Chatbots have become very sophisticated, no question. Without manual effort they allow us to produce vast amounts of text and source code. Humans are prone to make use of this opportunity for obvious reasons. So much, that the sheer volume of generated text renders it impossible for individuals to read it all. And even worse, chatbots tend to produce text with a rather low information-to-text ratio, making the reading process even more tedious. This puts readers in the difficult position of needing to consume inordinate amounts of text just to filter out the information of interest. In response to this explosion of textual data, readers also turn toward AI systems for content summarization and information extraction. The AI, thus, serves as intermediary between the massive volume of text and the human decision-makers, re-enabling an efficient decision-making process by providing condensed, relevant information.

Over time, we, thus, can expect a transition toward a Human-AI-AI-Human communication scheme. Unfortunately, all on the basis of vast packets of text in natural language. And, the ambiguous nature of natural language may lead to misunderstandings between AIs, as it does between humans. An intriguing alternative to natural language is the concept of common latent spaces, where information is stored in a more compressed format, as a set of numbers. Unfortunately, this format is incomprehensible to human beings. But, those common latent spaces could be engineered to show minimal ambiguity and to be independent of real languages, like English, German, or Chinese. Hence, as a side effect, language boundaries would be lapsed. These packages of numbers containing generalized information can then be queried by the receiver AI, extracting only the requested components while ignoring the rest. Common latent spaces, thus, offer an innovative approach to information exchange, potentially more efficient than traditional natural language communication. In this new order of communication we might overcome the problem of misunderstandings originating from our ambiguous natural languages and also fix the problem of overwhelming bodies of texts.

The idea sketched above is merely an interesting thought experiment. A lot of things most certainly may go wrong on all levels of society, should we really approach the era of Human-AI-AI-Human communication schemes. Note that problems are to be expected independently of the exchange format between the sender and receiver AI, be it natural languages or common latent spaces.