How do we judge
right and wrong?
Dilemmas and the science of moral judgment
Today’s question
What can a moral dilemma tell us about the moral mind?
By the end
Explain Greene’s design. Interpret its evidence. Use a comparison to test an explanation of moral preference.
The switch
A runaway trolley will kill five people on the main track. You can divert it onto a side track, where it will kill one person.
Assume the outcomes are certain and these are the only options.
Paraphrase of the switch case in Greene et al. (2001), p. 2105; May, pp. 164–165.
The footbridge
The trolley will kill five people. From a bridge, you can push a stranger onto the track. Their body will stop the trolley, saving the five. The stranger will die.
Assume this will work, you cannot stop it with your own body, and no other rescue is possible.
Write: appropriate / inappropriate / pass. Add one reason.
Paraphrase of the footbridge case in Greene et al. (2001), p. 2105; May, pp. 164–165.
Two questions
Hailey asks: Could emotion shape my judgment without my noticing, and should learning this change my answer?
The psychological question
Why might a person approve the switch and reject the push?
The ethical question
Is there a morally relevant difference that justifies these judgments?
The design problem
The number of lives stays fixed, but lots of factors are confounded between these two cases.
A feature could also influence judgment by eliciting emotion.
Greene et al. (2001), pp. 2105–2107; May, pp. 163–165, 171–172.
Could an automatic response interfere?
Greene’s proposal: some dilemmas provoke a stronger emotional response against the proposed harm. Approving that harm may require overcoming the response.
A brief demonstration of interference
Name each ink color aloud:
Greene’s two predictions
Neural prediction
Greater engagement of regions previously associated with emotion during personal moral dilemmas.
Behavioral prediction
Longer responses when participants approve the harmful action in those dilemmas.
Greene et al. (2001), p. 2106; May, pp. 165–168.
What participants actually did
Experiments
Nine undergraduate participants in each experiment.
Dilemmas per participant
Personal moral, impersonal moral, and non-moral cases.
Text screens per dilemma
Two screens describe the case; the third asks about an action.
Dilemmas appeared in random order. Experiment 2 used a revised set. The experiment compared categories of cases; the trolley pair motivated the hypothesis.
Greene et al. (2001), pp. 2106–2108, notes 10–13 and 22–24.
“Personal” is defined operationally
An operational definition tells researchers how to sort or measure something.
A proposed action counted as personal when it:
- Could be expected to cause serious bodily harm;
- Harmed an identifiable person or group;
- Did so through something other than redirecting an existing threat.
Other moral cases went into the impersonal category. Non-moral cases included choosing travel or shopping options.
Paraphrase of Greene et al. (2001), note 9, pp. 2107–2108.
Read the contrast before the brain map
Black bars are personal moral dilemmas.
- Some regions showed greater relative activity; others showed less.
- The authors linked these regions to prior studies of emotion and working memory.
- The right middle frontal result did not replicate in Experiment 2.
Greene et al. (2001), Figures 1–2 and pp. 2106–2108. Figure extracts © 2001 AAAS.
Approving personal harm took longer
The reported response-time pattern fits the prediction of interference.
Within personal moral dilemmas, “appropriate” responses were significantly slower than “inappropriate” responses.
The other two conditions showed no significant response-type difference.
A later challenge to this result
Greene et al. (2001), Figure 3, p. 2107, © AAAS; McGuire et al. (2009).
Does slowing a decision change its verdict?
If emotion makes an answer take longer but leaves the answer unchanged, what has it influenced?
The decision process
A factor may affect response time, effort, or conflict while the same answer is reached.
The final verdict
To show that emotion changes approval, we need evidence about how judgments differ when emotional engagement changes.
What did Greene’s extra control address?
Experiment 2 included personal dilemmas in which approval was the majority response. Approval still took longer in the reported comparison.
This challenges the explanation that an unusual answer alone causes the delay. Difficulty, item selection, and the identification of emotion remain questions.
Greene et al. (2001), p. 2107 and note 24, p. 2108.
How far can the conclusion go?
| Claim | Assess it |
|---|---|
| A. The dilemma categories differed in measured activity. | Assess ASupported in these experiments. |
| B. Emotion helped produce the judgments. | Assess BSupported, but not definitive. Prior region-function evidence and the behavioral prediction contribute; correlation alone does not establish the causal role. |
| C. Emotion makes a judgment unreliable. | Assess CNot supported. |
| D. The study proves utilitarianism. | Assess DNot supported. But… |
Greene et al. (2001), especially p. 2107; May, pp. 157–162, 169–173.
Emotion can help a person decide
PCN: the somatic marker hypothesis
Learned affective signals can guide attention toward the costs and benefits of an option.
Patients with ventromedial prefrontal damage may retain many intellectual abilities while struggling with practical decisions.
May: inference can be automatic
An intuitive response may reflect learning about intentions, harm, and social rules.
Fast, emotional, learned, and reasonable are different descriptions. They need not come apart.
Jack, Kate, Kevin, Casey, William, and Emily.
PCN, pp. 340–345, especially 341–343; May, pp. 153–161, 171–173.
Moral capacities without a moral module
How can morality be a distinctive capacity if it uses circuits that also support other tasks?
Shared capacities
Understanding intentions, learning from harm, and regulating action can work together in moral judgment.
Questions paraphrased from Clare Diffin and Ava Vidal. May, chapter 6, pp. 151–160; connection to chapter 2, pp. 58–61. Comparison-group questions are methodological guidance.
From trolley problems to Moral Machine
Moral Machine asks people to compare outcomes of unavoidable crashes involving autonomous vehicles.
Its 2018 study reported roughly 40 million decisions from 233 countries and territories.
Open Moral Machine ↗External website · internet required
Look for the design
- What attributes vary together?
- What does a click measure?
- Who chose to participate?
Optional continuation: four classroom cases and a discussion of what preferences can tell us.
What should the vehicle do?
In every case, an empty vehicle’s brakes have failed. All people shown are adult pedestrians. The two outcomes are certain; everyone outside its chosen path survives.
Write your choice privately: stay / swerve / pass. Keep your reason until all four rounds are complete.
Record class tally
Enter aggregate counts from a show of hands. Leave blank until counted; use 0 for a counted category with no votes.
Original classroom cases. Counts are entered locally; the activity sends no responses to Moral Machine.
A choice needs a comparison
Percentages use stay + swerve responses only. Pass counts appear separately. Unrecorded rounds remain blank.
Rounds 1 & 2: number of lives
Swerve, then stay spares the larger group in both cases. A consistent route choice fits a different explanation.
Rounds 3 & 4: crossing signal
Swerve, then stay spares those crossing on the walk signal in both cases. Does changing the signal affect judgments?
What remains uncertain?
Would a majority vote settle the rule?
Suppose a majority prefers that a vehicle spare pedestrians crossing on the walk signal.
Therefore, the vehicle ought to be programmed to give them priority.
State a bridging premise
For example: majority preferences should govern this kind of design decision.
Test the premise
What about equal protection, the gravity of the violation, and obligations to people who did not choose the rule?
Connect this to Greene
Classroom argument; May’s debunking schema, p. 162. Greene et al. (2001), p. 2107, distinguishes explanation from moral prescription.
When should we distrust an intuition?
Greene’s later challenge
Automatic heuristics can mislead when applied to unfamiliar problems beyond the experience that shaped them.
Some judgments against sacrificing one for many may deserve less weight in those circumstances.
May’s reply
A novel story can involve familiar moral distinctions learned through experience.
Intuitions can change. Deliberation can also rationalize prejudice. Compare the reliability of the actual processes.
Can a feeling be wrong, and can it change?
Graham Ewers asks whether a correct choice can feel wrong. The feeling of rightness and moral correctness can come apart; defend the judgment with reasons.
Kate McDonough asks whether practice can change a response. Learned intuitions can be revised, but overriding a feeling does not itself improve a judgment. Training needs a defensible standard of improvement.
Greene’s later argument as presented and criticized in May, pp. 169–173. This argument goes beyond the conclusions of the 2001 experiment.
What did we learn from the choice?
Choose Greene’s experiment or our vehicle activity. Complete these three sentences:
The observation
“The participants…”
The interpretation
“One explanation is…”
The next step
“To conclude more, we would need…”
A good experiment helps distinguish rival inferences.
Sources
- Greene, J. D., et al. (2001). An fMRI investigation of emotional engagement in moral judgment. Science, 293, 2105–2108. Paper. Figures 1–3 are cropped from the assigned copy, with no data redrawn or values estimated.
- May, J. (2023). Neuroethics: Agency in the Age of Brain Science, chapter 6, assigned pp. 149–173. Chapter. Central connections: affect and inference, pp. 153–161; debunking, p. 162; sacrificial dilemmas and dual-process theory, pp. 163–168; reliability, pp. 169–173.
- Purves and colleagues. Principles of Cognitive Neuroscience, 2nd ed., chapter 10, assigned pp. 340–345. Somatic markers and decision making; limits of simple region-to-emotion assignments.
- Awad, E., et al. (2018). The Moral Machine experiment. Nature, 563, 59–64. Paper · Author-hosted PDF · MIT overview. Supplementary instructor source.
- McGuire, J., et al. (2009). A reanalysis of the personal/impersonal distinction in moral psychology research. Journal of Experimental Social Psychology, 45, 577–580. Paper. Supplementary qualification of the original response-time result.
- Crockett, M. J. (2013). Models of morality. Trends in Cognitive Sciences, 17, 363–366. Paper. Optional assigned reading; discussion extension follows.
Classroom trolley paraphrases and vehicle cases are teaching examples. Classroom tallies are separate from published data. The main lecture is sections 1–14 and ends with the Moral Machine demonstration. Sections 15–19 are optional continuation; sections 20–21 contain sources and the Crockett extension. Student questions are paraphrased from the Week 3 submissions reviewed September 15, 2026. Hailey Montz and Chloe-Jo Bassim were carried forward from Session 6.
What does the system compute?
Model-based
Uses a representation of how actions lead to outcomes to evaluate what may happen next.
“If I do this, what follows?”
Model-free
Uses values learned for actions in particular situations from their reinforcement history.
“This action has been bad in situations like this.”
Pavlovian
Favors approach or withdrawal in response to predicted rewards or harms.
“Approach this; pull back from that.”
Crockett proposes interactions among these systems. A philosophical theory and a psychological mechanism need not map one-to-one.
Crockett (2013), pp. 363–366. This is a proposed explanatory framework, not a result of Greene’s 2001 study.