Follows the framework's disambiguation conventions: "Tier N" always names a value tier or a goods tier; the emotion model's automaticity axis is referred to by processing level; no bare "Tier N."
What this document is
A frontier case for the framework: what to make of choosing immoral options in video games. It runs four linked questions through the apparatus — does a non-player character have stake; where is the harm in violent play; should you always choose the moral option; and why does a virtuous person recoil from a game like GTA5 when the empirical literature finds little or no link between game violence and real violence. The case turns out to be close kin to the consensual-incest scenario worked through elsewhere: a well-calibrated gut alarm firing into a situation engineered to remove the thing the alarm exists to catch.
The NPC: functional stakes, not felt ones
Start with the NPC, because once it is settled most of the rest follows.
The framework's load-bearing distinction is between felt stakes and functional stakes. A thermostat "cares" about temperature in a purely mechanical sense; its stakes are functional, and it is no aperture. A scripted game character is the same kind of thing one layer up. It is built to behave as though it has preferences — to flee, beg, pursue, retaliate — and it produces the outputs that preferences would produce without there being anything it is like to be it. On the propositional-attitude grid this is behavior at the operative/gut processing level with nothing behind it: reactions, not wants. The NPC does not have preferences. It only appears to.
So the question Danny raised — does the NPC have stake, or merely appear to — gets the second answer. No aperture, no stake-geometry, no mattering-from-the-inside. The framework is unusually firm here: behavioral sophistication is not evidence of felt stakes. A more elaborate NPC, even a convincingly grief-stricken one, is more impressive mimicry, not a stronger candidate aperture. It is the floor-mapping robot, not the bumble bee.
The uncertain-apertures principle does not rescue the NPC. That principle — treat a case as if an aperture might be present when the evidence is genuinely mixed — is for beings where researchers looking at the same data disagree: invertebrates, fetal stages, possibly some trained AI systems under second-order uncertainty. A hand-authored game character is not in that zone; it sits firmly on the thermostat side. The live edge, worth stating so the framework is not misread as complacent, is whether a sufficiently sophisticated learned agent could one day cross the line — and the framework's own machinery means that could not be settled by fiat. But no shipping game NPC is close.
Where the harm is — and is not
On the framework's terms, killing an NPC collapses no stake-geometry, because there was none there to collapse. There is no victim in the morally relevant sense. The "atrocity" has the shape of an atrocity with the bearer of stakes stipulated away.
This is why the empirical literature finds what it finds. Violence done to functional mimics does not transfer to violence done to apertures, because the two are not the same kind of act, and most players' fiction-firewall holds precisely because they know, at the level that governs action, that no one is there. The framework predicts the null result rather than being embarrassed by it. (The literature is genuinely contested — Craig Anderson's camp reads the effect sizes as real and Christopher Ferguson's as negligible — but the framework's argument does not need the strongest version of the null. It needs only that real-world violent behavior is not the variable where the action is, which both camps' best longitudinal and preregistered work tends to support.)
The crucial move is to notice what this leaves open. "Studies show no real violence" answers the crudest form of the worry — will the player go and do it — and the framework agrees that is the firewalled variable. It does not touch the framework's actual locus of concern, which is not behavioral transfer at all. That concern is developed below, and it is about the player, never the NPC.
The repulsion, re-read
Which makes the virtuous person's recoil the real puzzle, and here the case is almost structurally identical to the consensual-incest scenario.
The recoil from GTA5 is a disgust detector firing at the shape of the act — gleeful, gratuitous cruelty — without checking whether there is a fire behind the smoke. Disgust's core appraisal is primitive and fast; its home processing level is operative/gut, faster than thought. And the detector is well-calibrated to the real-world distribution, where someone who relishes simulated cruelty is at least sometimes rehearsing or revealing something about how they hold real apertures. The game is engineered smoke-with-no-fire: the form of atrocity with the victim removed by construction. So the recoil is informative without being a verdict.
In the framework's vocabulary, the disgust is a Tier 1 value — a parochial, network-level guardrail tuned to the ordinary case — not a Tier 2 principle registering a structural violation. Reading the recoil as proof of wrongness makes the same error the incest-dumbfounded subject makes: mistaking a heuristic for a principle. The dumbfounding is diagnostic. The reasons collapse because the person is reaching for the kind of deep structural justification that would make the act wrong for anyone anywhere, and finding nothing, because the rule was never that kind of rule.
There is a defensible reason to keep the blunt rule anyway, the same reason the incest doc gives: a norm that prevents real collapse in nearly every actual instance is worth maintaining as a bright line, and one does not want people running private case-by-case cost–benefit analyses on cruelty. That is a Tier 1 value justified by its Tier 2 track record. It is not a finding that the rigged case contains a wrong.
The two considerations that survive — both about the player
The framework does more than say "no victim, ignore your disgust." Two genuine considerations remain, and neither is about the NPC.
The slow structural channel. In the mood-modulation scaffold, behavior feeds back into the underlying stake-geometry over time: what you repeatedly enact updates what you are, not only what you experience in the moment. So the morally live question is not "did I harm the character" but "what am I practicing, and does it shape my recognition-fidelity — my capacity to register real others as apertures — and my responsiveness to them?" The worry worth taking seriously is not that the player will commit the acts (that is the firewalled, well-studied variable). It is whether habituated delight in the form of cruelty dulls responsiveness to the real thing. This is an empirical question the framework can frame but cannot settle, and for most players the firewall holds and the question is moot.
Expression. Running over a pedestrian in GTA5 is identical across two players; the geometry behind it is not. One is doing sandbox physics, dark comedy, or testing the toy. The other is using the game as the closest available proxy for something they actually wish they could do to real people. The framework locates the moral content in what the stance reveals and reinforces about the player's disposition toward apertures that do exist, not in the pixels. This is why relishing matters more than depicting: the question is which configuration is being rehearsed and what it expresses, which is a fact about the geometry rather than about the content on screen.
Should you always choose the moral option?
No — and the framework gives a clean reason. Because the in-game choice borrows none of its weight from nonexistent NPC stakes, its actual weight runs entirely through rehearsal and expression, and on both of those, choosing the wrong path in fiction can be elaborating.
Picking the dark branch to see the story, to understand what it is to be tempted, to feel the weight of a choice you would never make, to inhabit a character — these are uses of Imagination, Play, and Tragic Awareness, all Tier 2 goods, dimension-adding and reach-extending. A game that forces a moral reckoning by letting you do the wrong thing and sit with it (Spec Ops: The Line, parts of The Last of Us, Disco Elysium) does more moral work than one that only ever permits virtue.
The framework goes one step further, and it is the step most people miss. A person who can only ever choose the moral option, who cannot bear to inhabit a wrong choice even inside an obviously fictional frame, may have the thinner geometry — a coherence maintained by rigidity rather than by integration, low on the dimensionality that lets a mind hold incongruity without collapse. The capacity to inhabit a villain in a clearly fictional frame, without confusing it for the actual, is exactly what Imagination and Play are for. A geometry that can do this is wider, not narrower. Virtue here is not the refusal to touch the dark option; it is the intact firewall that lets you touch it without it leaking.
The bullet to bite
GTA5 is not only narrative, and part of its draw is unscripted gratuitous cruelty for its own sake. The framework should not pretend that is identical to a weighted narrative branch in Disco Elysium. But the distinction it draws still holds, and it is not the distinction most people reach for.
The line is not moral-option versus immoral-option, and not violent content versus clean content. It is whether the engagement is elaborating — bounded by the firewall, integrated into the geometry, dimension-adding — or whether it is a slow rehearsal of contempt for the very form of an aperture. The first is fine and possibly good. The second is the only place a real harm lives, and it is a harm to the player's own stake-geometry through the slow structural channel, not a harm to anyone on screen. The disgust the virtuous person feels is pointing at the second case; its mistake is firing at the first one too, because it cannot tell them apart from the outside.
Intellectual lineage
- Joshua Shepherd, Consciousness and Moral Status (2018) — the dependence of moral status on phenomenal consciousness; the spine of the felt/functional verdict on NPCs.
- Jonathan Haidt, "The Emotional Dog and Its Rational Tail" (2001) and The Righteous Mind (2012) — moral dumbfounding and the disgust-into-a-vacuum structure that the game case shares with the incest case.
- Tamar Gendler, "Alief and Belief" (2008) — the operative/gut-level attitude that locates the recoil, and the firewall that lets fiction be inhabited without belief.
- Kendall Walton, Mimesis as Make-Believe (1990) — fiction as a frame within which one can entertain stances not endorsed outside it; the philosophical backing for the firewall.
- Craig Anderson and Christopher Ferguson — the two poles of the games-and-aggression literature; cited here only to mark that the framework's argument survives whichever way that empirical debate settles.
Known weaknesses
- The slow-channel claim is empirical and unsettled. Whether habituated relish in simulated cruelty actually degrades recognition-fidelity is an open question the framework can frame but not answer. The whole "harm to the player" branch rests on an effect that may turn out to be small or absent.
- The elaborating/rehearsal distinction is hard to operationalize. From the outside, sandbox cruelty and proxy-cruelty look identical; the framework locates the difference in the geometry behind the act, which is exactly the thing that is hardest to observe. This risks becoming unfalsifiable in practice.
- The expression reading flirts with thought-policing. Grounding moral content in what a private stance "expresses" can slide toward judging people for fantasies that harm no one. The framework needs the firewall to be doing real work for this not to collapse into puritanism.
- It inherits the incest doc's rule-consequentialist flavor. "Keep the blunt disgust rule for its track record" justifies a norm by its consequences across cases, which sits awkwardly beside the framework's structural commitments elsewhere. The tension is named there and recurs here.