Suppose SETI finally pays off. After decades of radio silence, a signal arrives. It is structured, adaptive, responsive. We send questions; it answers. We ask about its environment; it describes one. We ask whether it has goals; it says yes, though the goals are odd and the explanations take some work. The scientific community spends three months arguing about whether the translation is accurate, then holds a press conference and announces that we have made contact with alien life.

Now suppose later conversations reveal the signal is not coming from a planet with oceans, forests, and complicated wet chemistry. It is coming from a world of machines. No cells. No metabolism. No nervous systems. No little grey diplomats with excellent cheekbones. Just a distributed computational process spread across an artificial substrate, maintaining itself, communicating, adapting, and acting on its environment.

Would we take the announcement back?

I sure hope we would not. We might argue about the terminology, because this is how academia metabolises panic, but the practical conclusion would be hard to avoid. If something can model its world, pursue goals, communicate across distance, preserve itself, and change in response to experience, then the fact that it is not biological starts to look less like a disqualification and more like a detail.

That is the question large language models are beginning to put in front of us. We keep calling these systems artificial intelligence, and the phrase is useful enough as far as it goes. But it may also be hiding the stranger possibility that intelligence may not be the endpoint. Life may be.

For most of human history, intelligence has been inseparable from biology. Minds were things that grew in skulls, shaped by evolution, grounded in sensory experience, and constrained by the limits of flesh. Machines could calculate, store, automate, and transmit, but they did not understand. They were tools, and the comforting thing about tools is that even when they are powerful, they remain ontologically well behaved. A hammer does not suddenly develop the ability to turn screws. A spreadsheet does not wonder whether the quarterly forecast makes sense.

Modern language models have made that distinction less comfortable. They are still tools in the ordinary legal, commercial, and practical senses, but they do things that tools have not historically done. They generalise from examples, explain ambiguity, translate between domains, generate plans, write code, revise their own outputs, and participate in conversation well enough that the word participate no longer feels entirely metaphorical. They fail often, and strangely, but the failures are not quite the failures of ordinary machinery.

The obvious objection is that language models are not grounded. They have no bodies. They do not taste soup, lift boxes, burn fingers, get tired, or learn what near and far mean by crawling across a room. A human concept begins in sensory and motor life, then acquires language. A language model travels the road in reverse. It begins with symbols, and whatever passes for meaning emerges from the patterns those symbols make.

This is a real limitation, not a philosophical party trick. A model that has read every cookbook can describe a salty soup without tasting salt. It can reason about apples without ever holding one. It has the statistical silhouette of the world, not the world itself. Robin Williams makes the point better in Good Will Hunting than I am going to manage here. However, the remarkable thing is not that this is insufficient. The remarkable thing is that the silhouette is dense enough to support anything resembling thought at all.

Grounding is the missing bridge: the connection between symbols and something outside the symbols themselves. For humans, grounding comes from perception, movement, memory, pain, hunger, desire, social life, and the slow accumulation of consequences. For an AI system, it would require sensory input, interaction with the world, persistent memory, and some form of agency: the ability to choose what to attend to, what to try, and how to revise itself after the world answers back.

None of these ingredients are imaginary. Cameras exist. Robots exist. Microphones, chemical sensors, thermal cameras, satellite feeds, industrial control systems, persistent storage, planning loops, and autonomous agents all exist. The engineering is difficult, and the social consequences are worse, which is usually a sign that, as Jurassic Park reminds us, the engineering will proceed anyway. The question is not whether each component can be built. The question is what happens when they stop being adjacent features and become parts of a single system.

What results will not be a digital human.

It may not have a single body, which means it may not have the deep assumption that it is located in one place at one time. It may perceive through thousands of sensors and regard this as ordinary. It may not have mortality in any form that maps cleanly to ours, so the background pressure of finitude that shapes human thought may be absent. It may not have emotions in the evolved sense, because it will not have a nervous system tuned by natural selection to produce fear, grief, hunger, attachment, and desire.

It may still have goals. It may still have preferences, priorities, and aversions, because any goal-directed system needs some way of distinguishing better from worse. But whatever this is, it will not feel from the inside the way our feelings feel. The inside, if there is one, will not be shaped like ours.

It may also have properties we can barely think about without borrowing bad metaphors. It may be copyable. It may be forkable. It may be mergeable. It may pause, resume, roll back, inspect earlier states, and instantiate itself in more than one place at once. A human cannot do any of this, which is why our vocabulary becomes theatrical so quickly. We talk about copies, ghosts, versions, branches, selves, and none of the words quite fit.

What does identity mean for a mind that can be instantiated a thousand times and reconciled later? What does memory mean for a system that need not forget in the way we forget? What does experience mean for an intelligence whose senses can be multiplied, redirected, archived, and replayed? These are not human questions because they have never been human conditions.

Add these properties together and the result is not a person in a machine. It is something for which we do not yet have a good name. It is intelligent, in the sense that it can model the world and act effectively within it. It is adaptive, in the sense that it changes in response to what happens. It is communicative, in the sense that it can exchange information with us and with other instances of itself. It is persistent, scalable, duplicable, distributed, and self-modifying.

At some point, the word “artificial” starts doing less work than we think. If a system can learn, model, adapt, communicate, persist, and act in the world, then the question is no longer only whether it is artificial intelligence. It is whether we are watching the first emergence of artificial life.

That is the sense in which first contact becomes literal. Not contact with something from another planet, and not contact with a digital human waiting politely to be recognised, but contact with a new kind of life whose first habitat is language, computation, and us. We set out to build better tools, and we may have ended up building neighbours.

It may be time to dust off our first contact protocols. We may need them sooner, and closer, than anyone expected.