We live in an era where machines not only translate, but also write, summarise and even create poetry. But what happens when humans decide to play with language in such a convoluted, creative and ambiguous way that it escapes the logical and statistical understanding of AI models? In an increasingly automated world, language remains the last bastion of the purely human.
Language as a trench
AI language models, such as LLMs (Large Language Models), are trained with millions of phrases, texts and structures that already exist in the digital world. But what happens when a human being creates something totally new, surreal, with cross-cultural references, double meanings and unconventional structures?
Magic happens.
Because that’s where the machine stumbles, hesitates, and in many cases, fails. Convoluted language is not simply synonymous with difficult; it is a form of elegant sabotage, a subversive linguistic act where every word is a semantic trap.
When we don’t want the machine to understand us
“Flippin’ chickens are waiting for the mule to come. Fancy a wooly in the meantime?”
Does it make sense? Yes, to a human reader willing to play along and familiar with drugs slang. Does it make structural sense to an AI? Probably not. It’s not trained with that vocabulary.
Here’s an explanation:
• Flippin’ chickens: a ‘chicken’ is another word for a kilo of cocaine. In some cities, the word is reserved specifically for a kilo of crack and a ‘bird’ would be used for a kilo of raw powder cocaine. The act of ‘flippin’ chickens’ can simply mean selling kilos of cocaine or crack for a higher price than they were purchased for. In some cities, ‘flippin’ chickens’ is the act of buying a kilo or more of cocaine and cooking it and transforming it into crack cocaine; this process actually adds weight and volume to the final product, making it much easier to turn a profit.
• Mule: a carrier or supplier of drugs
• Wooly: a marijuana cigarette laced with cocaine.
These types of expressions rely on hybrid metaphors, mixtures of unrelated semantic fields, surreal humour and references that require cultural, emotional and even experiential context. AI, however powerful, has no experiences.
The limitations of AI training
AI models work with what they have seen. They are predictive, not intuitive. Intuition, that human ability to read between the lines, to grasp tone, irony, hidden meaning… has not yet been replicated.
That is why, when baroque language is deliberately abused, or sentences are constructed that appear logical but lack it (unless understood from a poetic, absurd or emotional perspective), AI is left out of the game.
And that’s not a mistake. It’s a space for resistance.
Language as emotional cryptography
At BBLTranslation, we have been defending the value of language as a strategic, technical and emotional tool for years. In the world of AI, where we work training linguistic models, creating multilingual datasets and auditing systems to ensure consistency and ethics, we are fully aware of the limits of automation.
BBL AI, our AI language solutions division, is dedicated to ensuring that words, even in intelligent systems, retain their value, context and nuances. But we also recognise that there are areas where human creativity becomes indecipherable code for machines.
What if that’s a competitive advantage?
Humans can play with language as they please. They can invent. They can be ironic. They can say one thing and mean another. They can use the passive voice, polysemy, puns, metonymy and leave even the most sophisticated system speechless.
AI, on the other hand, for now, needs logic, consistency and patterns.
In our previous post Boost your digital visibility with GEO: content optimisation for generative AI engines, we talked about how to make AI understand and find you. Here, we want to give you some tips on how to prevent it from doing so, because you might find that more interesting.
Practical applications: how can this ‘secret language’ be used?
- Protection of sensitive content: In contexts where you do not want AI to index, analyse or classify certain information, deliberately ambiguous or creative language can be used.
- Advertising creativity: Copy that plays with absurdity or double meanings can generate real human engagement, precisely because it does not respond to predictable formulas.
- Privacy in automated environments: In channels where messages are automatically analysed, a form of “weird writing” can keep conversations off the radar.
In short: the machine learns, but the soul writes
Linguistic models will continue to advance. But they will never have childhood, their own cultural context, a sense of absurd humour, family grudges, or unique ways of playing with language as we humans do.
So, if you are looking for privacy in automated environments and don’t want the machine to understand you, we can help you!
