Two jobs under one title
Asking whether AI will replace translators treats two different jobs as one. Written translation is text in, text out, and machine systems are fast and cheap at it. Live interpreting is a person in a room, reading tone, accents, overlapping speech and what a speaker actually meant. The task mix above splits along that line, and so does the pressure on pay.
Routine written work is where erosion shows first. Translating a product manual, a standard contract clause or a help article can now start as machine output that a person checks. Compiling glossaries and checking technical terms is also partly automatable, because the raw material is already in text form. Agencies buying that work at volume feel the cost gap, and entry-level document jobs are the ones that thin out.
The other half of the work resists the same treatment. Rendering spoken statements in real time, relaying a speaker’s intent rather than their literal words, and interpreting in court, hospital or immigration settings all carry consequences for a mistake. Someone qualified has to stand behind the meaning. Sign language interpreting is visual and spatial, not a stream of text, which puts it further again from what text models do well.
What AI does, what it assists with, and what people keep
Share of task time AI can handle without a person: 24%. That group is the predictable written work, such as translating documents between languages and drafting first-pass terminology lists from existing material.
Share where AI assists but a person stays in charge: 53%. Here the pattern is post-editing: a model produces a draft, and the translator fixes register, fixes false friends and adapts references so the text lands with the audience it is written for.
Share that still needs a person: 23%. Simultaneous interpreting in live settings sits here, along with sign language work and any assignment where a certified interpreter is required to take responsibility for accuracy. Our overall coverage figure for this job is 46 out of 100, and how coverage is measured explains what that counts.
The evidence is thin, and that matters
Evidence grade for quality parity: D. In our scale, that is the grade for work with no direct, published test of AI against qualified people doing this job under real conditions. So we publish no parity number for interpreters and translators. Claiming one would be guessing.
What would settle it is specific: blind comparisons of certified court and medical interpreters against live speech translation on real assignments; error studies that count consequential meaning errors rather than fluency ratings; and audits of post-edited translation quality at agency scale. Until that exists, the honest reading is that machine output is clearly usable for some text and unproven where accuracy has legal or clinical weight. Our quality parity method sets out the grading, and the wider scoring method covers the rest.
When the balance could shift
Most likely between 2036 and 2046 (8 in 10 of our scenarios). The replacement-year method explains what that window is based on.
Two things could pull it earlier. First, no hardware is needed: the robotics tier for this job is none, so adoption only depends on software and a device someone already owns. Second, the cost gap is wide, and the tools are cheap enough that buyers can try them on low-risk content without a procurement fight.
Two things hold it back. Certification and liability rules in courts, hospitals and government settings specify a qualified human interpreter, and those rules change slowly. And the hard inputs stay hard: strong accents, crosstalk, regional dialects, low-resource languages and sign language are all weak spots for systems trained mostly on clean text and clean audio.
What to do: If you work mainly on bulk document translation, move deliberately toward interpreting, certification or review work before the rate pressure reaches your niche.
How to stay needed in this job
Lean into the parts of the work that sit in the needs-a-person group. First, live interpreting where someone must answer for accuracy, especially legal, medical and public-service assignments. Second, cultural adaptation, where the job is to make a message work for an audience rather than to mirror its words. Third, final review and sign-off on translated material, including terminology that a client has to defend.
Two skills raise your floor. One is disciplined post-editing: knowing fast where machine drafts fail, and pricing that work honestly. The other is a credential in a regulated setting, because certification is what makes a human requirement enforceable rather than optional.
If you are weighing a move, these related jobs sit close to this one: court reporters and simultaneous captioners, editors and technical writers. For context on the wider group, see the media and communication workers family and the information sector. You can also put this job beside another on the job comparison tool, or see where written-language work falls on our list of jobs most exposed to AI.
For scale: the US employed about 52,060 interpreters and translators at a median wage of $60,170, with projected employment growth of 2% from 2025 to 2035 (BLS, 2025). That is slow growth in a job where the work inside the title is being reshuffled faster than the headcount moves.