Drawing on his work in AI alignment, Eric Heinis - design engineer and founder of Ethic LLC - argues that encoding values at the semantic level is fundamentally unstable: any intelligence capable enough to construct a narrative justifying a rule-break will eventually do so. The architecture has to do the work, not the words.
His framework is spatial. Think of moral behaviour as paths through a landscape shaped by the systems we inhabit. In healthy conditions - high capacity, low rigidity, high reversibility - the terrain is relatively flat and people have genuine agency. But under pressure, the landscape deforms: valleys deepen into canyons, options narrow, and what’s even thinkable shrinks.
The design response is to preserve four conditions that keep moral emergence possible: adequate tempo for complexity to be metabolised; reversibility so that being wrong isn’t catastrophic; the legitimacy of withdrawal, recognising that inaction is sometimes the right choice; and tolerance for unresolved meaning, resisting the pressure to close questions before they’re ready.
Eric is a design engineer and founder of Ethic LLC, currently developing Raya, a relational AI architecture oriented toward meaning-making and wisdom.
