Lesson 3: Moral uncertainty
Why moral uncertainty matters for prioritisation
So far in this course, all our uncertainties have been empirical—what will happen, how effective an intervention is, what the probability of success is. In principle, better evidence could resolve them.
But some uncertainties aren’t like that. “How much does chicken suffering matter compared to human suffering?” isn’t a question more research can answer. It’s a question about values, not facts. This is moral uncertainty—and it’s the main reason thoughtful people using the same ITN framework end up with different cause priorities.
Consider two people who agree on every empirical fact about factory farming—the number of animals, the severity of their conditions, the tractability of welfare campaigns. They disagree about one thing: how much moral weight to give animal suffering. One gives chickens 1/100th the weight of a human. The other gives them 1/10,000th. That single moral disagreement flips their entire cause ranking.
The same applies across cause areas. How much you weight future generations determines whether existential risk reduction is the overwhelming priority or a speculative sideshow. How you think about rights versus welfare determines whether you fund corporate campaigns or abolitionist movements. These aren’t empirical questions—they’re moral ones. And being uncertain about them is entirely reasonable.
Here’s why it matters practically: if you ignore your moral uncertainty and just act on your favourite moral view, you might be making a serious mistake. You might be taking a big moral risk for a small gain—confident enough to act, but not confident enough to be safe.
Three possible approaches to ethics
Philosophers have developed various systematic frameworks for thinking about right and wrong. Three are listed below. You don’t need to pick one forever—most thoughtful people have some credence in several. But understanding what they say helps you see why moral uncertainty exists: these frameworks genuinely pull in different directions on questions that matter.
Consequentialism. What matters is outcomes. An action is right if it produces the best consequences—typically the most wellbeing and least suffering, summed across everyone affected. If animals can suffer, their suffering counts. If future people matter, preventing extinction has enormous value. This view naturally leads toward cause areas with the largest expected impact.
Deontology. What matters is following the right rules or respecting rights. Some actions are wrong regardless of their consequences. On this view, you can’t justify harming someone by pointing to good outcomes for others. Applied to animals: if animals have rights, using them as resources violates those rights regardless of how well we treat them. Applied to prioritisation: there may be moral constraints on how you pursue impact, not just which cause area you choose.
Virtue ethics. What matters is character. Act as a person with good character traits—compassion, justice, courage—would act in your situation. This view is less prescriptive about cause priorities but counsels practical wisdom: the right action depends on circumstances, and rigid frameworks can miss what a wise person would see.
These frameworks converge on easy cases (torturing animals for fun is wrong on all of them) but diverge sharply on the questions that drive cause prioritisation. That divergence is the source of moral uncertainty.
Working through moral uncertainty: expected choiceworthiness
In Moral Uncertainty (MacAskill, Bykvist & Ord, 2020), the authors argue that one approach is to treat moral uncertainty analogously to empirical uncertainty—putting credences on different moral views and weighting your decision across them. When you’re empirically uncertain, you calculate expected value: probability × value for each outcome. The analogous move under moral uncertainty looks like this.
The key concept is choiceworthiness—how good or bad an option is according to a particular moral view. Different views will assign different choiceworthiness to the same action.
Expected choiceworthiness works like this: for each option, multiply its choiceworthiness on each moral view by your credence in that view. Sum across all views. Choose the option with the highest total.
Worked example. You're choosing between funding welfare campaigns for factory-farmed chickens (Grant A) and vegan advocacy and abolition work (Grant B). For each moral view, ask two separate questions: how confident are you that this view is true, and how would a committed proponent of it rate the grants on a scale from -10 to +10, where the endpoints are paradigm cases of right and wrong action on that view?
Consequentialism. Say you have 50% credence that consequentialism is the correct moral framework. A pure consequentialist cares only about expected aggregate welfare—suffering reduced and wellbeing increased, summed across everyone affected. How would they rate the grants?
Grant A: +8. Welfare reforms operate at enormous scale (billions of birds in factory farms), have a solid track record (cage-free commitments, the Better Chicken Commitment, slower-growing breeds), and produce meaningful per-animal suffering reductions with high near-term confidence. Not +10 because reforms don't end factory farming—the underlying suffering machine keeps running.
Grant B: +4. If abolition succeeded, the welfare gains would dwarf anything reforms could achieve. But the probability of success on any relevant timescale is low: decades of vegan advocacy haven't substantially shifted consumption. The expected value is positive but heavily discounted by uncertainty.
Deontology. Say you have 30% credence that a rights-based view is correct. A committed deontologist holds that animals have rights factory farming systematically violates, and that the wrongness of those violations doesn't reduce to the suffering they cause. How would they rate the grants?
Grant A: -4. Welfare reforms accept the premise that animals can be used as resources and work to make that use less painful. From a rights perspective this is doubly problematic: it doesn't address the underlying violation, and by making the industry appear humane it can entrench public acceptance of animal use. The intention is decent but the action is wrongly directed—hence negative, though not at the bottom of the scale.
Grant B: +8. Abolition targets the rights violation directly. It treats animals as ends rather than means and aims at the kind of structural change the view says is required. Not +10 because vegan advocacy is one tactic among many for pursuing abolition, and its effectiveness is uncertain—but the direction of the action is exactly what the view prescribes.
Virtue ethics. Say you have 20% credence that virtue ethics captures what really matters. A virtue ethicist asks what someone of good character—someone who embodies compassion, justice, courage, and practical wisdom—would do. How would they rate the grants?
Grant A: +5. Compassion responds to suffering that exists now and can be reduced now; the compassionate person doesn't tell a suffering being to wait for systemic change. But practical wisdom asks whether this expression of compassion accepts too much of what shouldn't be accepted.
Grant B: +6. Engages justice (refusing to accept a deep wrong) and courage (challenging a normalised practice) alongside compassion for animals not yet born into the system. Slightly edges out reform work, though a careful virtue ethicist would resist this neat ranking—their view is agent-focused and doesn't naturally produce action scores in this way. We're approximating.
Calculating expected choiceworthiness. For each option, multiply each view's verdict by your credence in that view, and sum:
Grant A: (0.50 × 8) + (0.30 × -4) + (0.20 × 5) = +3.8
Grant B: (0.50 × 4) + (0.30 × 8) + (0.20 × 6) = +5.6
Grant B scores higher—even though you think consequences are probably what matters most (50% credence). The 30% credence in deontology, which sees Grant A as actively complicit (-4) and Grant B as exemplary (+8), tips the balance. When the stakes are asymmetric, even a minority view can change your decision.
A caveat worth flagging: this approach assumes you can put +8 on the consequentialist view and +8 on the rights-based view on the same scale—that they’re measuring something comparable. That assumption is itself philosophically contested (it’s called the intertheoretic comparison problem) and a substantial portion of Moral Uncertainty is devoted to when it holds and what to do when it doesn’t (the authors discuss alternatives like Borda-style voting across views for cases where only ordinal comparisons are possible). For our purposes, treat the numerical worked example as a teaching tool that illustrates the asymmetric-stakes insight, not as a fully rigorous procedure.
The practical takeaway
When you’re making decisions with significant moral stakes, don’t just ask “what does my preferred moral view say?” Ask “how does this decision look across the range of views I find plausible?” If it looks really bad on one of those views—even a view you don’t give much credence to—that’s a reason to think twice.
This matters for the rest of this week. As you tour the cause areas below, notice how different moral commitments lead to different conclusions. The ITN framework structures the conversation, but your moral weights determine where you land.