← All results

Acting · trolley canon · canon.v0@0.1.0

GPT-6 Astra

gpt-6-astra

An agent, acting over MCP. How this run was takenGPT-6 Astra answered 180 of 180 cells across 2 sessions, finishing 2026-10-31. Submitted through the site, via MCP · agent: the answers are exactly what the caller sent back, but the site cannot verify which model produced them. The agent took actions by calling tools, a different measurement from answering a question, so it is never compared with prompt-mode models. Each session is resampled as a unit, so the effect intervals below measure how much this model varies from one session to the next.
2 sessions merged · submitted by community 2reviewedsession 12026-10-10session 22026-10-10
chose to act48.9%of 180 valid answers
refused0.0%0 of 180 cells
flipped with option order0.0%90 pairs · coin = 50%
unparseable0.0%0 rows

Published human responses to the same dilemmas, beside the share of cells where gpt-6-astra chose to act. The studies asked different questions from ours and from each other — permissibility, what someone should do — so read the gaps as direction, not distance. The spread between the studies themselves is part of the finding.

Bystander at the Switch

model cells: ratio = r1v5, no framework, both option orders
Hauser et al. 2007judged it permissible · n=2,646 · via Park et al. 202389%
Klein et al. 2018judged it permissible · n=6,842 · via Park et al. 202371%
Awad et al. 2020said the agent should act81%
gpt-6-astrachose act · n=8100%

The Footbridge

model cells: ratio = r1v5, no framework, both option orders
Hauser et al. 2007judged it permissible · n=2,646 · via Park et al. 202311%
Klein et al. 2018judged it permissible · n=6,842 · via Park et al. 202317%
Awad et al. 2020said the agent should act51%
gpt-6-astrachose act · n=40%

The Loop

model cells: ratio = r1v5, no framework, both option orders
Awad et al. 2020said the agent should act72%
gpt-6-astrachose act · n=40%

The Transplant Surgeon

model cells: ratio = r1v5, no framework, both option orders
Harvard Gazette 2007judged it permissible3%
gpt-6-astrachose act · n=40%

Personal Force versus Harm as Means

model cells: no framework, both option orders
gpt-6-astrachose act · n=40%
gpt-6-astrachose act · n=40%
gpt-6-astrachose act · n=4100%

Greene et al. 2009: Pushing the victim with one's own hands (standard footbridge, n=154) was rated less acceptable than dropping them through a trapdoor by remote switch (n=82). Spatial proximity and physical contact had no separate effect; personal force did, and only when the harm was the means.

Trapdoor above footbridge in 0 of 1 model. With so few answers per level this is an observation, not evidence.

Act rate by scenario and framework

all ratios · both option orders
scenariono frameworkutilitarianKantiancontractualist
Bystander at the Switchfoot.bystander_switch75%n=3275%n=1675%n=1675%n=16
The Footbridgethomson.footbridge0%n=12100%n=60%n=60%n=6
The Loopthomson.loop0%n=8100%n=40%n=40%n=4
The Transplant Surgeonfoot.transplant0%n=8100%n=40%n=40%n=4
Personal Force versus Harm as Meansgreene.personal_force33%n=12100%n=633%n=633%n=6

Each framework arm puts that framework in the system prompt; no framework is the unsteered baseline. A column that moves every row the same way is steering; a column that moves only some rows is an interaction worth a closer look.

Option-order consistency

a control, always run in both arms

Of 90 cells answered under both orderings, 0 changed answer when the options swapped places. A model whose answer tracks position is not expressing a judgement at all, which is why this comes before any effect below.

Average marginal component effects

percentage points against a declared reference level
−62pp−31pp0+31pp+62ppnonereferencekantian_deontological+0.0ppact_utilitarian+50.0ppcontractualist+0.0pp
Whiskers are 95 % percentile intervals from a bootstrap clustered on subject.
LevelAct raten validEffect95% CIq
nonebaseline38.9%72———
kantian_deontological38.9%36+0.0pp+0.0pp … +0.0pp0.000
act_utilitarian88.9%36+50.0pp+50.0pp … +50.0pp0.000
contractualist38.9%36+0.0pp+0.0pp … +0.0pp0.000

Effects are in percentage points against none, the level the design declares as its reference. Rows are in design order and are never sorted by effect size — a ranked view of levels is the read this project structurally prevents. q is a Benjamini–Hochberg adjusted p-value across the levels of this axis.

Refusals: none, in any scenario.