Should AI be Trusted as it Gets Better at Translation?

“Is There a Translator in This Text?” That question, followed by “How generative AI flips the entire notion of authorship on its head” framed a panel discussion moderated by Slator’s Head of Advisory, Esther Bond.

The discussion brought together Gabriel Fairman, Founder and CEO of Bureau Works, Wada’a Fahel, Senior Director, Engagement Marketing-Transcreation and Global Ops at Zendesk; and Katherine Elkins, Author, Researcher, Director and Professor at Kenyon College Digital Humanities Lab.

The panelists first talked about their own experience and interest in generative AI, which Fahel described as being “inevitable.” For localization, specifically, Fahel explained that AI can amplify and accelerate processes. Zendesk already had a good foundation with a network of vendors, resources like style guides and terminology, and Bureau Works as a tool. “In 10 months, we were able to roll out four AI use cases that reduced time to market and cost by an average of 30%,” she added.

Elkins explained that in her case, as a natural language processing (NLP) scientist and author, she gets to examine models from safety and performance perspectives. 

Fairman sees Bureau Works at the intersection of what Zendesk does with messaging and branding and what Kenyon does in academia: “With generative AI … it’s no longer a question of whether we use it or not … The question is, how are we going to shape our interaction with it?”

Iterative Pulverized Learning

Elaborating on that interaction, Fairman discussed what he called “the prompt response paradigm.” “We call it ‘iterative pulverized learning,’ where an engine is capable of deciphering human intentions at a very granular level. … It’s this continuous flow of very small interactions. That’s the direction that we see things going, at least when it comes to copywriting and authorship.”

To Fahel, the interaction is a source of efficiency. At Zendesk, the expectation is not to replace copywriters, authors, or translators. “AI is here to support. The efficiency is time to market. We are able to process shorter-form content, long-form content … Aim for progress, not for perfection. Because if you aim for perfection, you’re not going to get it,” she added.

Elkins addressed the question of what the responsibility in setting up translators and linguists for success looks like by stating that “people are going to have the human-in-the-loop … because it’s very easy as things become automated and we begin to trust AI to kind of just rubber stamp what’s coming out.” 

Fairman echoed the words of Farhel regarding the importance of having a very good foundation, something that reaches a new level with generative AI: “So if you have any deviations, for instance, let’s say your TM is full of gender bias … all your feeds are going to be full of gender bias, and it’s going to be hard to steer it … if your entire foundation is driving it in another direction,” added the Bureau Works CEO.

In closing, Fairman emphasized the need for people to experiment with AI and see what they can do with it. He added that although there are drawbacks, there are also opportunities in the short and the long term.