AI has been with us for years. Car navigation systems, traffic cameras that read license plates, Google Translate, and even the home thermostat that knows who is home and who isn’t: we might call it something else, but it’s all AI — nothing more, nothing less.

The reason AI has gained so much extra attention since 2022 is all about the rapid development of deep learning. Don’t ask me how, but deep learning uses linear algebra and statistics to process vast amounts of (human-generated) data and extract patterns from it. In short: it trains and teaches itself.

Large Language Models (LLMs)

This is where so-called Large Language Models come into play. These LLMs, from companies like Google and OpenAI, have been trained to such an extent that they can now perform tasks like photoshopping, composing music, automating robots, making medical diagnoses, mapping climate developments, and coding software.

As deep learning continues to analyse more and more data and trains itself, many researchers are grappling with the question of what AI will ultimately be capable of and whether more deep learning leads to ‘better’ results.

The central question is: what does ‘better’ deep learning mean? Is it ‘better’ if it scores increasingly higher on a math or law exam, or is it ‘better’ if it starts coming up with creative ideas on its own?

What Can LLMs Do and What They Can’t (Yet)?

Current models excel at generating essays, creating TikTok videos, and summarising a legal dissertation. All of this is made possible by the patterns they extract from their training data. However, these models fall short in determining which data is logical, applicable, and appropriate. Logic and reasoning based on trade-offs are still largely missing.

In that context, it’s currently unclear whether the present LLM models will ever be capable of consistent reasoning. It might even require entirely different and yet to be developed models to achieve that level of cognitive ability.

The Role of Humans

As long as AI’s limitations persist, human logical reasoning and creativity will be essential to fully harness the benefits of both. Or, as Albert Einstein once said (and he was usually right): if you keep using the same method, you’ll keep getting the same result. In short: creative thinking should never be excluded.

You might think this sounds self-serving, but there is now serious research on this. Studies (source: April 2024 npj Digital Medicine) show that combining AI with qualified human intellect — including reasoning skills, intuition, creative thinking, and logic — significantly improves data output.

This has been demonstrated in medical, cultural, mechanical, and legal research, as well as in the translation industry.

…And What Does That Mean for Translation?

In translation, context, nuance, creativity, logic, and localisation (cultural adaptation) play a critical role. We’ve seen empirical evidence that some of our clients, based on their experience with AI, have developed a more nuanced perspective on its applications.

After experimenting with various translation software (which is, of course, an AI product), the conclusions are, let’s say, mixed. This doesn’t apply to less complex translations like emails, simple word lists, or a website for furniture, offering colour choices for a new chair. However, for business-critical texts that need to be precise, where nuance, logic within context, specialised knowledge, effective communication with the target audience, and the use of commercial keywords are essential, AI alone can fall short.

That’s where clients also look for human intervention. I deliberately write ‘also’ because clients don’t necessarily abandon AI in translations, but they are looking for ways to combine it with human correction and text editing.

Translation Kings’ Role When AI Translations Are Desired

Translation Kings has been offering this service since last year: a machine translation with post-editing. We (or you, if you wish) handle the machine translation and then have it post-edited against the source text by a professional translator. For the reasons described above, this gives the client a fast, cost-effective, and quality translation. While it’s not an ISO 17100-certified translation, it’s good enough for the client’s intended purpose, and without the shortfalls which AI has not yet overcome.

As long as AI cannot solve problems on its own and still requires human intervention for guidance and correction, Translation Kings embraces AI technology in any application our clients desire.