A Procedural Critique of Deontological Reasoning

Key insights

6 takeaways

01
A player's underlying maxim can only be inferred imperfectly from observed behavior, because of the problem of induction, so outside observers will often misidentify it; the game is built to make this difficulty visible.
02
The prototype is a mixed-initiative authoring tool in which rules are learned from player commands (reward, punish, kill, teleport, reverse, split, attract, repel) rather than predefined by designers, with a basic routine converting commands into persistent if-then rules.
03
Operational rules start absent, so collisions and breaches have no consequences until the player issues rule-changing commands, and unplanned play quickly produces unbalanced, degenerate rulesets such as runaway scores and chain-reaction splitting that freezes the engine.
04
The authors read these degenerate outcomes as evidence of a weakness in Kant's categorical imperative, while acknowledging the system remains an unfinished prototype.
05
Drawing on Bogost's procedural rhetoric and Sicart's ethics of game design, the work argues that moral maxims are often indeterminate and that acting morally under such conditions is hard enough that most play yields unengaging rulesets, treating unbalanced design as an ethical failing.
06
The authors suggest that viable development would require stronger induction methods (such as decision trees or falsificationist rule design) and a gamified goal structure, and they publish the prototype to encourage further work.

Summary

Togelius's prototype, presented at DiGRA 2011, is both a procedural implementation and a critique of Kant's categorical imperative. Its central thesis is that a player's underlying maxim, the principle guiding their actions, can be inferred from observed behavior only imperfectly. Because of the problem of induction, an outside observer will often misidentify that maxim, and the game is designed to make this difficulty visible. The work positions itself as a prototype authoring tool for mixed-initiative rule generation, in which a system learns rules from player commands rather than from predefined design.

The game's ontology consists of a player agent and colored "things" governed by simple physics and heuristics. Operational rules are initially absent: collisions and breaches carry no consequences until the player issues rule-changing commands such as reward, punish, kill, teleport, reverse, split, attract, or repel. A basic rule-induction routine converts these commands into persistent if-then rules. Unplanned play quickly generates unbalanced and degenerate rulesets, including runaway scores and chain-reaction splitting that freezes the engine. The authors read these outcomes as evidence of a weakness in the categorical imperative, while acknowledging that the system remains an unfinished prototype.

Analytically, the work draws on Bogost's procedural rhetoric to argue that the system exposes two points: the maxims of actions are frequently indeterminate, and acting morally under such conditions is so difficult that most play yields unengaging rulesets. The discussion connects these findings to Sicart's ethics of game design, treating unbalanced design as an ethical failing. The authors suggest that stronger induction methods, such as decision trees or falsificationist rule design, and a gamified goal structure would be necessary for the system to become a viable tool or game, and they publish the prototype to encourage further development.