Artificial intelligence promises to interpret your events for a fraction of the price. But when the WHO tested the best platform on the market, only one interpretation out of 90 passed. Here are five reasons the machine still can’t replace a human interpreter.
Las Vegas has earned its place among the world’s great business capitals. In 2024 alone, the city drew nearly six million people to its conventions and events, generating roughly $16 billion in economic impact. It’s hard to picture a crowd that size, so here’s a comparison: that’s about the same as the total number of tourists who visit all of Colombia in a year—just over 6.5 million people.
As a corporate and event destination, Las Vegas pulls in millions of executives, speakers, negotiators, and attendees from a huge range of countries and cultures—and that’s exactly where the language barrier comes in. How do you get everyone in the room to understand each other perfectly, and in real time? There’s really only one dependable way to pull it off: professional, certified interpretation—the kind Language Up provides—for the event and the people running it.
And yet, plenty of people still ask: why pay for human interpreters when any smartphone already “translates on its own” thanks to artificial intelligence?
It’s a fair question. We’re in the middle of the ChatGPT, Gemini, and Claude revolution, and these tools, among others, are now part of everyday life. But handing professional interpretation over to AI is a risky bet. Case in point: the National Association of Judiciary Interpreters and Translators (NAJIT) titled its 2026 annual conference “Beyond AI: The Irreplaceable Human Element.”
In 2025, the interpretation team at the World Health Organization (WHO) put one of the highest-rated AI interpretation platforms on the market through a rigorous test across all six of its official languages. The results were damning: out of 90 interpretations, only one earned a passing grade. The average quality score came in at 46%, and every single one contained an error serious enough to put a reputation at risk.
Here’s why the machine still isn’t ready to sit in the interpreter’s booth—backed by the data, and by the field experience of Ximena Chica, director of Language Up.
True simultaneity
A human interpreter speaks almost at the same moment as the speaker, with a lag of just a couple of seconds. AI doesn’t work that way. It runs through three steps—it turns speech into text, translates the text, and then turns it back into speech—and to do all that, it has to wait for the sentence to end. The result is closer to consecutive interpretation than to simultaneous.
“AI has to wait for the sentence to finish before it can translate. It’s always two or three sentences behind. That’s not simultaneous interpretation.” — Ximena Chica
That delay isn’t just cosmetic. In the WHO test, the lag stretched past 32 seconds, compared with a maximum of about 5 seconds for a human interpreter. As a result, the last sentences of each remark often didn’t get interpreted at all.
Picture that playing out in an hour-long business meeting where two sides are negotiating the price of a shipment of tires for the coming year. The buyer makes an offer and waits for a response—and if the other side has to wait for a lagging block of text to catch up two sentences later, the conversation stalls out. No back-and-forth, no chemistry, no close. And if a presenter is walking through a slide full of figures while a digital interpreter runs three sentences behind, by the time the listener finally “gets” the number, everyone’s already on the next slide.
Register, culture, and regional speech
Interpreting isn’t swapping words from one language to another—it’s carrying a message that’s shaped by who’s speaking and how. The register—the tone, the vocabulary, the cultural background—of an executive with a PhD is not the register of a technician on the shop floor. Someone from northern Mexico doesn’t use the same expressions as someone from Argentina. A single cultural reference can be the beating heart of a great speech, or a misunderstanding big enough to sink the deal.
“A machine still can’t tell registers apart. It can’t distinguish someone from a Latin American subculture from an English speaker, for example; it doesn’t pick up on regionalisms, or on ethnic and cultural differences. Understanding those nuances well enough to carry the message faithfully is exactly what we do.” — Ximena Chica
The WHO study drives this home almost to the point of caricature. When a speaker from Bangladesh closed with “Joy Bangla”—a national slogan that means “Hail Bengal”—the AI mistook it for the name of the session’s chair and, in the languages that mark gender, treated him as a woman. Proper nouns met the same fate: “Brunei Darussalam” came out as “the brunette Russel,” Greece became “Chris,” and Haiti became “Heidy.” A human interpreter who doesn’t recognize a phrase repeats it in the original language, asks for clarification, or carefully leaves it out. The machine just makes something up.
Humor and emotional weight
The best speakers don’t inform—they persuade. And they do it with humor, with emphasis, with pauses, with emotion.
“A machine simply can’t carry humor. A good interpreter is also a bit of an actor: if the speaker is angry, you sound angry; if they’re sad, you sound sad. Good interpreters are theatrical. A machine is flat—it talks like a GPS.” — Ximena Chica
The WHO evaluators described the AI’s voice as extremely monotonous and expressionless—impossible to follow for more than a few minutes. Understanding a text is one thing. Getting a room of 300 people to laugh at the joke, feel the story, and applaud the closing line is something else entirely. That’s the difference between an event that merely “works” and one that actually connects.
Figures, data, and specialized terminology
This is where AI goes from awkward to dangerous. Every industry has its own language inside the language: in a commercial contract, the Spanish word “capacidad” doesn’t line up with “capacity.” In medicine, in insurance, in the automotive world, there are so many technical terms for such precise concepts that one wrong word can flip the entire meaning of a clause or a diagnosis.
“You have to train the machine on the specific vocabulary of each context, and that takes an interpreter. For a conference of mechanics, or one built around a particular disease, nobody is going to fully train an AI model before every presentation. So you’re stuck with a tool that’s generic and imprecise.” — Ximena Chica
In the WHO test, numbers came out wrong—especially the ones with a lot of zeros, and even dates. A reduction “of about 70%” turned into “to about 70%.” “Transmission of polio” got confused with something about “transportation.” In a boardroom or an operating room, mistakes like these aren’t anecdotes—they’re lawsuits.
Accountability: who do you sue when the machine gets it wrong?
Here’s something almost nobody factors in when they’re comparing prices: a professional interpreter is responsible for what they say. In court, for example, they’re legally accountable for their work.
“An interpreter can be held legally responsible for a mistake. A machine can’t be held responsible for anything. That’s why lots of courts won’t even accept AI’s written translations.” — Ximena Chica
The U.S. court system backs her up. The National Center for State Courts (NCSC) doesn’t mince words: AI should not replace human interpreters for real-time spoken interpretation in court proceedings, given the high risk of errors around context and nuance. NAJIT requires every AI-generated translation to be reviewed by a human translator. And the American Translators Association puts it bluntly: when AI makes the mistake that leads to a mistrial, a wrongful conviction, or harm in a hospital, the company that sold the software will already have disclaimed all liability. The bill, on the other hand, lands on someone very real.
AI isn’t the enemy—it’s just the wrong tool for this job. Let’s be honest: AI is genuinely useful in our line of work—for written translation, say, or as a prep tool ahead of an assignment. But it’s the wrong tool to hand a live speaker over to. That still takes human judgment you can’t replace.
“With AI, I can put together a glossary in minutes. If I’ve got a car brand conference, I ask it for an English–Portuguese glossary and it gives me great material. It would be crazy not to take advantage of this technology for things like that. But it’s a help—not a tool that can replace us.” — Ximena Chica
That’s exactly what separates a professional event from a gamble. AI translates words. An interpreter translates intentions, reads the room, lands the humor, guards the numbers, and puts their name behind every sentence. When there’s a contract, a reputation, or an hour you can’t get back on the line, that difference is everything.
Language Up provides professional simultaneous interpretation for corporate events, trade shows, and business meetings in Las Vegas. Have a multilingual event on the calendar? Let’stalk.
