability
The low-stakes tier is disappearing now — translation earbuds and meeting-platform features are already substituting for hired interpreters in routine settings, and agency volume work erodes sharply before 2030. Certified court and medical interpreting hold through the decade on regulatory and liability grounds, though even there AI handles overflow and off-hours demand. The profession compresses toward its credentialed, high-stakes core.
For casual and routine use, they already are — travel, basic business calls, everyday customer interactions. For courtrooms, hospitals, and negotiations, no: those settings demand certified accuracy, confidentiality, and a human accountable for errors. The honest picture is a shrinking middle — the hourly generalist market — with a durable certified core above it.
Only with a direct route to the protected tier. Training years for general business interpreting means competing with free software that improves monthly. Training toward court certification, medical interpreting, or a rare language pair is a different bet — those markets are gated by credentials and liability, and may actually tighten as generalists exit the field.
Because the tolerance for error is effectively zero and someone must be accountable. Legal proceedings involve idiom, evasion, register shifts, and answers where a subtle mistranslation changes meaning under oath. Courts require certified interpreters partly so errors are attributable and challengeable. No jurisdiction is eager to let an appeal turn on what a speech model guessed.
The routine tier is changing now — real-time translation is embedded in phones, earbuds, and video platforms people already own, so substitution requires no purchase decision at all. Agency and volume work will erode visibly this decade. Certified legal and medical work is on a much slower clock, governed by regulation and institutional risk tolerance rather than model quality.