Translated's data says AI will match human translators by 2030

Whenever people debate the "technological singularity," it usually sounds like black hole physics: everyone agrees there's an event horizon out there, but nobody has a ruler to measure how close we actually are.

Rome-based translation company Translated decided to build that ruler. Since natural language is arguably one of the toughest hurdles for machines, they used the time human translators spend fixing AI output as a practical stopwatch.

Popular Mechanics detailed the company's findings based on tracking their AI's performance from 2014 to 2022. By analyzing over 2 billion professional post-edits across eight years, Translated mapped out a concrete timeline for when machine translation might match human quality.

The "Time to Edit" stopwatch

Translated calls its metric Time to Edit, or TTE. Basically, it calculates how many seconds a professional human editor needs to check and fix each word generated by an AI translation system.

The benchmark to beat is roughly 1 second per word, which is what it typically takes one professional translator to edit another human's draft.

Back in 2015, human editors spent about 3.5 seconds per word reviewing machine-translated suggestions. Today, that number has dropped to just 2 seconds.

As Translated CEO Marco Trombetti pointed out at a conference in Orlando and on podcasts, daily progress is so subtle you barely notice it. But zoom out over a decade, and the gap between machines and human pros is steadily shrinking.

If that curve holds, Translated projects its AI will perform as well as human translators by the end of the decade, if not sooner.

Fluent speech isn't necessarily true intelligence

Equating human-level translation with the singularity, however, raises plenty of eyebrows.

Mastering human language is certainly a frontier challenge, but generating hyper-accurate sentences doesn't automatically mean a system has achieved artificial general intelligence. Researchers still can't even agree on what "intelligence" actually means.

Even if true singularity remains elusive, reaching human parity in spoken and written translation is still a seismic milestone on its own.

Why it matters

For anyone working across borders, languages, or global content, Translated's data confirms what many already feel: machine translation is rapidly graduating from crude rough draft to production-ready output.

Translators won't vanish overnight, but the job is fundamentally shifting from drafting sentences to high-level cultural editing and liability checking.

We think Translated's approach is genuinely clever because it takes the slippery, overhyped concept of "singularity" and anchors it to a concrete, commercial metric backed by 2 billion real-world edits. It's far more grounded than synthetic benchmark scores.

That said, the trap is over-extrapolating. Mastering translation syntax and vocabulary doesn't mean AI can reason, plan, or think like a human. Wrapping a translation trendline in apocalyptic "singularity" packaging feels more like marketing than science, but the underlying fact remains: within just a few years, language barriers may effectively be history.