Menu
Young H., CEO |
November 15, 2024
The CEO of a translation company assures you: artificial intelligence will replace translators.
Humanities interpret, describe, and define human actions and the principles of the world from a subjective perspective. Fields like philosophy, literature, art, history, politics, linguistics, and religion all fall within this domain. Therefore, the value of the humanities is not determined by the subject being observed but by how the observer defines it. For instance, Han Kang’s Nobel Literature Prize was awarded for “powerfully expressing historical trauma and human frailty,” a value rooted in the humanities.
On the other hand, science rigorously excludes subjective thoughts and experiences, instead of pursuing objective and verifiable principles. Regardless of what anyone argues, the speed of light in a vacuum is precisely 299,792,458 meters per second, and according to the second law of thermodynamics, entropy in the universe will never reverse. If this principle were false, it would imply time flowing backward or decaying corpses coming back to life. Even when the Vatican claimed the Earth was the center of the universe, our planet was still orbiting the Sun at a speed of approximately 30 kilometers per second.
Most translators have a background in the humanities. In contrast, artificial intelligence (AI) is the product of mathematical calculations developed by scientists. By converting data into numbers, arranging them in matrices, stacking these matrices, and calculating the relationships between these numbers as vectors, AI emerges. While those in the humanities may describe or evaluate this vast collection of computations based on their values, hopes, and interpretations, their subjective views do little to aid in understanding the potential and limitations of AI developed by scientists.
When CDs replaced LP records, many people argued—and some still do—that analog sounds were superior to digital, claiming they were more “human.” From a scientific perspective, analog signals are continuous waves, much like sound waves. Conversely, digital signals break down those waves into 60 segments per second (60Hz) and convert the pitch of each sound into 65,536 levels (16-bit) using electrical signals. While the result is a clearer sound, some dislike it for being more “mechanical.”
However, if you think about it, even the fundamental aspect of humanities, “writing,” is digital. Humans first communicated through speech (analog signals) and later invented rules to record those sounds using specific symbols (digital characters). This means the word “mother” is read as “mother” by anyone who encounters it. In the same vein, musical scores are fundamentally digital. They transform intangible sounds into nine distinct pitches by adding two semitones between seven primary notes and using a system of lines and scales to represent the entire range of music. In addition, note lengths are indicated by the shapes of “bean sprout heads,” while tempo matches the ticktock of metronomes. This standardization of pitch and duration, regardless of performer or instrument, finally enabled variations and ensemble performances. Thus, while music itself is highly analog, it was born from the highly digital framework of sheet music.
Furthermore, the human body itself is synthesized based on the digital nucleotide code known as DNA. The thoughts and feelings derived from this are essentially electrical signals connecting neurons in the brain, circuits formed through external stimuli and learning. AI, as described earlier, is modeled after this. Just as we cannot precisely calculate which neurons connect to produce the emotion we call “love,” we also cannot know the exact process AI uses to generate its answers.
The reason for this lengthy preamble is to emphasize that, regardless of the professionalism and emotional nuance that humans may bring to translation, AI will undoubtedly replace humans, at least in “commercial translation.” It’s inevitable, just as dawn breaks, no matter how one twists a rooster’s neck. The reasons are clear: automated translation is much faster, far cheaper, and continually improving in quality. While those with a humanities mindset might question, “Is this something that should be said by the CEO of a company that has been in the translation business for over 20 years?” from a scientific perspective, whether I stay silent or not, every day, countless graphics processing units (GPUs) are being produced for AI training, and engineers are ceaselessly training AI models and developing services based on them. Even the stock prices of power companies have surged because of the increased demand for electricity to run these massive GPUs. AI is becoming more sophisticated and intelligent, rapidly approaching a level where it can outperform humans in translation.
We’ve been consistently emphasizing the impact of AI and sharing relevant visions and solutions. (Translation Night 2023)
Even now, in translation bid venues, most judges, often professors, heavily emphasize how many professional translators a company has, even though most of them are unfamiliar with CAT tools (computer-aided translation tools: software that aids translators in leveraging previously translated materials for efficiency, which is becoming less relevant with the rise of AI). It feels like having horsemen evaluate the performance of cars.
It’s true that AI’s expertise and accuracy still fall short of humans, often failing to grasp context. However, this comparison is unfair from the perspective of AI. Humans assign hundreds of people by field and language, each responsible for a specific area, while AI has to handle everything on its own. Yet, I confidently assert that more than half of the hundreds of translators working with our company do not match the level of AI translation in terms of expertise and accuracy, including the much-vaunted “context understanding.”
Understanding context has always been a chronic issue, even for human translators.
Context encompasses various variables, not just the “emotions” the author intended. Cultural context, historical context, and generational or gender-related nuances are all factors to consider. Can we truly say that human translators have been free from errors in understanding context? Consider how, for the longest time, the translation of the Filipino everyday question “Anong ulam mo?” has defaulted to “What viand are you having?” for some reason. While viand is a word, rarely do you encounter its usage, and it doesn’t quite capture the nuance that ulam refers to the food you eat with rice, whether it’s meat, vegetables, or something else. Translators would do well to finally retire “viand” and translate the question to something like “What rice meal are you having?” Thus, I do not agree that AI makes contextual errors more frequently than humans do; rather, it simply makes different kinds of errors. There’s a paradox where what’s easy for humans is difficult for computers, and vice versa. AI consistently scores better than the average human in objective tests involving contextual understanding, such as essay questions. Additionally, while it’s difficult to correct the contextual errors humans make, which often devolve into unproductive debates, AI can acknowledge its mistakes immediately, correct them, and evolve into a better version within a few months.
In conclusion:
The time will soon come when we say, “Translation is a task for AI, just like calculations are done by calculators.” The transformation where one or two reviewers now review and refine translations done by AI—tasks that previously required 10 translators—is already happening. The top 10%–20% of translators will transition to jobs as reviewers or prompt engineers, commanding higher pay because of increased productivity, while the bottom 80% will lose their jobs rapidly.
Though this is a future that seems daunting and undesirable, denying it won’t stop the water from flooding into a leaky boat. Recognizing reality is essential to find ways to survive. Let me propose three practical solutions for now.
Further discussions on this will follow in the future.
