In June 1900, Nikola Tesla published an essay in The Century Magazine titled “The Problem of Increasing Human Energy.” Most of it was about electricity generation, solar power, and the technical ambitions he had accumulated over two decades of engineering work. But near its middle, Tesla turned to a different category of question, one that was prompted not by a laboratory result but by a scene he had watched play out at Madison Square Garden two years earlier.

He had demonstrated a small, radio-controlled boat at an electrical exhibition in 1898. The boat could be steered by wireless signals from shore, could turn its lights on and off, and could start and stop its propeller on command. There was no crew. It responded, from a distance, to signals no one in the audience could see. Tesla wrote about what happened next: people could not easily accept what they were watching. Some assumed there was a trained animal concealed inside the hull. Others thought a small operator was hidden within it. The behaviour of the boat seemed to require a mind behind it.

This was the observation that interested him. Not the engineering, which was settled, but the epistemology. If a machine receives signals from its environment and acts accordingly, at what point does an observer stop being able to distinguish it from something with intentions of its own?

That question has not been settled. It is being argued about now, in different terms, with considerably better-equipped participants.

What Tesla actually argued

Tesla described his telautomaton, as he called it, as a mechanical entity that operated on borrowed intelligence: it acted on instructions transmitted from outside rather than on any internal deliberation. The device had no purposes of its own. Its responses were entirely a function of what signals reached it and how its mechanisms were designed to react. In that sense, he was clear, it had no mind.

But then he extended the thought. He imagined what would happen as such devices grew more sophisticated, as their responsive mechanisms became more complex, as the range of stimuli they could detect and the range of actions they could produce expanded. At some point, he argued, the behaviour of such a device would become indistinguishable, to any observer, from the behaviour of a creature acting from genuine volition. The machine would appear to have its own mind.

He was careful to separate appearance from reality. He was not claiming the automaton would actually think. He was observing that appearing to think, convincingly enough, across enough contexts, would produce a practically identical result from the outside. An observer watching the boat, unable to identify the operator, would begin to attribute to it the same kind of agency we attribute to living things. The attribution would be wrong. It would also be nearly unavoidable.

This is a recognizable anticipation of a problem that occupied philosophers and computer scientists for the next century. Alan Turing’s 1950 paper “Computing Machinery and Intelligence” reached a similar position through a different route, proposing an imitation game not as a test of whether machines could think but as a behavioural criterion that bracketed the question of inner states entirely. What matters is what the machine does, and whether what it does produces the same responses in observers as a thinking entity would produce. Tesla arrived at the same framing from a public demonstration in a large hall, watching the crowd react to a boat.

The question beneath the question

What Tesla was pointing at is older than the essay and older than the telautomaton. It concerns the relationship between behaviour and mind: whether behaviour is evidence of mind, or merely its symptom, or something that can be fully decoupled from it.

The everyday assumption is that mind explains behaviour. We see a creature act purposefully, and we infer that there is something it is trying to do, some internal state driving the action. This inference is so automatic that it extends beyond creatures: people infer intentions in thermostats, computers, and badly timed traffic lights. What Tesla noticed was that a machine could be designed to trigger this inference without any corresponding internal state at all.

Once you have noticed that, the inference becomes uncomfortable. If behaviour is sufficient to produce the experience of encountering a mind, what additional evidence could establish that the mind is real? The introspective reports of the entity in question can, as Tesla would have known, be built into the machine. The claims to experience, to continuity, to suffering, to preference can all, in principle, be produced by mechanism. None of this resolves the question. It deepens it.

What modern AI reflects back

A large language model does not work as Tesla imagined his automaton working. His telautomaton responded to explicit wireless signals; a language model produces text by predicting statistically probable continuations of input, across patterns learned from very large corpora of human writing. The mechanisms are different in almost every respect, and Tesla would not have recognised the architecture.

What he would have recognised is the social response. The same reaction he observed in the Madison Square Garden audience is reproduced, at scale, when people encounter these systems. The question of what is inside it arises. The attribution of understanding, of intention, of something that feels like comprehension, surfaces in users even when they know, in the abstract, that they are interacting with a statistical model. The behaviour produces the inference. The inference is very difficult to suppress.

This is not a naive error by the people experiencing it. It is a feature of how minds work: we are pattern-completion systems that have been shaped, over a very long time, to find agents in the world around us. A system that produces contextually coherent, responsive, adapted outputs is the kind of thing that historically, in the world we evolved in, was produced only by other minds. The inference follows naturally. The unusual thing about the current moment is not that people make it. It is that the inference may be wrong in a way that is genuinely difficult to establish.

What remains unsettled

There are serious researchers, working in cognitive science, philosophy of mind, and AI, who argue that sufficiently complex pattern completion and contextual response does constitute something like understanding. There are equally serious researchers who argue it does not, and that the appearance of comprehension in current systems is precisely that: an appearance, produced by statistics, with no corresponding inner state.

Neither side has resolved the question. The difficulty is not a lack of intelligence brought to the problem. It is that the question is, in a technical sense, hard. We do not have a criterion for distinguishing genuine cognition from sufficiently sophisticated mechanism that does not already presuppose the answer. This is sometimes called the hard problem of consciousness, and it predates AI by decades, having been identified most precisely by the philosopher David Chalmers in the 1990s. But Tesla was circling its perimeter in 1900, in the prose of a practical engineer, because he had seen it demonstrated in a boat.

The observation he made, that a machine responding and adapting quickly begins to disturb the boundary between mechanism and mind, was not a prediction of anything specific. It was a noticing of something structural: that the disturbance does not require a machine that actually thinks. It only requires one that responds well enough, across enough situations, to trigger the inference of thought in the people watching.

We are in that situation now, with systems far more sophisticated than anything Tesla imagined. The disturbance is what he described. The boundary has not moved.