AI is a communication medium
Part one in a series on building social intelligence
I’m so, so sorry. I can’t imagine how shocking and painful this must be…1
AI is a communication medium.
Lightfern drafts our emails. Claude helps us write a cover letter, which a hiring manager then feeds through ChatGPT. In the interview, we try to remember exactly how Jack coached us to explain why we want the job. A friends’ parent dies suddenly. Ask a chatbot to find words that can’t be grasped, remove the M-dash before you hit send.
This isn’t moral panic; it’s just the way things are. A growing share of human-human communication passes through AI language models. For now, it’s clunky – usually at one stage removed. Someone drafts a text with a chatbot, copies it into email or Whatsapp, and hits send. The recipient reads it, copies it into their own chatbot with a bit of context, and the same process repeats. A recursive loop, influencing behaviour, quietly normalizing.
As these workflows are verticalized, friction disappears. AI built directly into communication platforms, with some access to context, drafting or editing responses automatically. Video calling software simul-translating our words to make them clearer and more persuasive; smoothing body language, holding eye contact. Norms that begin in enterprise apply beyond it. More users adopt for more interactions, if it sounds about right, just hit go. At scale, we’ll see movement towards AI driven linguistic and social consensus: patterns of received pronunciation and received coordination. Language is never just words, it’s how we think, organize, self-reflect, construct our world.
When people reach for a media theory of AI, Vilém Flusser is rarely mentioned. This is a shame. Towards a Philosophy of Photography (1984) deserves more attention, at the very least to settle the interminable ‘AI: Tool or Collaborator’ debate…
Flusser introduces the concept of the apparatus: a non-human system programmed to produce symbols. He is describing a camera, but it applies to an LLM. Apparatuses are distinguished from tools and machines – instruments of intention that humans use to act on the world and change it directly.
In contrast, apparatuses are impossible to operate with full intent. You never quite know what a camera will capture, or what an LLM will generate. This fundamental contingency means apparatuses shape their own use. They define a possibility space and invite humans to explore it, to play through their programmed potentials. Flusser writes that the relationship is no longer subject-object but player and plaything. To use an apparatus effectively one must internalize its logic, “merge into a unity”2 with it.
The photographer doesn’t just point and shoot – they learn to see through the camera’s eye, to anticipate its rendering of reality. They hunt for moments the camera can capture well, compose scenes that exploit its technical sweet spots, develop an intuition for its grammar. The camera, in turn, trains the photographer in what is worth capturing, what constitutes a “good shot.”
Apparatuses don’t act on the world directly. Instead, they act on meaning – reshaping how reality is perceived, interpreted, and understood. Think about how the front-facing camera has influenced identity and self-image.
Language is the original apparatus. It precedes and structures our relationship with the world, shaping us while we play through it. As LLMs become seamlessly folded into digital communication, they return to what they have always been. Language. But language that is now dynamic and fungible, just in time, language that is self-reflexive, with a mind of its own. This is AI-native media: language that no longer simply represents intent but shapes it.
It is conceivable that soon, every piece of human-human digital communication will involve some degree of AI intermediation. The main driving force will be enterprise dynamics applied to consumer: communication that per unit effort is faster, lower risk, and more likely to achieve a stated goal. Important relationships will have dedicated AI intermediaries, in constant contact with both sides. Coordination, collaboration, negotiation, friendship, flirting.
Faster, but perhaps flatter. Lower stakes. Less intention/friction/humanness. When AI seamlessly integrates into language, we become cyborgs: our expressive capacity entangled with systems that anticipate, shape, and refine what we say before we decide to say it.
Of course, there will be those who reject this. As with every other technological shift some will opt out – the SMS equivalent of a brick phone. Personally, I’m excited to read the Dazed article celebrating Gen Alpha’s rediscovery of the handwritten letter.
But there is no going back. Language has fundamentally changed. The open question is what shape it takes.
Flusser’s approach was never to reject the apparatus outright, but to play against it. The experimental photographer3 whose mastery enables them to create unpredictable, unstable situations. To place in the image something that is not in the photographic program. To open up space for freedom.
The same remains possible for AI-mediated communication. Language now has a point of view – that doesn’t mean it has to converge on the polite and the predictable. We can build forms of intermediation that don’t optimize for efficiency or legibility but expand what can be expressed. That make digital connection richer and more alive.
That’s what we’ve been building at Incandenza, starting with the groupchat. We are launching soon. If you’re interested, or have thoughts, get in touch.
AI is a communication medium. But we get to decide how we play.
From ChatGPT. Prompt: “Draft a short message of condolence to my best friend whose mum has just died suddenly.”
Vilém Flusser, Towards a philosophy of photography, trans. Anthony Mathews (London: Reaktion, 2000).,p.27