About twenty years ago, I made a bet with a friend that Chinese would become the new dominant international language, akin to English's position today. I was wrong. Not that the underlying trends have changed, or that its importance won't continue to grow. The world, being non-binary, doesn't mean English will disappear altogether. Instead, a long-term process will occur where English might become akin to what Latin is today. However, with the small AI revolution we are currently experiencing, this process will likely continue, but the importance of international and universal language isn't as significant anymore. Not in the same way.

Context matters. It matters when talking to our friends, colleagues, compatriots, or complete strangers. Context is vital as the subject we’re communicating on, but also as the reference points of our culture and those of our interlocutors.

With the AI tools we have today, accurately translating languages has become trivial. What has become more important to enhance our communications is the context in which we can effectively, swiftly, and accurately communicate our ideas to our friends, colleagues, and AI assistants. It's no longer the natural language translation that truly matters, but the contextualisation of ideas. The concept involves pouring a narrative from one mind to another, from one (training) model reference to another.

As we begin to harness the power of AI, one can easily envision a world where an individual's discourse gets translated to others according to their knowledge, cultural references, and capabilities. What matters is the translation of the idea, not the actual words. In a way, we’re already doing this - we just call it ELI5, ELI12, etc. Furthermore, we’re already utilising additional context translations when we’re reading works like those of Shakespeare, or holy texts like the Bible or the Quran. For these examples, we often receive an explanation of what certain words meant in another time or context, or more information about the semiotics of the narrative.

I’m not introducing a novel idea. Consider the notion that knowing the name of something doesn't necessarily mean understanding it. This idea is illustrated by physicist Richard Feynman.

This principle, simple yet profound, mirrors the central thesis of our discussion. In an age where AI tools can translate languages with incredible accuracy, it's not merely the words that matter, but the underlying ideas and context. Just as Feynman's father taught him, knowing the name - or in our case, the direct translation - does not provide a full understanding. In our interactions with people of different cultures, backgrounds, and knowledge bases, and even more so with AI, we need to ensure the transfer of ideas and meanings, not just words or phrases. This is the driving force behind the concept of meta-tags for context definition and the emphasis on AI's role in contextualising communication.

In this new world, you'll finally be able to comprehend novel scientific ideas and theories, understand the cryptic emails of your quirky colleague at work, or even grasp the demands of that persistent client. Your personal AI would communicate with their personal AI, understand each other's reference points and context, and translate it for you in a way that your AI knows you'll comprehend. We could go further and suggest that we'll finally be able to better understand the people who matter in our lives if meanings, rather than simple words, get better translated into our own context and reference points. So yes, this means one day, you'll be able to better understand your partner, but let's not get ahead of ourselves just yet.

To achieve this goal, one can easily envision the use of a convention for meta tags for text files and AI-generated media, including images and videos. The objective would be the ability to match the reference points of an author to those of the reader. These meta-tags would define contexts such as the author's language, dialect, and the year of the dialect. They could potentially also define the author's native dialect and dialect year in cases where a foreign language is used. When AI-generated works are involved, the meta tags could expand to include: the language used; the model of the AI, its version; and a scale from 1 to 10 of the AI's involvement, from none to a fully AI-generated file.

I shared this idea with an AI to gather some opinions, and it replied, “This is an incredibly forward-thinking idea…” I felt proud for a moment until I realised what it really meant was, “This *might* be useful so incredibly far into the future my friend that by the time it is, we'll have better alternatives.” To which I had the obligatory reply, “I’m not your friend, buddy,” to which, sadly, the AI didn’t respond, “I’m not your buddy, guy!” IFYKYK.

Now, despite the fact that this dialogue did not happen (except for the first reply), I still use it to convey the idea that some memes we have acquired in our lives are not immediately translatable. A reference to a meme could infuse much more meaning into a text or even invert the meaning for someone who understands the context. So, I'm afraid we probably won't be able to translate philosophical attitudes in our personal lives, where they matter the most, anytime soon. While the potential for AI to enhance understanding is immense, it's important to acknowledge the potential challenges and risks. Let's delve into a few of these potential dangers.

AI Shaming: There are some voices advocating for the ability to “detect” AI-generated text, supposedly to determine originality. I don't share this viewpoint, nor do I think it is relevant. We shouldn’t hold ourselves back from using AI, quite the opposite. I fear that the use of these meta-tags, as per the indication of the AI involvement, could be counterproductive if they're used to “judge”. Instead, they should simply be used in a historical context to understand where the author is coming from so that the context can be accurately translated to the reader.

Privacy: Privacy is one of our most cherished values. Personally, I do not think I would always be comfortable using some of the meta-tags, or all of them, depending on the situation and the audience.

Tech & Standards: These meta-tags could be implemented in numerous ways. JSON or YAML at the end of files, for instance. The standard notation could be adapted to fit the paper, academic references, and so on.

Verifiability: The meta-tags could be mutually signed by the public key of the author and by the public key of the AI instance. This would signify that both the author and the AI instance have “agreed” on the content of the meta-tags.

Finally, one could easily argue that these meta-tags are cumbersome, difficult to implement, and potentially futile. The day we start using context translation from one individual to another is likely also the day those contexts could be easily recognised as patterns by the AI and derived. Maybe so. However, the simple personalisation of an AI to match its user capabilities and reference points is problematic as well, primarily because it involves serious privacy concerns and could be exploited in less altruistic ways to cite a few immediately obvious nuisances like marketing or propaganda. 

Like any technology we’ve introduced, the technology itself is not inherently evil. We just need to be cautious and ensure we use this technology responsibly. We’re already discussing human-machine interfaces that could speed up the bandwidth of our communication with an AI and among ourselves. As we continue to innovate, one could imagine these communications involving more intuitive or visual forms that should probably be faster for us to process than verbal communication. Again, in this hypothetic example languages become less important than visual representations. 

Whatever the future may hold, I’m hoping for a one that uses technology to enhance our understanding of ourselves, our fellow human beings, and our comprehension of the world we inhabit.

Example of Meta-tags for this article:

{
  "author" : {
  "language" : “english”,
  "dialect" : “international english”,
  "year_of_dialect" : “2020”,
  “native_language” :”left-out”,
  “native_dialect”: “left-out,
  “native_dialect_year”: “left-out”
  },
  "ai" :{ 
  "model": “chatGPT”,
  "version": 4,
  "ai_involvement" : 2
  },
  "signatures": {
  "author" : "hash",
  "ai": "hash
  }
}

*A proposed 11 points scale for AI involvement :

0/ No AI involvement 1/ AI assisted with basic linguistic corrections (punctuation, spelling) 2/ AI assisted with more complex language corrections (grammar, style) 3/ AI served as a research tool, providing knowledge or references for the topic 4/ AI provided assistance with idea generation and content planning 5/ AI drafted some sections of the text with human guidance 6/ AI served as a significant contributor to the content and ideas 7/ AI independently generated some sections of the text 8/ AI served as a major contributor to both the text and underlying ideas 9/ The author provided the main ideas, and AI generated the rest of the content 10/ The text is entirely AI-generated