Part 2: Why Conference Interpreting is the Wrong Model for AI and The World

What are the problems and opportunities presented to the interpreting profession by the new large language models commonly referred to as “AI”? This essay follows on Part 1, which set out to establish a few key distinctions between the high-profile type of interpreting many people are familiar with and the regular everyday type of interpreting whose presence or absence can determine the life chances of minoritized language users. The primary audience for these essays are the conveners, prospective and future members of the Interpreting SAFE AI Task Force.

Below is an initial foray into engaging a careful selection of key points from Brad Smith’s (2019/2021) Tools and Weapons: The Promise and The Peril of the Digital Age.

While I’m sympathetic to a critique of this tech insider’s description of the ethics of building and deploying AI as “utter[ly] self-congratulatory and self-aggrandizing” (personal correspondence, June 29, 2023), I am more inclined to engage the ideas in this book as an “assembly of stories describing how Microsoft, often in partnership with other technology companies and governmental agencies, has dealt with the problems and opportunities digital technology creates” (Anthony Joseph Juben, review for the Association for Computing Machinery).

Maximizing the Strengths of Conference Style Professional Interpreters

The first thing that stands out are the predictions and statistics. For instance, five years ago Smith projected that “…the ability of computers to think means that we’ll see jobs at risk that involve brains…For example, given the rapidly improving ability of AI to translate human languages, it seems inevitable that the jobs of human interpreters will increasingly be at risk” (p. 270). Smith’s prediction is based on performance achievements from nearly a decade ago: “Microsoft’s Speech Recognition Tech is Officially as Accurate as Humans” (Dom Galeon, Futurism, October 20, 2016), and “Microsoft Researchers Achieve New Conversational Speech Milestone” (Xuedong Huang, Microsoft Research Blog, Microsoft, August 20, 2017).

The caveat for replacing human interpreters by AI depends upon which tasks can be automated and which cannot be automated (p. 271). In their speed competition with the KUDO language translation AI, Barry Slaughter Olsen and Walter Krochma concede that the machine is better with regard to information content, especially at faster speeds. Well. Speed and information content are discrete “tasks” that compose singular components of the wholistic function of communication. Unfortunately, both are dimensions of competition that interpreters will lose. This fact begs questions of why and how a focus on information content and the accompanying valuation of ‘speed’ have become so deeply ingrained in the interpreting field. Ethically, let’s ask who does speed serve and what does it achieve? Also ethically, if we exclusively prioritize information over the other aspects of communication — such as the unfolding relationship among humans in interaction with each other over the course of time — where does this lead?

Ethical Principles of AI — please meet the Ethical Principles of Community Interpreting!

According to Smith, responsible developers of AI are lobbying for regulation by the government in order to establish “meaningful human control” of AI (p. 237), particularly with regard to military actions but also with regard to protecting and guaranteeing the exercise of democratic capacity by current and future generations. Governmental regulation is a matter of values and political will to establish fairness, reliability and safety, privacy and security, inclusion/belonging/opportunity, transparency, and accountability (p. 230–232).

Can the Interpreting SAFE AI Task Force can generate policy guidelines and principles for when live human interpreting is required that sync within the ethical principles and policies framework already developed within the realm of responsible AI? If so, this will be a genuine and enduring contribution to human society.

Enter Community Interpreting and Social Models of Interpreting

Smith concedes that “there are certain tasks that AI likely won’t perform well. Many of these involve soft skills such as collaboration with other people, which will remain fundamental in organizations large and small… AI…is unlikely to excel in providing… empathy…” (p. 272). Community interpreting requires soft skills, such as those that manifest along three primary axes of interpretive action in the important theory of rolespace (Lee & Llewellyn-Jones, early slidedeck, 2014). This is the first theoretical model of interpreting that provides a paradigmatic framework for the social aspects of interpreting. Unlike the more visible model of conference and diplomatic interpreting that is typically promoted by media (like @Wired), community interpreters have always already been in the nitty-gritty interpersonal-intercultural during interpreted interaction.

According to Olsen, in his comments about the speed competition with KUDO, facilitating or mediating mutual understanding and connection is ostensibly off limits in the diplomatic, business and academic settings of conference interpreting — whether or not participants achieve comprehension is not the interpreters’ concern. This attitude of ‘hands-off’ interpreting has long been challenged by — in particular — Deaf participants and professionals within community interpreting, who have called it a machine model (Baker-Shenk, 1986 and 2014). Further, the myth of neutrality and/or objectivity that rationalizes laissez-faire interpreting has been most assertively articulated by academics within sign language interpreting (e.g., Metzger 1999), the precepts of this pragmatic realism have been a constant minority theme in interpreting studies (Roy, 2000; Wadensjö, 1999).

Prescribing a Sustainable and Resilient Future of Interpreting with AI

Smith elaborates that “…ethical views ultimately rest on broader human rights and philosophical foundations. This makes it essential to connect these topics to an understanding of the world’s diverse cultures, as well as the varying laws and regulatory approaches this diversity creates“ (p. 238).

Here is where interpreters as a profession — all of us together, conference style and community style — have the organic skill that AI lacks. As a field, we are positioned by virtue of decades of lived professional practice “to promote a thoughtful, respectful, and inclusive…conversation” (Smith, 2021, p. 241) about the pragmatic features of regulation and policy that can uphold the dignity and life chances of billions of people communicating with each other using different languages.

Let’s create an exemplary process according to the principles we want to invite others to follow.