“Much Ado About Nothing?”
For several years, artificial intelligence was presented as the ultimate solution to translation challenges. Fast, inexpensive, instantly available, it was supposed to render human translators and professional translation agencies obsolete. This promise was widely echoed in the media, in the marketing rhetoric of AI developers, and in the innovation strategies of businesses.
Yet, since 2024, a clear reversal has emerged: clients disappointed by machine translation are returning to specialised human providers. Not out of conservatism, but because they have seen, in practical terms, the limitations of AI when applied to translation.
Is this the end of the love affair with AI?
Machine Translation and AI: A Persistent Confusion
It is essential to reiterate a reality that is often misunderstood: an LLM-type AI does not “understand” a text. It does not reason, nor does it analyse the legal, commercial, or cultural intent of a document. It calculates linguistic probabilities based on existing corpora.
This works relatively well for:
- simple texts,
- generic content,
- isolated sentences,
- internal needs with no legal or reputational stakes.
But professional translation involves much more. It requires:
- terminological precision,
- consistency across lengthy documents,
- consideration of cultural context,
- accountability in the event of an error.
However, AI assumes no responsibility.
The Concrete Limitations of AI Translation
In practice, feedback is now recurring. Businesses are identifying:
- subtle but legally critical mistranslations,
- inconsistent translations from one paragraph to another,
- approximations with technical or regulatory terms,
- linguistically plausible but false “hallucinations”.
These errors are not always immediately apparent. They may come to light during a legal review, an international negotiation, an audit, or sometimes, too late.
It is precisely this silent and insidious nature that is problematic. A poor translation does not always fail conspicuously. It can undermine a contract, distort a strategic message, or cause lasting damage to a brand’s image.
Even AI Developers Acknowledge These Limitations
The main players in the sector, including OpenAI, explicitly acknowledge in their documentation that their models can produce incorrect or misleading information and should not be used for critical purposes without human validation.
Academic institutions and regulatory bodies take the same view. Several studies highlight that language models produce plausible but unguaranteed responses, especially in specialised fields such as law, finance, healthcare, or technical domains.
In other words, AI-powered machine translation is a tool, not a guarantee of reliability.
Much Ado About Nothing: An Overhyped Promise
The current situation strongly recalls Shakespeare’s phrase, Much Ado About Nothing. Much talk, much enthusiasm, but a more nuanced reality.
AI translation is not useless. It is effective in certain specific cases. But it has not replaced professional human translation, contrary to what was announced.
The market has not collapsed. It has refocused.
The Return to Professional Translation Agencies
For several months, agencies like Alpis have observed a clear phenomenon: clients who had switched en masse to AI are returning to human solutions. They are seeking:
- reliability,
- terminological consistency,
- sector-specific expertise,
- contractual responsibility,
- genuine data confidentiality.
This return is not ideological. It is pragmatic. Businesses have tested. They have assessed the risks. They have adjusted.
The True Future of Translation
The future of translation is not a battle between humans and artificial intelligence. It rests on a carefully managed complementarity.
AI can speed up certain processes, assist production, and facilitate pre-translation. But final validation, semantic judgement, and responsibility must remain human.
Because translating is not simply about transposing words.
It is about conveying meaning, law, strategy, and sometimes major financial stakes.
Conclusion
After the media frenzy, the market is returning to a more realistic perspective. AI-powered machine translation has found its place, but it has not replaced human expertise. The wave has passed, the profession endures.
In a complex international context, professional translation remains a lever for security, credibility, and performance. It is precisely in this area that specialised providers like Alpis continue to deliver value.
Sources:
OpenAI – Safety & Limitations Documentation
Stanford University – On the Dangers of Stochastic Parrots (Bender et al., 2021)
MIT Technology Review – Why AI Hallucinations Are a Feature, Not a Bug
Harvard Business Review – When AI Gets Language Wrong (2023–2024)
European Commission – Liability and AI-generated content