
What happened
Todd Hughes, who trains language tutors at Rosetta Stone, told Fortune that AI can build vocabulary and aid comprehension, but proficiency still requires live human interaction. He says he's not ready to use AI for speaking practice.
Why it matters
Hughes draws a line between fluency (smooth, fast speech) and proficiency (understanding, finding words, and adapting when talk goes off script). Real conversation changes mid-flow, and he says AI doesn't grasp social cues yet.
What to watch
Duolingo chief product officer Cem Kansu told Fortune that real-time translation can help you understand someone now, but can't replace the cultural understanding and satisfaction of speaking yourself. The test is whether tools can teach tone, not just words.
WHO IT HITSLanguage learners who rely on apps and translation earbuds should hear this as a warning: tools can smooth an exchange or build vocabulary, but they don't prove you can handle an unscripted conversation. Corporate trainers and language-course providers may need to keep human practice in their programs.
Ask the AI about this article →
Summaries like this, in your inbox every morning.
Hughes's own language history shows why a single label rarely captures ability. He describes his Spanish as near-native, can read and understand French extremely well but makes mistakes speaking it, and finds Danish reading easy while listening and pronunciation remain a challenge. That mix is why he approaches proficiency through four core skills—reading, writing, listening, and speaking—rather than a single number.
He has worked with technology in language education since 1995, when he wrote his dissertation on using it to acquire Spanish vocabulary. That long view shapes his wariness: AI can act like a huge library that never closes, generating reading material and exercises, but he sees a gap when it comes to speaking practice and the social cues around words.
The stakes for learners hinge on purpose. Duolingo says many people learn English for high-stakes reasons like a better job or higher education, and for them an accurate translation still doesn't demonstrate proficiency. If tools keep improving at vocabulary and comprehension but remain weak on tone and unscripted exchange, the likely split is between travelers who need a quick bridge and learners whose goals require sustained human practice.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
Alphabet's second-quarter revenue grew 24% to $119.8 billion, while free cash flow turned negative $5.9 billio…

On September 8, Mad Money's Jim Cramer said Super Micro has accounting "irregularities," said he cannot recomm…

Google Research released TimesFM-3, a 330 million-parameter forecasting model trained on over one trillion dat…

Twenty-five Fields Medal winners, including Terence Tao, signed a joint statement warning that AI companies tr…

Meta is asking individual contributors in its Applied AI division whether they want to return to manager roles…

Strong demand for AI computing is causing a capacity squeeze across the passive-component industry, according…
