Nick Beckstead’s finished but as-yet unpublished dissertation has much to say on this topic. Here is Beckstead’s summary of chapters 6 and 7 of his dissertation:
[My argument for the overwhelming importance of shaping the far future] asks us to be happy with having a very small probability of averting an existential catastrophe [or bringing about some other large, positive “trajectory change”], on the grounds that the expected value of doing so is extremely enormous, even though there are more conventional ways of doing good which have a high probability of producing very good, but much less impressive, outcomes. Essentially, we’re asked to choose a long shot over a high probability of something very good. In extreme cases, this can seem irrational on the grounds that it’s in the same ballpark as accepting a version of Pascal’s Wager.
In chapter 6, I make this worry more precise and consider the costs and benefits of trying to avoid the problem. When making decisions under risk, we make trade-offs between how good outcomes might be and how likely it is that we get good outcomes. There are three general kinds of ways to make these tradeoffs. On two of these approaches, we try to maximize expected value. On one of the two approaches, we hold that there are limits to how good (or bad) outcomes can be. On this view, no matter how bad an outcome is, it could always get substantially worse, and no matter how good an outcome is, it could always get substantially better. On the other approach, there are no such limits, at least in one of these directions. Either outcomes could get arbitrarily good, or they could get arbitrarily bad. On the third approach, we give up on ranking outcomes in terms of their expected value.
The main conclusion of chapter 6 is that all of these approaches have extremely unpalatable implications. On the approach where there are upper and lower limits, we have to be timid — unwilling to accept extremely small risks in order to enormously increase potential positive payoffs. Implausibly, this requires extreme risk aversion when certain extremely good outcomes are possible, and extreme risk seeking when certain extremely bad outcomes are possible, and it requires making one’s ranking of prospects dependent on how well things go in remote regions of space and time.
In the second case, we have to be reckless — preferring very low probabilities of extremely good outcomes to very high probabilities of less good, but still excellent, outcomes — or rank prospects non-transitively. I then show that, if a theory is reckless, what it would be best to do, according to that theory, depends almost entirely upon what would be best in terms of considerations involving infinite value, no matter how implausible it is that we can bring about any infinitely good or bad outcomes, provided it is not certain. In this sense, there really is something deeply Pascalian about the reckless approach.
Some might view this as a reductio of expected utility theory. However, I show that the only way to avoid being both reckless and timid is to rank outcomes in a circle, claiming that A is better than B, which is better than C,. . . , which is better than Z, which is better than A. Thus, if we want to avoid these two other problems, we have to give up not only on expected utility theory, but we also have to give up on some very basic assumptions about how we should rank alternatives. This makes it much less clear that we can simply treat these problems as a failure of expected utility theory.
What does that have to do with the rough future-shaping argument? The problem is that my formalization of the rough future-shaping argument commits us to being reckless. Why? By Period Independence [the assumption that “By and large, how well history goes as a whole is a function of how well things go during each period of history”], additional good periods of history are always good, how good it is to have additional periods does not depend on how many you’ve already had, and there is no upper limit (in principle) to how many good periods of history there could be. Therefore, there is no upper limit to how good outcomes can be. And that leaves us with recklessness, and all the attendant theoretical difficulties.
At this point, we are left with a challenging situation. On one hand, my formalization of the rough future-shaping argument seemed plausible. However, we have an argument that if its assumptions are true, then what it is best to do depends almost entirely on infinite considerations. That’s a very implausible conclusion. At the same time, the conclusion does not appear to be easy to avoid, since the alternatives are the so-called timid approach and ranking alternatives non-transitively.
In chapter 7, I discuss how important it would be to shape the far future given these three different possibilities (recklessness, timidity, and non-transitive rankings of alternatives). As we have already said, in the case of recklessness, the best decision will be the decision that is best in terms of infinite considerations. In the first part of the chapter, I highlight some difficulties for saying what would be best with respect to infinite considerations, and explain how what is best with respect to infinite considerations may depend on whether our universe is infinitely large, and whether it makes sense to say that one of two infinitely good outcomes is better than the other.
In the second part of the chapter, I examine how a timid approach to assessing the value of prospects bears on the value of shaping the far future. The answer to this question depends on many complicated issues, such as whether we want to accept something similar to Period Independence in general even if Period Independence must fail in extreme cases, whether the universe is infinitely large, whether we should include events far outside of our causal control when aggregating value across space and time, and what the upper limit for the value of outcomes is.
In the third part of the chapter, I consider the possibility of using the reckless approach in contexts where it seems plausible and using the timid approach in the contexts where it seems plausible. This approach, I argue, is more plausible in practice than the alternatives. I do not argue that this mixed strategy is ultimately correct, but instead argue that it is the best available option in light of our cognitive limitations in effectively formalizing and improving our processes for thinking about infinite ethics and long shots.
Nick Beckstead’s finished but as-yet unpublished dissertation has much to say on this topic. Here is Beckstead’s summary of chapters 6 and 7 of his dissertation: