The first error of our era of anxiety is grammatical. We say “AI thinks”, “the agent decided”, “the model understands”. Business adopts the jargon because it flatters transformation decks. Politics adopts it because it outsources responsibility. Complex-systems people adopt it because “emergence” makes surprise sound like ontology. The machine has not crossed a threshold. We have... into systematic misnaming.
AI is not an agent. It has no body, no biography, no mortality, no debt to tomorrow. It does not live with its outputs, suffer their consequences, lose a reputation, or repent. The thing you call "the AI" is a stochastic parrot, a conditional probability distribution over token sequences. It is trained on the symbolic remains of human activity. It produces judgement-form because judgement left traces. It imitates deliberation because deliberation has a style. It answers fluently because institutions converted thought into documents, policies, tickets and speeches.
That is power. It is not agency.
For leaders, the distinction is not academic. Treat AI as agent and governance collapses. Accountability shifts onto a tool that cannot answer. Strategy becomes ventriloquism. “The system recommended” becomes the alibi. Firms call it autonomy when they mean accelerated bureaucracy. Governments call it technological progress when they mean probabilistic administration. It is human responsibility laundered through an artificial substrate no one is forced to own.
“Intelligence” is no rescue. Intelligence presupposes a bearer of ends. The substrate has none. It has objective functions, reward traces, guardrails, rankings, temperature, prompts, policies — not purposes. Ends belong to vulnerable beings and their institutions. A model may optimise language around a value; it cannot be bound by the value. It can simulate prudence; it cannot be wise. It can mimic conscience; it cannot be guilty.
The right category is infrastructure: symbolic substrate, not synthetic colleague. A prosthesis of structure, the way the printing press was. Treat it as compiler, database, market signal, court record, logistics network: powerful, partial, structuring, corruptible, never sovereign. It conditions action; it does not author it. It expands structure; it does not replace agency. Its outputs deserve audit, not deference. Its fluency deserves suspicion, not awe.
The danger is not that AI will become a god. The danger is that leaders exhausted by responsibility, will enthrone an instrument and call abdication innovation. The substitution is technological pneumatology: consecration of unredeemed ideology as redemptive mediation. Naming it correctly is the first act of using it well.
AI is substrate. Keep it there.
#AI #Leadership #AIGovernance #DigitalTransformation #PhilosophyOfTechnology
Q: I'm curious as to why you landed on using the word "substrate" rather than "infrastructure"?
A: Because Infrastructure (like roads, servers, or cloud platforms) is something you build on top of. It’s usually passive: it provides capacity, connectivity, or storage. You use it, but it doesn’t fundamentally reshape everything running on it. Substrate, on the other hand, comes from science—it means an underlying layer that actively enables and shapes what grows or operates on it. Think of it like soil for plants, or a biochemical medium in biology. It’s not just support—it influences behavior, outcomes, and evolution. AI isn’t just one layer (like cloud computing); it seeps into every layer—apps, workflows, decision-making, even user interfaces. It doesn’t just host applications; it changes how they are designed and behave. For example, software becomes probabilistic and adaptive instead of strictly rule-based. And AI systems can be embedded anywhere—inside tools, APIs, chips, and processes—like a material you can mix into everything.
Plus most importantly, unlike traditional infrastructure, AI can generate, “decide”, and act, which makes it feel less like plumbing and more like an active operator.