The Incremental Slide

In the AI Story

The slide's danger is its invisibility. Unlike a single dramatic decision, which can be evaluated against one's values and refused, the slide presents each step as a minor extension of the previous step. The comparison is never between the current self and the original self — that comparison would trigger resistance. The comparison is between the current step and the last step, which is always small, always manageable, always reasonable given what has already been accepted.

Glover observed the mechanism in dozens of cases where ordinary people ended up participating in extreme harm. The Rwandan genocide proceeded through language first — the reclassification of Tutsis as inyenzi, cockroaches — before it proceeded through violence. The language deposited the sediment that made the violence thinkable. The Nazi bureaucratization of murder proceeded through administrative euphemism, each euphemism a small semantic adjustment that preserved the moral self-image of the functionaries processing the paper.

In the AI context, the slide operates through a sequence of adoptions. The first use of Claude Code to generate boilerplate feels benign — the builder retains judgment, direction, understanding. The second use expands to features the builder understands but finds tedious. The third expands to domains the builder does not fully understand — a backend engineer generating frontend code she could not write herself, deploying logic she has not independently verified. Each step is marginal. Each step seems reasonable. The cumulative trajectory — from tool-assisted tedium reduction to tool-dependent production in domains the builder cannot comprehend — is a gradient of diminishing moral ownership.

Segal's own confession in The Orange Pill fits the pattern: the acknowledgment that he built products knowing they were addictive by design, understanding the engagement loops and dopamine mechanics, because the technology was elegant and the growth was intoxicating, and because someone else will build it if I do not. Each justification is a step. Each step makes the next one easier. The self that emerges at the bottom of the slope is not the self that stood at the top, and the difference is invisible from inside.

Origin

The concept emerged from Glover's work on how ordinary people came to participate in the twentieth century's atrocities. Interviewing perpetrators and reading their testimony, he noticed the uniform absence of dramatic moral crossings. No one described a moment of decision. Everyone described sequences — small accommodations that accumulated, small numbings that deepened, small justifications that cleared the path for larger ones.

Related concepts exist in moral psychology — moral licensing, foot-in-the-door compliance, Milgram's observation that the incremental structure of his experiments produced obedience that would have been refused in a single step. Glover's contribution was to integrate these mechanisms into a unified framework of moral identity formation: the slide is not just a compliance effect, it is a self-construction event. The person at step seven is a different person from the person at step one, and the difference is what made step seven available.

Key Ideas

Slope, not cliff. The geometry matters. A cliff produces resistance; a slope does not.

Local comparison only. Each step is evaluated against the previous step, never against the original baseline. This is what makes the gradient invisible.

Sediment enables descent. The self reshaped by step one is a self for whom step two is easier. The descent accelerates.

Language comes first. In Glover's cases, linguistic change — the reclassification of persons, the adoption of euphemism — reliably preceded behavioral change. The vocabulary prepares the ground.

Interruption requires external vantage. Because the slide is invisible from inside, interruption requires mechanisms that introduce the distant view — mandatory review against original values, contact with persons who have not traveled the slope, institutional structures that restore the comparison to baseline.