Cluelessness: can we know the effects of our actions?
Many people argue that the effects of our actions are so diverse and unpredictable that it’s impossible to know whether they’re ultimately good or bad — so it doesn’t make sense to struggle to increase our impact.
The pop version of the criticism is that if you save a life, that person might become the next Hitler — so your well-intentioned action might have actually resulted in the deaths of millions.
This version of the criticism is silly: although there’s some chance that the person whose life you save will be the next Hitler, it’s so low that the possibility doesn’t have much effect on the expected value. And just as you might save the next Hitler, you might also save the next Norman Borlaug, who saved hundreds of millions of lives.
But the silly version of the argument does point at something real.
Even if we set aside extreme scenarios like saving baby Hitler, every action we take causes ripple effects that extend indefinitely into the future.
For instance, if you save a life, that person will likely have children (which could be good or bad for that child), and then those people will have children, and so on and so on.
These ripple effects could have either enormously good or bad consequences — and we don’t know what they’ll be. This appears to make people who want to do good “clueless” about the ultimate effects of their actions.
How might we respond to this argument?
An initially attractive response is ‘cancelling out.’ Suppose we know that an action has some positive short-term effects, but also many unknown effects. Because the unknown effects are unknown, they’re just as likely to be good as to be bad. That means that in expectation they’re zero. So, in practice, we can (probably) ignore the unknown effects and just count the known positive effects.
To apply that to our earlier example, if you save someone’s life, that person is just as likely to go on to do a huge amount of good as they are to do a huge amount of harm — so these potential longer-term effects cancel out, and you can focus on the known goodness of saving that individual’s life.
However, not everyone thinks this response works. Philosopher Hilary Greaves has proposed a more advanced version of the puzzle, which she called “complex cluelessness.” She has an argument to suggest that we can’t be confident the unknown effects indeed cancel out. I won’t sketch this argument here – see her talk for a description – but let’s assume it’s true for now.
If this argument is correct, it leads us to a radically sceptical position: that we can’t know whether the effects of any actions are ultimately good or bad from an impartial perspective.
This conclusion would apply well beyond people interested in 80,000 Hours, effective altruism, or longtermism. Rather, it would apply to any attempt to appeal to consequences. For instance, it would mean that we don’t know whether helping an old lady across the street is ultimately good or bad for people; whether the government policies you support are good or bad; or whether it’s typically good or bad to lie, cheat, and murder.
How might we respond?
One option is to give up on impartiality. If you have a narrow ethical circle (for instance, just your friends and family, or just people alive today), then it’s still practical to track the effects of your actions.
However, giving up on impartiality means we would be indifferent to the suffering of anyone not in our narrow circles, which we find deeply unattractive.
We think the right response is to be extremely humble about what we can ever know, but then do our best to work out which kinds of actions have the best long-term effects (those that don’t cancel out), and focus on those.
Professor Greaves doesn’t accept the sceptical conclusion either. She’s still not sure what the right conclusion is, but thinks “going longtermist” is one promising route. This means caring about all the effects of our actions — including the long-term ones — and then doing our best to figure out the most important drivers of positive long-term effects.
For instance, Nick Bostrom has argued that whether an intervention increases or decreases existential risk is the most important determinant of its long-term impact. The hope would be that although most of the effects of our actions are unknown, we might still be able to assess actions in terms of whether they increase, decrease, or have no effect on existential risk, and therefore still make progress in finding the best actions.
There are other schools of longtermism that focus on other key drivers of long-term effects, and we hope that we’ll learn more about the best proxies for long-term impact as those schools develop.
One interesting implication of this response to cluelessness is that it means there are ultimately no ‘high-confidence’ or ‘evidence-based’ ways to do good — at best, we can rigorously measure the short-term effects of a course of action.
But, as we’ve shown, these are only a tiny minority of all the consequences, and we can’t simply ignore all the long-term effects. Rather, we need to use our judgement to assess the long-term effects too. And because the long-term effects comprise the vast majority of all the effects, this is where the action is. Focusing only on ‘evidence-based’ interventions means ignoring most of what matters.
This debate is a long way from settled. If you’d like to learn more about the responses to cluelessness, we recommend Professor Greaves’s excellent talk, which you can watch or read a transcript of here: Evidence, cluelessness, and the long term.
Learn more
- Video and transcript: Evidence, cluelessness, and the long term by Hilary Greaves
- Podcast: Philosopher Hilary Greaves on moral cluelessness, population ethics, probability within a multiverse, & harnessing the brainpower of academia to tackle the most important research questions
- Podcast: Tackling the ethics of infinity, being clueless about the effects of our actions, and having moral empathy for intellectual adversaries, with philosopher Dr Amanda Askell
- Cluelessness by Hilary Greaves
Read next
This is a supporting article in our advanced series. Read the next article in the series, or here are some others you might find interesting:
