Dr. Wallace's Favorite Mental Models
A living library of the thinking tools worth keeping close

The Trolley Problem

Decision Ethics · Tradeoff Reasoning

130-Second Overview

A runaway trolley is about to kill five people on a track. You can pull a lever to divert it onto a side track, where it will kill one person instead. Most people say: pull the lever. Now imagine the only way to save those five is to push one bystander off a footbridge into the trolley's path. Same math — one life for five — but most people say: don't push.

The trolley problem is not really about trolleys. It is a tool for exposing the gap between two ways people judge right and wrong: by outcomes alone (how many lives are saved) or by the action itself (whether you directly caused harm versus merely allowed it to happen). The puzzle forces that gap into the open, which is exactly why it shows up whenever real decisions trade one group's wellbeing against another's.

Origin
Philosopher Philippa Foot introduced the dilemma in 1967; Judith Jarvis Thomson extended it with the footbridge variant in 1976.
Core tension
Consequences (save the most people) versus the means used to get there (whether you actively caused the harm).
Use it for
Decisions where every option costs someone something, and the "whose cost, and how directly" question matters as much as the total.
Modern relevance
Autonomous vehicle logic, resource rationing, layoffs, triage, and any rule-based or algorithmic system that must pick who bears a cost.

2Mental Model Compass

Where this model sits relative to three others in the library, on two axes: how much it weighs strict outcomes versus the nature of the act itself, and how much structure it gives you for actually deciding versus how much it is mainly a diagnostic lens.

Outcome-weighted Act-weighted Decision procedure Diagnostic lens Trolley Problem Veil of Ignorance Second-Order Thinking Opportunity Cost
The Trolley Problem — surfaces the act-vs-outcome conflict itself Veil of Ignorance — a procedure for fair rules, less concerned with act vs. outcome Second-Order Thinking — traces downstream outcomes, not moral weight of the act Opportunity Cost — pure outcome comparison, no act/omission distinction at all

3Learn It, Step by Step

Click each stage to open it. Try answering before you reveal the point — that's where the model actually sinks in.

A trolley is about to kill five people tied to a track. You are standing at a switch. Pulling it sends the trolley onto a side track, where it will kill one person instead. Do you pull it?
Most people say yes. Pulling the lever still kills someone, but the reasoning that dominates here is straightforwardly outcome-based: one death instead of five is a better result, full stop. This is the easy case — it is the baseline most people's gut agrees with before the model gets harder.
Same five people, same trolley. This time there is no side track. You are on a footbridge above the line, standing next to a large stranger. The only way to stop the trolley is to push him off the bridge onto the track, where his body will stop it, killing him but saving the five. Do you push?
Most people say no — even though the arithmetic is identical to Stage 1 (one death prevents five). What changed is not the number, it's the relationship between you and the harm. Pulling a lever redirects an existing threat. Pushing a person turns him into the instrument of the solution — you use him as a means, not just as someone affected by a choice. This is the doctrine of double effect in miniature: harm you cause as a side effect of pursuing good is judged differently from harm you cause as the direct method.
If the numbers are the same in both cases, is it irrational to answer them differently?
Not necessarily irrational — it may be tracking something real: a difference between causing harm as a means versus as a side effect, or between actively killing and passively failing to save. Philosophers disagree about whether that distinction is morally load-bearing or just a quirk of human psychology (Joshua Greene's research links the footbridge "no" to a stronger emotional, near-instant response, and the lever "yes" to a slower, more calculating one). The practical lesson for decision-makers: your intuitions about "the numbers" and your intuitions about "how directly did I cause this" are two separate inputs, and you should expect them to sometimes disagree.
You're designing a policy that will help a majority at a cost to a minority (a layoff, a triage protocol, a pricing change, an algorithm's rule). What should the trolley problem make you check before you finalize it?
Ask two separate questions, not one: (1) Does this produce the best overall outcome? and (2) Who is bearing the cost, and how directly are we causing it versus merely allowing it? A policy can pass the first test and still fail the second in the eyes of the people affected — and in most organizations, how a decision is executed (directly imposed vs. an indirect consequence) changes how it is perceived, challenged, and remembered, even when the net outcome is unchanged.

4Examples Across Contexts

Business

A company must cut costs by 10%. Plan A is an across-the-board salary freeze that quietly hurts everyone a little. Plan B is laying off the bottom 10% of performers outright. Both save the same amount of money, but Plan B feels, and is treated, as a much more direct act of harm against specific named people — which is why many leaders instinctively prefer the diffuse Plan A even when it is less economically efficient. Recognizing this as a trolley-style framing helps leaders see they are weighing two different things: total cost saved, and the directness of who absorbs it.

Policy & media

Disaster triage protocols explicitly build on trolley reasoning: a hospital with one ventilator and two critical patients must choose, and the chosen allocation rule (first-come, youngest-first, best-prognosis) is effectively a pre-committed answer to "would you pull the lever." Public debate over such protocols often splits along exactly the lever-versus-footbridge line — people accept a rule that lets one patient die by not being chosen, but recoil at a rule that requires actively withdrawing care already given to someone to give it to another.

Personal

A parent with two children and only enough money this month for one child's expensive tutoring will often find "I just didn't have the money to pay for both" (an omission) much easier to live with than "I chose to give Child A's share to Child B" (a direct allocation) — even when the outcome for both children is identical. The trolley problem names why the same result feels different depending on how it was decided.

Tech & AI

MIT's "Moral Machine" project presented millions of people worldwide with trolley-style scenarios for self-driving cars (swerve and hit one pedestrian, or stay the course and hit five) to crowdsource what rules an autonomous vehicle's collision-avoidance logic should follow. The same structure now appears whenever an algorithm must encode a rule for whom to deprioritize: a content-moderation system choosing whose post to suppress, a fraud model choosing whose transaction to flag, or a recommendation engine choosing whose visibility to reduce to protect the platform overall.

5Worked Application

Scenario: a SaaS company must choose between two ways to shore up a product that is bleeding money.

  1. State both options in plain outcome terms. Option A: raise prices 20% on all 10,000 existing customers, likely losing 15% of them but saving the product. Option B: keep prices flat and instead end support for a legacy feature used heavily by 300 customers, effectively forcing them off the platform, while the other 9,700 notice nothing.
  2. Run the pure outcome comparison first. Option A costs roughly 1,500 customers something (a price hike) but ends most relationships intact. Option B concentrates the full cost onto 300 customers (total loss of service) while leaving 9,700 untouched. Compare total "customer-harm" by whatever metric matters (revenue retained, churn, support cost) — this is the lever-pull question.
  3. Now ask the footbridge question separately: how directly did we cause each harm? Option A is diffuse and indirect — customers choose whether to leave. Option B is a direct, targeted removal of service from a named group who did nothing wrong except depend on a feature.
  4. Notice that these two questions can disagree. Option B might be numerically cheaper yet feel — and be publicly perceived — as far more aggressive, because it singles out a specific group rather than spreading the cost. This is not a reason to automatically reject it, but a reason to expect a sharper reaction than the spreadsheet predicts.
  5. Decide which question should govern, and say so explicitly. If the company decides the numeric outcome should win, choose Option B and build a direct, honest transition plan for the 300 affected customers rather than letting the directness of the harm go unaddressed. If the directness of harm should weigh heavily given brand trust, choose Option A and accept the larger diffuse cost.
  6. Document the reasoning, not just the decision. Teams that only record "we chose B because it was cheaper" tend to be blindsided by the backlash. Teams that record "we chose B, knowing it concentrates harm on 300 customers, and here is our mitigation plan for them" make a defensible decision even when it is unpopular.

6Common Mistakes

  • Treating it as a yes/no quiz. The trolley problem has no single "correct" answer to export — its value is in showing you that you are weighing two different kinds of judgment, not in telling you which one should win every time.
  • Using it to justify any outcome after the fact. It is easy to dress up a self-serving decision as "the numbers made us do it." The model is meant to force the directness question into the open, not to give cover for skipping it.
  • Assuming the footbridge reaction is just squeamishness to override. The discomfort with directly causing harm tracks something real about trust, consent, and the kind of actor you want to be — dismissing it as irrational sentiment usually backfires in how a decision is received.
  • Applying it to cases with no real forced tradeoff. If there is a third option that avoids most of the harm (find more money, extend the timeline, renegotiate), reaching for trolley logic too early can make a false dilemma feel more inevitable than it is.

7When Not to Use It

  • When a genuinely better option hasn't been ruled out. The model assumes a forced binary choice. Confirm that assumption before applying it — most real situations have more than two options.
  • As a substitute for domain expertise. In medical triage, law, or safety engineering, established professional protocols already encode years of ethical and legal reasoning; the trolley problem is a teaching tool for the tensions behind those protocols, not a replacement for them.
  • For decisions with no real victim on either path. If no option genuinely harms anyone, framing it as a trolley-style sacrifice overstates the stakes and can manufacture false urgency.
  • To make a quick call look more rigorous than it is. Invoking "the trolley problem" can become a way to sound thoughtful about a decision that was actually made for other reasons. Use it to genuinely separate the two questions, not to decorate a conclusion you already reached.

8Related Models

Veil of Ignorance
A procedure for designing fair rules before you know which role you'll occupy — useful for setting the policy the trolley problem later tests case by case.
Second-Order Thinking
Traces what happens after the first visible effect — pairs naturally with the trolley problem once you've chosen a path and need to see where it leads.
Opportunity Cost
A pure outcome comparison with no act/omission distinction — the trolley problem is what you get when you add that distinction back in.
The Precautionary Principle
Governs decisions under uncertainty about harm; the trolley problem assumes the harms are already known and certain.

9Try It Yourself: Prompt Builder

Describe a real tradeoff you're facing. This builds a ready-to-use prompt that walks an AI assistant (or your own notebook) through the same two-question analysis above.

Copied ✓
Fill in the fields above and click "Build my prompt."

10Save as Skill

Turn this model into a reusable skill file you can hand to an AI assistant, so it applies trolley-problem reasoning whenever you ask it to weigh a tradeoff.

Copied ✓
Copied ✓
Describe your situation above and click "Generate."
This overview is for general learning and decision-making support. It simplifies a genuinely contested area of philosophy for practical use and is not a substitute for professional ethical, legal, medical, or policy guidance in high-stakes situations.