The Delegation Schism

The Delegation Schism is the divide between the two sentiments regarding delegating any given task to whatever form of artificially intelligent aid. The crux of this schism is the inability of the Accelerants and the Preservationists to see eye to eye. A truce, however, is possible.

A Tale of Two Camps

On one hand, the Accelerants see delegation to AI as obvious progress: delegating the tedious, compressing the mundane, removing friction, and moving faster. On the other hand, the Preservationists argue that between the point at which delegation is instructed and output being produced, something is lost in that making. More often than not, that thing remains unnamed. One camp sees delegation as the removal of unnecessary friction, and one sees the friction as a feature and not a bug. To get a negotiation started between the camps is to shelter any given task under one of two umbrellas:

This classification matters due to delegation being a non-neutral act. To delegate a task to AI is to make a bet, whether consciously or otherwise, about where the task’s value lies. This is the Delegation Wager.

The wager can be mapped simply:

The wager is not always lost. Much of modern life belongs plainly in the upper-left square. There are tasks whose value is preserved, even improved, by delegation. To pretend otherwise is to confuse friction with depth.

Verdict: Completion Found Not Guilty

The thing about human attention is that it is by no means infinite; the reserves are finite, and wasting this finite resource on low-formation friction is not noble. Thus, the Accelerants have a case: that not every task is worthy of a soul, and in that case, they plead not guilty to delegation misallocation. When tasks are mere connective tissue, their completion becomes the means and the end. Only a usable result is needed. An organized inbox, a calendar plan, a low-stakes administrative email are all non-sacred completion tasks and are there to support meaningful work as opposed to being the meaningful work themselves.

An Accelerant is not necessarily someone avoiding effort by nature; they just may be someone allocating effort (another finite resource) to places in more need of it. To delegate a Completion task is far from abandoning or cheapening human work; it simply protects bandwidth and attention for work that actually requires a human.

A task does not become profound just because a human performed it inefficiently and, in a manner, subpar. However, it is this very logic that liberates attention in one case and hollows out meaning in another. Accelerant logic begins to fail when the manner in which Completion tasks are taken care of becomes a blueprint for all other tasks alike, and when Accelerants assume that every act can be judged by the existence of an output.

At some point, delegation crosses a certain threshold.

The Enactment Threshold

In the court of formation, the Preservationists are also found not guilty. An apology is not remorse-shaped lexicon. An apology is confronting harm. A condolence note is attention to grief. A wedding vow is obligation and love. These are Formation tasks. They do something to the doer, and upon their completion, the doer is not the same. This notion is anything but a modern one; Aristotle called it habituation: the self is not declared but practiced into being, one enacted act at a time.

To better understand why Formation tasks are formative for their respective doers, and why delegating them is ill-advised, it’s helpful to dive into thought. For the sake of the argument, Charlie will serve as the subject in a loosely staged dialogue after philosopher Robert Nozick’s “The Experience Machine” put forth in his 1974 Anarchy, State, and Utopia.

A machine is placed before Charlie.

In the machine, Charlie would be plugged in and would float, experiencing a preprogrammed life.

The machine offers the experience of a life, but not its enactment. Nozick’s thought experiment is not about artificial intelligence per se, but it does clarify the arena AI delegation now enters. The question that is now faced by all becomes: Does the value of an act survive when the doing is removed from the doer?

The bone of contention here is that the loss is rarely visible in the produced artifact. When AI delegation fails crudely, the problem announces itself: the answer is false, the summary is off, the tone is cold. But when delegation succeeds (as it increasingly does with each more capable model released), the loss becomes harder to detect. A generated apology may sound more graceful, a generated dissertation may read smarter, a generated condolence note may arrange tenderness better than grief can. In each of these cases, the artifact may improve while the enactment disappears. Surface becomes shinier and more reflective, as it becomes shallower. These are the diminishing returns beyond the Enactment Threshold: that something is produced as output, yes, but that very something is produced without the person being formed by its production.

The threshold holds at even the smallest of scales. Consider the difference between typing out a wall of text and pasting it. The output is identical to the byte, but the doer isn’t. Typing forces a passage through the material at the speed of hands, and somewhere in that duration, thoughts form. Most of the thoughts that form are noise, that much is undisputed, but occasionally one makes itself known that could not have arrived any other way. Formation, then, is not a currency to be earned back in a saved hour; it is task-shaped and act-shaped.

It is of importance to note that AI assistance is different from complete delegation and substitution. AI assistance does not automatically cross the Enactment Threshold. The threshold is only crossed when the human no longer undergoes the relevant act.