A unique metric suggests AI could achieve translation accuracy comparable to professional human editors by the end of this decade. This development offers a concrete perspective on the concept of technological singularity.
Singularity is the hypothetical moment when AI surpasses human intelligence and begins to transform society at an unprecedented pace. Predicting its arrival is extremely difficult because, like a black hole’s event horizon, its limits and effects remain largely unknown. Researchers are seeking tangible signs that AI is approaching this threshold, using measurable benchmarks to track progress.
Language, one of the most natural human abilities, is considered a critical test for AI. Achieving human-level accuracy in translation not only represents a technical milestone but also signals broader capabilities in artificial general intelligence. Success in this area could profoundly affect communication, business, and social interaction across the globe.
Measuring AI Progress With Time to Edit
A Rome-based translation company, Translated, developed a metric called Time to Edit (TTE) to assess AI’s performance in translation. According to Popular Mechanics, TTE calculates the time professional human editors take to correct AI-generated translations compared with human-produced ones. From 2014 to 2022, Translated analyzed over 2 billion post-edits to track the AI’s improvement.
Marco Trombetti, CEO of Translated, explained at a 2022 conference in Orlando, Florida, that language is the most natural and challenging aspect of human intelligence. “Nonetheless, the data Translated collected clearly shows that machines are not that far from closing the gap,” Trombetti said. This metric offers a novel, data-driven approach to estimating how close AI is to performing at human-level skill in language translation.
Closing the Gap Between Machine and Human Translators
The Time to Edit metric revealed significant progress in AI translation. In 2015, human editors required about 3.5 seconds to edit each word produced by machine translation. By 2022, this number had dropped to roughly 2 seconds per word, indicating steady improvements in accuracy. Trombetti highlighted that daily changes may seem subtle, but over several years, the cumulative progress is striking.
Although excelling at translation does not automatically signify full artificial general intelligence, it represents a measurable step toward human-level capability. Machines achieving human-level translation skills demonstrate that AI is moving closer to performing complex cognitive tasks traditionally associated with humans.
Potential Impact of Human-Level Translation AI
An AI capable of translating speech with human-level accuracy could have profound societal effects. It could facilitate global communication, reduce language barriers, and enhance cross-cultural understanding. While this achievement does not guarantee the arrival of full technological singularity, it shows that AI is already reaching milestones that could transform daily life and professional practices.
Trombetti emphasized that this accomplishment is notable regardless of whether AI reaches full artificial general intelligence. Effective translation systems can already impact education, international diplomacy, and commerce, providing a practical benchmark for evaluating AI’s evolution.