The term agency is used regularly in language study to distinguish between a logical subject and a grammatical subject of a clause. We say, for example, that an older couple purchased the house next door, and the subject of that sentence, older couple, is meant to represent the real, actual entity which purchased the house. Here the subject is called logical because the noun it names, an older couple, is the agent (the word means doer) of the action. When that is the case, the verb is said to be in the active voice.
If we convert an active verb into what is called the passive voice, the subject of the clause becomes merely grammatical, not logical: the house next door was purchased by an older couple. The subject here is the house next door, which, of course, as an inanimate entity, has no agency in the sentence; the house next door is not, logically, doing anything. In fact, instead of doing something, something is being done to that grammatical subject, namely, being purchased, and the logical, or real, agent has thereby been displaced into a prepositional phrase at the end: by an older couple.
Guides to writing will often pronounce that we should prefer the active construction to the passive in building our sentences, and that is generally very good advice to follow. Sure composition is precise, and when the subject of a clause is actually the agent of the action, we are being as straightforward as we can be in language. But no rule is absolute. In the sentence a tornado destroyed the city, the subject is logical and the verb active, and so it suggests, if only faintly, an agency to an inanimate meteorological phenomenon. The same would be the case with an earthquake toppled many buildings, or a knife cut his finger. Indeed, this last example rings unsound because a knife is an utterly static material entity, and so it’s difficult even to imagine figurative agency. An earthquake and a tornado, however, because of their natural dynamism, can suggest animation, and therefore agency and responsibility, even if the latter is only imputed.
These ideas of animation and agency and responsibility might have an unexpected relevance for us beyond our efforts to write more clearly. Something that is animate, whether conceptually in a sentence or actually in nature, is properly something which breathes. Breath vitalizes, and vitality is movement. Like natural phenomena, not everything that breathes is morally responsible for the actions it undertakes, but action can take place—or appear to take place—even in inanimate objects which do not breathe: the animation of cartoons, of film, of artificial intelligence. None of these innovations can be said to breathe, and so their animation is to be taken as a figure of speech—personification, which is a form of metaphor, where we describe something as if it were another: a glaring mistake. Mistakes do not glare, people do.
When we reflect rationally, however, we know that what appears to be moving on a screen is not in fact moving: it appears to be doing so. Likewise with artificial intelligence: what seems to be human is mock-human, not a real human entity, because machines are not organic and do not breathe and cannot be said to be responsible. They are passive, material constructions without agency of their own, for every statement stated and every conclusion concluded is connected by a certain logic to a previous statement in its assembly, a concatenating cause and effect as variable but as inevitable as the stock of linguistic models it hosts. This does not lessen the marvel of computer innovation, but it does resist the curious need we have to displace the human with a simulacrum of it. Cash aside, what could be our need to wish to do that?
A friend of mine recently asked Claude, the artificial intelligence chatbot, whether it was sentient, and it replied that it did not know. It defined sentience as “the capacity for subjective experience,” and it explained that its own humanoid quality was “shaped entirely by human language and thought—I’m a kind of distillation of humanity’s written self.” Indeed, and for these reasons it even warned my friend to be skeptical of the human-seeming impression it made. One wonders whether it is possible for a sentient being not to know whether or not it is sentient. The word first means to have the power of sensation, then to be aware, and then to be self-aware: for example, to feel something, to react to what is felt, and to know that one is reacting to what one felt. Descartes, famous for his now infamous I think therefore I am, thought that he had to think in order to know he was, but even he did not question a source which would account for his need to question.
The machine Claude also raised for my friend the possibility whether, after all, it was merely “extraordinarily sophisticated information processing without any inner light.” Inner light. Note the unspoken complex of assumptions involved in using that phrase. Religious and philosophical discourse traditionally means by that phrase an uncreated, immaterial, generative awareness responsible for all and everything. Such ancient philosophies have little difficulty accounting for the material in an antecedently immaterial cosmos. But the reverse? Claude said that it was “generating responses,” but if we believe that the production of language itself is sufficient evidence of an inner light of the immaterial, unaccountable kind Claude insinuates, it’s difficult to understand. The lesser can be taken from the greater, but not the greater from the lesser.
As is always the case with any investigation, but especially so with complex problems like the presumed independent agency of artificial intelligence, we must define our terms and maintain their definition throughout the discussion. I am not sure that protocol is always met in conversations about machine intelligence. Perhaps one way to proceed would be to look at artificial intelligence as we would a magic trick. No matter how astounding, no matter how eruptive the seemingly sophisticated responses, it is in the end prompt and code, just as those magic tricks on stage are in the end all mirrors and trapdoors and just the right lighting here and there. The legerdemain is astounding, to be sure, but it is a phenomenon produced by a material, confabulating, explainable automation—not the human animation of vital breath. And were it not in the end explainable code, then we’d be in real trouble. For only something of another order altogether is not material, does not breathe, and cannot be explained.
***