Concerns with Longtermism
I am fairly new to EA, but as I read What We Owe the Future and related topics, the issues here strongly impress themselves on me. I welcome feedback, whether you can clarify why I am wrong, or offer support.
Discounting future good
MacAskill says (p. 11) “I am not claiming that the interests of present and future people should always and everywhere be given equal weight.” However, he does not so far as I know talk about a discount rate for future good versus present good, even though economics clearly does discuss this.
Imagine that I can donate to save a life today for $4000. And imagine that I can invest that $4000 and earn 2% (for example) above the rate of inflation. Then in 35 years, I would have $8000, and could save two lives. If those future lives are worth the same as the life today, then of course I should never save a life today, but invest to save twice as many lives 35 years from now. But by the same logic, I shouldn’t save lives then, but keep investing. This leads to the absurd conclusion that I should never donate to save lives, as I could always do more good in the future.
The logical conclusion is that I should discount future good I might do by at least the real rate of investment return. I should value saving a life today twice as much as saving a life in 35 years, assuming my 2% figure.
Uncertainty
There is another way in which good I plan to do in or for the future should logically be discounted. Assume I know a charity which has a proven track record of saving a life for $4000. I can have a high degree of confidence that my contribution of $4000 will save very close to one life. But instead, I decide to invest my money to save two lives in 35 years. There is no certainty that I can confidently save two lives with my $8000 35 years from now. Opportunities might be better or worse. But uncertainty is not as valuable as certainty when trying to do good.
This is especially true if I am taking an action (say, donating $4000) today in a way I believe will save two lives in 35 years. The organization to which I donated might collapse, or the problem they are addressing might become harder to solve, or wars might break out rendering them incapable of saving those lives. So even more strongly, the further in the future I am trying to do good by an action now, the less confident I can be that my planned outcome will be successful.
Imagine a philanthropist 140 years ago planning to make the world better in 2022. In the year 1882, she might decide that cleaning up manure from horse-drawn wagons in city streets would make life vastly better. She has no way of anticipating two world wars, automobiles, airplanes, telephones, computers, the Internet, or many other changes in the world that lead the problems of our time to be very different, or at least need different tools, than the problems she sees as most critical. I think we greatly over-rate our ability to see what the future looks like. Simply read any 50-year predictions from the past and you will see this. As a result, I would suggest that an appropriate discount factor should be applied for uncertainty in the future as well.
Implications for longtermism
Let us assume that I am able to make an intervention of some kind that has a certain benefit, say adds one QALY to some individual, every year from now on as long as people are around. How does my impact change depending on how much longer humanity survives?
Using just my conservative discount rate of 2%, my intervention this year is worth 1 QALY, when the result for next year adds .98 to make 1.98 QALYs. The third year adds .9604, making 2.9404 QALYs. On year 194, I surpass 49 QALYs. But no matter how far in the future I go, I will never surpass 50 QALYs.
This simple and practical perspective says to me that while the future is absolutely important to attend to, especially the near future, there is no benefit whatsoever in trying to analyze a good for more than 200 years. In fact, because of uncertainty, it appears to be overoptimistic to assume my actions today can have meaningful impact even at this timeframe.
Climate Change
In the Center for Climate and Energy Solutions document “Discounting the Benefits of Climate Change Mitigation: How Much Do Uncertain Rates Increase Valuations?” it is said, “Underlying existing climate change models is a specific set of assumptions regarding emissions levels, economic growth and flexibility, technological innovation, climate change policies, and the magnitude of climate change damages. Though this set of assumptions varies from model to model, each includes a discount rate, which is used to compare costs and benefits over time. The discount rate tells us how high future benefits need to be to justify spending a dollar today. While there is considerable debate regarding the appropriate discount rate to apply to any cost-benefit analysis conducted across generations, most climate models choose one rate (2-7 percent is a common range) and hold it constant over the time horizon of the model.” (https://www.c2es.org/document/discounting-the-benefits-of-climate-change-mitigation-how-much-do-uncertain-rates-increase-valuations/#:~:text=While%20there%20is%20considerable%20debate,time%20horizon%20of%20the%20model.)
Uncertainty is also the reason that while the effects of climate change will be much greater in the 22nd century than they are in this century, analyses almost all only go out to the year 2100.
The topic of discount rates for climate change is widely discussed, for example in “The Choice of Discount Rate for Climate Change Policy Evaluation” (https://media.rff.org/documents/RFF-DP-12-43.pdf) by Lawrence H. Goulder and Roberton C. Williams III.
Conclusion
My point in the Climate Change example is simply to show that this discounting approach is mainstream and worthy of being considered. Based on any reasonable discount rate, it would appear to me that there is no meaningful way in which considering the good we might accomplish for any time more than 200 years in the future (and realistically, any time more than 100 years in the future) can be meaningful as a basis of making moral decisions.
The logic here seems off. Suppose we discover a new unusually profitable investment that gets you a real rate of return of >100%[1]. Will you suddenly update towards believing that saving a life next year is only half as valuable as saving a life today? More broadly, it seems suspicious that you should peg your (moral) discount rate to be always greater than (empirical) rates of returns, regardless of the facts on the ground. Like surely at some level of investment returns donating later is more valuable than donating now?
Which is plausible for some human capital interventions, for example getting training for job negotiations.
We should take into account the possibility of having more influence in the future (e.g., from investment or learning more) and the possibility of having less influence in the future (e.g., from uncertainty about the future or the best opportunities becoming unavailable). This doesn’t mean that we should morally value future good differently from present good. (At 2% discounting, a single life 2000 years ago was far more important than everyone living today, which is an unappealing conclusion to me.)
There’s some literature on why we should basically just use the most pessimistic case for discounting, since it dominates over expected future impact. And that likely means we should do no discounting.
Could you please point me to some? Thanks!
There’s some arguments in favor and against time discounting linked from here (and more scholarly sources linked from those links): https://forum.effectivealtruism.org/topics/temporal-discounting
I’ve never personally looked into discount rates in any depth, and see the entire topic as rather beside the point; Scott Alexander best explains why: https://forum.effectivealtruism.org/posts/KDjEogAqWNTdddF9g/long-termism-vs-existential-risk
It’s not that a life 2000 years ago was more important than everyone living today, but rather that someone 2000 years ago trying to do good would be more effective trying to save one life that year than trying to save humanity 2000 years in their future (i.e. today).
That seems true. But I think my comment is a true response to
I find this argument compelling and have thought the same thing before.
Keen to see what comes up in response to this.
This is similar to this post from just a couple of weeks ago—you may also be interested in the comments on that post.