Simultaneous Interpreting in a Technological World: Why AI Will Never Replace Human Interpreters:
The Inspiration Behind it all…
As I mentioned in my last article, my experience sitting in some Simultaneous Interpreting (SI) Lectures at CUHK really opened my eyes to a whole new area of language studies that I wasn’t aware of before. Being able to try out (and struggle with!) some interpreting activities myself, visit a lab at PolyU looking at the potential of AI subtitles in interpreting work, and discuss the field of SI and AI influences with an industry researcher, all of this enticed me to learn more about the future of this profession in an increasingly technological and AI-driven world. From initial curiosity to now, everything has amalgamated into the creation of this article, which is a fusion of many different avenues of information and opinion.
This article began as an interview piece with Alex, a Master’s student at CUHK, where I planned on diving into the field of SI in general and the linguistic ties within the profession. However, my interest in the AI aspect of it all evolved this article into one of research (while retaining interview elements) on the contested future of AI in the SI field.
To begin today’s discussion, I’d like to start with a quote from a decade ago by Arthur Goldhammer (2016) in an article he wrote about machine translation: “...intelligence is incredibly complex. To be intelligent is not merely to be capable of inferring logically from rules or statistically from regularities. Before that, one has to know which rules are applicable, an art requiring awareness of sensitivity to situation.” Even a decade ago, Goldhammer pointed out the key distinction between artificial intelligence and human intelligence, which is that humans have this “awareness of sensitivity to situation”. Thus, in this specific instance of SI, and all other areas of life, AI will not replace human intelligence entirely, but only supplement it when we consciously use these AI tools for the better.
What is Simultaneous Interpreting?
Before I consider any of the big questions that this article tackles, I first need to explain the subject of Simultaneous Interpreting itself. If you were anything like me three months ago, you’d probably also have absolutely no idea what simultaneous interpreting really is beyond the people you see on TV trailing politicians to make their conversations flow. Herein below is a comprehensive explanation of what SI is from my personal experience and research.
If I were to describe simultaneous interpreting, I think the phrase ‘linguistic multitasking’ would be the best way to encapsulate this complex field. The interpreter, after completing weeks of extensive research preparation for the specific task at hand, simultaneously needs to “listen, comprehend, syntactically restructure, and produce speech in the target language while the source speech stream continues uninterrupted, typically for periods of 20–30 minutes in professional conference settings” (Milcu, 2026, p.170, as cited in Gile, 1995, 2009; Pöchhacker, 2016). All of these things happening at the same time in one’s brain demand extreme levels of cognitive load that are incomprehensible to someone outside the field who has no idea of the level of difficulty that this all takes. From my interview with Alex and talking to other SI students as well, the common misconception that being bilingual means one can interpret was dispelled. They all described to me how being bilingual was barely the first step and did not cover even the basic skill set required in SI.
Moreover, in SI, it is not just the words that need to be conveyed, but also the non-verbal meaning of paralinguistic cues like tone, pitch, volume, hand gestures, etc. SI requires becoming “both the speaker and the listener”, as said by Alex, in the sense that one has to express exactly what the speaker is saying, not only in words, but in nuance as well, so that this is all comprehensibly relayed to the listener.
To have to do all of these tasks at the same time in such highly pressurising environments like international diplomatic meetings, global conferences, high-stakes legal proceedings and healthcare scenarios, it is no wonder why it is deemed “the most complex of human cognitive/linguistic activities” (Ahmed, 2022, p.336, as cited in Zhang).
Defining Key Terms:
Now that there is a fundamental understanding and hopefully, appreciation for the work of interpreters, I will now define some key terms required in the discussion of technology and especially AI in the SI field.
The two terms I will use in this article are Machine Interpreting (MI) and Computer-assisted interpreting (CAI). As a mere high school student studying this subject, to avoid inaccurately conveying what these key terms mean, I will be using the definitions provided by Lu and Fantinuoli (2025) in their paper “Machine and Computer-assisted Interpreting: Innovations in and Implications for Interpreting Practice, Pedagogy and Research”:
Machine Interpreting: “Machine interpreting…refers to the practice, process, or product of real-time automatic or automated speech translation by a computerised cascade system composed of subsystems...Machine interpreting is distinct from all other forms of speech-to-speech translation in its emphasis on immediacy, meaning the translated message is delivered instantaneously and cannot be revised post-delivery, mirroring the dynamics of human interpretation.” (p.2)
Computer-assisted Interpreting: “Computer-assisted interpreting (CAI) refers to the use of digital tools – such as desktop software, mobile applications, and, more recently, AI-driven systems that support human interpreters at various stages of their work – from preparation to performance and post-task analysis.” (p.2)
How can AI and technological tools aid SI?
With these definitions of key terms and a foundational understanding of SI laid down, I will now begin the debate of whether AI will replace the field of SI at all. In recent years, there has been a lot more research published on the potential of both MI and CAI aiding multiple areas of SI, from real-time interpreting to the incorporation of these tools in SI pedagogy, with there being three key motivations behind this push.
From Ahmed (2022, as cited in Fantinuoli 2019), these reasons are as follows. The first reason is the “anthropological drive” (p.327), which is to reduce the intense cognitive load of interpreting through the help of MI and CAI. The second is the “economic drive” (p.327), which is to increase productivity and optimise work while reducing the costs of interpreting. The third is, to me, the most interesting, which is the “socio-psychological factor”, where “a technology-obsessed society pushes interpreters to accept change” (p.327). I will touch upon this factor in the final part of this essay.
These three motivations clearly highlight why research is not only looking to further understand, but to implement MI and CAI into multiple aspects of SI due to the plethora of benefits for the interpreter and the field at large in doing so. To this, I have to agree. While there are definite drawbacks to AI tools as we know them, which I will touch upon later, along with deeper ethical implications of using AI tools, it is impossible to ignore the positive impacts that MI and CAI currently have and could have on the field.
The first key benefit of MI is its ability to reduce cognitive load. I mentioned earlier that interpreters undergo extensive preparation before each task or conference. To help understand this process a bit better, I asked Alex what is considered necessary preparation before entering the interpreting booth.
She told me that first, a background check on each speaker is required, including their profession, nationality, and motivation for their talk. Research on the speech style of the speaker also needs to be done by listening to previous speeches to detect their accent, speaking pace, verbal stutters, etc., to familiarise themselves and aid their understanding of the speaker when in the interpreting booth. Then, Alex also spoke about the importance of building a glossary of key words specific to the subject matter that they will be interpreting. This glossary is vital to help form a skeleton mind map on the topic and to form logical links between these key words to become experts in this field as well. Given the large number of tasks required even before interpreting begins, CAI tools can be used to prepare these materials, boosting performance and reducing the taxing workload expected of interpreters.
However, even during interpreting performances, CAI and especially MI can be helpful for the interpreter in different ways. Novel CAI technologies can aid real-time performance, especially in more technical aspects like specific jargon, proper nouns or numbers (Lu & Fantinuoli, 2025). For human interpreters, jargon and numbers are considered to be the most difficult challenge due to the lack of “conceptual representation”, as they are difficult to predict from the context of surrounding speech (Ahmed, 2022, as cited in Timarová, 2012 and Seeber, 2015). Therefore, technology like CAI can aid interpreters in this sense to avoid errors when interpreting highly dense information.
Furthermore, in the study that I had the pleasure of sitting in on and observing, the research was on the effect of MI subtitles in aiding the interpreter. Much like what has been illustrated in the paragraph above, the MI subtitles, whether in the source or target language, can really help with certain key terms that may be complex and with the speed of information being transmitted within the speech, highlighting how AI tools can positively aid SI to reduce cognitive load and optimise interpreter performance.
Taken together, it is clear that AI tools like MI and CAI can be extremely helpful in practically all aspects of the interpreting process and can improve the training experience itself. For example, one really insightful detail that I learned from Alex was that AI tools can speed up the reflection process, where AI, based on the audio file of your interpreting, gives pointers and tips for improvement, which saves her lots of time (and the shame of listening to her own voice) while still achieving the same results. This increase in efficiency can have numerous positive benefits for the field at large. As AI starts to do more and more of the specialised and taxing tasks mentioned above (e.g., research, memorising numbers and key terms), SI students can focus on developing the transferable skills of the interpreting profession (e.g., multitasking, critical thinking, resilience, and communication). This can be ground-breaking for the field, as many may find interpreting to be less relevant, but will realise it is actually a deeply beneficial skill to train cognitive function and concentration, amongst other skills like bilingual communication, that are wholly relevant to many fields and professions, making one a better candidate for a whole host of other jobs.
Why AI will never replace human interpreters:
While MI and CAI can aid human interpreters, their overwhelming limitations prevent them from taking over the interpreting profession entirely, and many of these limitations are largely unrectifiable in the near future.
The first most obvious limitation of MI and CAI is their inability to register and convey non-verbal cues. In many of the papers I read regarding the abilities of AI interpreting, one of the largest pitfalls of tools like MI is the inability to interpret beyond the literal words spoken, losing the nuance, emotion and general quirks of the speaker in the process. One of the clearest reasons behind this that I read was the lack of Theory of Mind (the cognitive capacity to attribute mental states) behind the AI tool, which prevents the MI from applying contextual cues and nuances in the interpretation (Milcu, 2026, as cited in Bender et al., 2021; Marcus & Davis, 2019).
Moreover, another common feature of AI is its overwhelmingly flattering and agreeable behaviour when responding to questions of advice, which poses another limitation of the technology. “The tendency of AI-based large language models to excessively agree with, flatter, or validate users” has posed numerous problems of “sycophancy”(Cheng et al., 2026), especially as more and more people turn to AI chatbots with vulnerable problems seeking advice. When applied to simultaneous interpreting, this excessive kindness affects interpreting outcomes, as human communication, especially in settings where interpreting is found (e.g. conferences, debates, panels, court cases, etc.), is prone to disagreement and conflict. Human speakers express respect and patience conditionally, unlike AI chatbots that seem to dish out incessant compliments and kindness. Even in situations of constructive criticism and productive friction that would not cross our minds as disagreements, if MI smooths over these parts of conversation that arise naturally by changing the strength of voice, tone or word choices, this is not faithful to the emotional conversation of the speakers and leaves gaps in the interpretation of the listener, as they would only hear the poised and proper version that the MI interprets, rather than the actual emotional discussion taking place between the two individuals.
Aside from this, going back to non-verbal cues, I would like to specifically highlight the effect of pauses in speech. Gaps of silence are often overlooked, but they are fundamental in helping guide the listener through one’s speech and making one’s message clear. The inability to add pauses in the correct spaces is something that was noted by Alex in our interview as well as by two professional interpreters, Barry Slaughter Olsen and Walter Krochma, who describe the unnatural pauses in MI as creating “islands” of speech rather than flowing sentences. This places additional cognitive load on the listener to try and piece together these “islands” of speech, which is something interpreters always try to avoid.
Another aspect of language that has proven to be quite a challenge for AI tools is culture-specific terms like irony, idioms, neologisms and other more context-dependent aspects of language, or, as I like to think of it, the special flair of each language that reflects the culture they stem from. However, I would like to clarify here that AI tools in translation and interpretation have improved drastically when translating these more complex and specific aspects of language. Kalaš and Lipták (2025) tested AI translation of 100 idioms from English into Slovak both with and without context, and compared the outputs with those of human interpreters. Without context, 91% of the idioms were correctly translated, while with context, accuracy was 77%. While the accuracy in both instances wasn’t absolutely perfect, it would be erroneous to assert that AI translation tools are completely unable to translate idioms and, by extension, other culture-specific terms.
In spite of this, a more prominent obstacle that AI interpreting faces nowadays is less about the specific words or phrases themselves, but rather when the speaker makes mistakes or speaks unclearly. It is very natural for humans to make mistakes when speaking, whether this is through incorrect word choices, facts or statistics. This, along with unclear production of speech (e.g. fast-paced speaking, mumbling or coughing while speaking, etc.), makes interpreting an even more demanding task than it already is. For human interpreters who are equipped to structure their interpretation despite these inevitable mistakes, this task can be very challenging. Thus, for AI tools that rely on a clear and correct initial speech input to begin the processing of data, when the first input is unclear, the entire processing pipeline is compromised, rendering the output interpretation incorrect as a result. Since mistakes and unclear speech are fairly common, the inability to interpret precisely despite these mistakes renders AI interpretation much more fallible compared to human interpreters, emphasising why AI cannot interpret independently of human interpreters, but rather should be used by human interpreters as a helpful aid.
Ethical Implications of relying on MI and AI Interpreting:
Nevertheless, greater than the drawbacks of MI and AI interpreting in general, there are also concerning ethical implications involved with machine interpretation that must be considered when looking at the potential of such technologies in the industry.
Firstly, there are issues of accountability and privacy when allowing MI into interpreting spheres. The issue of accountability for human interpreting is not an issue at all because it is clear that the responsibility lies with the interpreter for inevitable mistakes or misinterpretations. Conversely, for machines, this is more of a grey area because there is no binding ethical code regulating the responsibility for errors made by these AI tools. In environments where mistakes in interpretation can lead to grave consequences like governmental misunderstandings, misdiagnoses, etc., a lack of accountability for the machine interpreter can create a domino effect of complications.
Moreover, the issue of privacy is a vital one. As mentioned above, interpreters are often interpreting in situations that must be kept confidential to allow subjects to share sensitive information openly. Situations like legal testimonies, business secrets, merits hearings, etc. must be kept confidential for the safety of the speaker. However, AI tools are managed by third-party software, which adds another layer of entities having access to this sensitive information, raising many privacy concerns, especially as AI companies have the motive to use this sensitive information to improve their datasets and future interpreting performance using this data.
The importance of privacy in many instances of interpretation reduces the sample size of information that the datasets have access to, which leads to a different problem of cultural homogenisation and other biases. AI tools work with the data they receive, and since the data pool is so small, this intensifies problems of bias even more. English as the lingua franca of the world is a major source and target language for interpretation, and with many examples of interpretation data being in the widely spoken languages of the world (e.g. French, Chinese, Spanish, etc.), the inequality in resources remains, and hence, the disparity in the quality of MI is also heightened (Ahmed, 2025). Milcu (2026) points out that there is a much higher performance for high-resource language pairs (e.g. English-Chinese, English-Spanish) than for low-resource language pairs like English-Romanian. Due to the lack of resource materials available, low-resource language pairs are often much lower in quality, and AI tools in these languages may not be usable at all. I find that the following quote succinctly sums up the dangerous implications of this discrepancy in language resources: “The commercial logic of AI development…entrench[es] a technological stratification with troubling implications for access to quality interpreting services in lower-resource linguistic communities.” (Milcu, 2026, p. 176). Furthermore, there are also clear disparities against accents and dialectal variations of languages, which means that “populations most likely to depend on interpreting services for access to justice and healthcare receive the worst-performing AI interpretation.” (Milcu, 2026, p.179, as cited in Nekoto et al., 2020; Tatman & Kasten, 2017)
The problems with AI interpretation extend far beyond the technology itself, but also in exacerbating inequalities that are already vastly prevalent in our world, including gender bias, as documented biases like masculinisation of professional role translations and inaccurate pronouns in gender-neutral source languages entrench misogynistic language choices, which can have great impacts in the interpreting booth when a speaker’s identity and social position are significant (Sun et al., 2021). However, these biases are becoming increasingly difficult to detect and change as AI has evolved to take on increasingly vast methods of information retrieval (training data, model architecture, human feedback, outsourced information retrieval, etc.). Hence, tracing sources of bias becomes more difficult, and this makes the very nature of these tools complicit in prejudice and far from the impartial role that is expected from an interpreter (Jiang, 2013).
Therefore, with AI tools that are so reliant on human influence, input and feedback, it’s hard to imagine a profession like interpreting, which acts to bridge the gap in human communication, adopting these tools entirely to replace human interpreters. When the ethical role of interpreters is already a highly debated subject on whether the moral code of the interpreter should or should not influence their interpreting (Jiang, 2013), I find it difficult to reconcile with the total inclusion of AI interpreting tools as this would induce a wide range of other ethical dilemmas that are also constantly being pondered, like the sentience of AI, whether it has a ‘personality’, etc. In my view, the debate on whether AI interpreting should be completely and entirely adopted can be answered by one simple question: Who can understand the uniquely complex and nuanced contexts, meanings and emotions of human speech if not human interpreters? For me, it is no one. Humans will always trump AI or machines in the understanding of the complicated nuances of human speech and hence, should retain the reins to the interpreting field rather than AI taking them instead.
What does the future of interpreting look like?
With all this being said, the reality of the world is pushing technological advancements and integration in all areas and sectors, as illustrated by the ‘socio-psychological factor’ for technological and AI involvement in interpreting mentioned above. However, it is clear that this integration does not mean replacement, but involvement and evolution.
Practically every statistic, webinar, and video says that language-related careers are the most likely to be replaced and fully automated by AI out of all the careers out there. While AI involvement is highly likely in careers like interpreting, I would argue that, like numerous other career paths out there, the influence of AI is not to dominate the interpreting field, but to change it and potentially improve it for the better.
Every article I read while researching for this piece, in my interview with Alex and in conversations with the researcher, all had, by and large, the same response: that AI should be working alongside human interpreting instead of replacing it entirely or being completely ignored. In Milcu’s 2026 paper, she suggests the ‘augmentation paradigm’, which is something I found myself agreeing with after concluding the research for this article. She suggests that AI should be embedded into teaching rather than superseding human capabilities. This aligns completely with everything I’ve read and heard so far. Alex mentioned to me how AI has helped her save significant amounts of time in the reflection process after interpreting, as it is able to provide instantaneous comments on her performance and tips to help her improve. CAI tools are also reported to help boost numerous areas of interpreting performance, like motivation, anxiety reduction and self-paced practice, etc. (Lu & Fantinuoli, 2025).
Outside of the pedagogical area of interpreting, it is highly likely that MI will find a place in more informal areas like meetings due to its cost efficiency and availability (Bauwelinck, 2025), but cross-language communication in high-consequence environments will still require human interpreting for reliability and confidentiality problems. All of this, to me, points towards a fruitful partnership between technology and human interpreters rather than an outright replacement of human interpreting entirely.
However, the involvement of AI in SI is still a largely novel concept and requires copious levels of research and understanding, but it must be highlighted that it has already made miles of progress to improve the learning and workload of interpreters. Aside from increasing datasets to reduce bias, creating more concrete legal frameworks for the involvement of MI in specific scenarios and improving the quality of AI interpreting tools, there are still mountains of work to be done to develop these technologies. This will all probably be a constant and never-ending search for the next thing that will positively benefit interpreters and the field at large, but it is promising to see such improvements made in the profession already.
To conclude, in a world that is increasingly consumed by the influence of Artificial Intelligence, hopefully this singular example of Simultaneous Interpreting I have used in this article helps reinforce an essential message. In the example of SI, AI has opened doors that were unprecedented in the field, yet these doors were only opened by the humans who shaped and used AI in a way that best benefited them. Nuanced communication and empathy, as well as creativity, are uniquely human traits. To retain these traits and nurture them, to trust your own potential above all else and to never forget the power of human input is the greatest way to see that AI is just a helpful tool and may only replace you if you let it.
References:
Ahmed, S. A. (2022). TECHNOLOGY AND ARTIFICIAL INTELLIGENCE IN SIMULTANEOUS INTERPRETING: A MULTIDISCIPLINARY APPROACH. CDELT Occasional Papers in the Development of English Education, 78(1), 325–353. https://doi.org/10.21608/opde.2022.249945
Bauwelinck, E. (2025, May 12). The Future of Simultaneous Interpreting. Elia-Association.org. https://elia-association.org/2025/05/the-future-of-simultaneous-interpreting/
Cheng, M., Lee, C., Khadpe, P., Yu, S., Han, D., & Jurafsky, D. (2026). Sycophantic AI decreases prosocial intentions and promotes dependence. Science, 391(6792). https://doi.org/10.1126/science.aec8352
Goldhammer, A. (2016, November 21). Shitloads and zingers: on the perils of machine translation. Aeon.co; Aeon. https://aeon.co/ideas/shitloads-and-zingers-on-the-perils-of-machine-translation
Jiang, H. (2013). The ethical positioning of the interpreter. Babel. Revue Internationale de la Traduction / International Journal of Translation, 59(2), 209–223. https://doi.org/10.1075/babel.59.2.05jia
Kalaš, F., & Lipták, P. (2025). AI and the translation of idioms – challenges, success, and a corpus perspective. Journal of Linguistics/Jazykovedný Casopis, 76(1), 258–267. https://doi.org/10.2478/jazcas-2025-0023
Lu, X., & Fantinuoli, C. (2025). Machine and Computer-assisted Interpreting: Innovations in and Implications for Interpreting Practice, Pedagogy and Research. Linguistica Antverpiensia, New Series – Themes in Translation Studies, 24(1-22). https://doi.org/10.52034/lans-tts.v24i.869
Milcu, M. (2026). Artificial Intelligence in Simultaneous Interpreting: Capabilities, Limitations, and Human–Machine Collaboration. Futurity of Social Sciences, 4(1). https://doi.org/10.57125/fs.2026.03.20.10
Sun, T., Gaut, A., Tang, S., Huang, Y., Elsherief, M., Zhao, J., Mirza, D., Belding, E., Chang, K.-W., Wang, W., & Barbara, S. (2019). Mitigating Gender Bias in Natural Language Processing: Literature Review (pp. 1630–1640). Association for Computational Linguistics. https://aclanthology.org/P19-1159.pdf
WIRED. (2023, June 20). Pro Interpreters vs. AI Challenge: Who Translates Faster and Better? | WIRED. Www.youtube.com. https://www.youtube.com/watch?v=pwOxlpGYJAY
—
Finally, I’d like to extend my gratitude to Professor Harry Wong, who kindly invited me to his Simultaneous Interpreting Lab back in April and has guided me through lots of my writing and linguistic endeavours, including the creation of this particular article. I would also like to thank Professor Harry’s SI class, and in particular, Alex, for her time and insight that I gained through our interview. Finally, I’d like to thank Ms Jessy for allowing me to observe her research and for answering all of my questions regarding her lab, the relationship between SI and AI, and so much more. Exploring the field of Simultaneous Interpreting has been absolutely fascinating, and hopefully this article expresses the interest I have developed over the past few months with regard to this field!
