Three Myths about Generative AI and Interpreting
Generative AI tools such as ChatGPT (or DeepSeek, if you are Chinese) are rapidly changing the way we work and communicate. As an interpreter, I've been asked countless times about the impact of these tools on our profession. Will they replace us? If so, when? These questions have been swirling in my own mind too.
While I don't have definitive answers, I'd like to tackle the debate by outlining three often overlooked distinctions in these conversations.
Myth #1: Language = Written Text
Many people seem to forget that language exists in both spoken and written forms, and these forms differ significantly. Generative AI models are primarily "large language models," but perhaps a more accurate term should be "large written language models." They excel at processing and generating written text, producing impressive results in tasks like drafting emails or summarizing reports.
Machine translation tools like DeepL and Google Translate tend to produce high-quality translations of well-written, formal texts. But when faced with the loose, informal language of everyday conversation, their accuracy plummets. Deciphering intonation, stress, pauses, homophones, and the implied meaning behind words and cultural references all rely heavily on context. Spontaneous speech is often unstructured, with grammatical errors and even illogical jumps in thought, making it much more challenging for machines to understand. This distinction between spoken and written language is crucial for understanding the limitations of current AI tools in the context of interpreting.
Myth #2: Absolute Numbers or Relative Proportions?
Imagine I tell you that the number of interpreters will double in the future. You might conclude that the profession is thriving and not threatened by AI.
Now, imagine I tell you that 90% of all interpreting tasks will be handled by machines. You'd likely reach the opposite conclusion – that the profession is facing extinction.
Paradoxically, both scenarios can be true simultaneously. The overall demand for interpreting services might increase significantly, while the proportion of human involvement decreases. The difference between these two measures is all too often lost in the hyped discussion about the impact of AI on the interpreting profession - or any profession for that matter.
This 1953 ad compares slide rules to an IBM electronic calculator
To illustrate this, think of IBM's electronic calculator in 1953, which was advertised as being able to do the work of 150 engineers. Guess what? Technically, that claim was true. However, the nature of engineering work also evolved with the advent of personal computers, and the need for engineers increased exponentially in the subsequent years.
Think restaurants. Over the past few decades, machines have taken over a lot of the laborious work in food processing and preparation, right? But has the industry shrunk because of automation? Of course not. Thanks to higher disposable income, the sheer number of jobs in the restaurant ecosystem has grown significantly in the last few decades - not only men and women who wait tables, but also Uber Eats drivers, Yelp community managers, TikTok food influencers, and software engineers who make and maintain those apps.
I believe the same principle applies to interpreting: the sheer increase in global interactions—partially driven by technological advancement—will drive an unprecedented need for human-to-human communication globally. As a result of more international travel, trade and cultural exchange, the aggregate need for human interpreters will likely grow, not shrink.
Rony Gao providing consecutive interpretation for a delegation of Chinese engineers visiting Canada
Myth #3: Technological Maturity = Human Adoption
We often equate technological maturity with job displacement. However, widespread adoption of a technology is a separate process that usually entails enormous behavioral change.
For example, imagine a life-threatening medical emergency involving a beloved family member who is hospitalized in a foreign country. How would you prefer to talk to the medical staff? Would you be comfortable relying on AI to translate what the doctors tell you, or would you prefer a human interpreter?
Similarly, if your business were investing a half-billion dollars in an overseas venture, wouldn't you want a human interpreter to help handle the communication and negotiations before you are ready to make a deal?
When stakes are high enough, or matters personal enough, trusting a machine may be too huge of a leap. I'm afraid it may take a long time, or perhaps never, before people place full trust in AI for a meeting of such critical importance.
Conclusion
To assess the impact of AI on the interpreting profession, it is crucial to recognize
the unique characteristics of spoken language;
the difficulty in defining technological job displacement; and
the enduring value of human trust and connection.
I believe the future of language interpreting will continue to involve humans, but in a different way. My message to interpreters and interpreting students today is to embrace technology while honing our uniquely human skills – empathy, cultural awareness, and building trust and rapport with clients.
Rony Gao is a prize-winning conference interpreter, certified translator and communications consultant based in Toronto and serving clients worldwide. He is a member of AIIC and regularly provides consecutive and simultaneous interpretation services for the Government of Canada, international organizations and global leaders in business, technology and academia.
