The haste consideration

Interesting argument. I think it’s a strong consideration against investing money for the future, but less so against eg studying. There are a few problems with using it to argue against other forms of deferred payoff:

a) Improving your social standing improves the proportion of rich/intelligent people whose ear you’ll be able to bend. For eg when Warren Buffett said ‘tax us more’ in the NYT it was headline news. When the average low-wage socialist says ‘tax us more’, they’re one of a very large crowd who go largely ignored.

b) The replacement effect and the need to eat - if you’re talking about spending your normal working hours on persuasion then if you’re not getting paid for doing so, you’ll pretty quickly starve - or you’ll need to work other hours, so you’ll simply have inverted your daily pattern. But if your job is to be paid to persuade people of effective altruism, then chances are someone else would be doing it if you weren’t. Granted, when EAs are displacing each other the replacement effect is much diminished since you’ll only be displacing them into some other form of effective altruism, so if you happen to be better than them at persuasion it’s a strong argument that you should take the job. But by definition most EAs will not be among the most persuasive of EAs, so in fact most EAs should prefer that this job went to someone else.

c) So assuming you’re not one of the most persuasive of EAs, you’ll need to find other work. During this work-time you could be trying to persuade people, but if it’s a low-pay career then they’ll be people who have less capacity to effect change, and perhaps less rationality about how to do so, assuming that there’s a correlation between salary and intelligence. In any case, you could also try to persuade people to become EAs while you were studying.

d) This is still susceptible to recursive reasoning - rather than making a subsistence income and trying to persuade people yourself (especially since we’re assuming here that persuasion isn’t your specialty), you might do better to maximise your income now and pay for better influencers to do the job for you. Now you need to look at how much studying would improve your income (or rather your disposable income) and how effective influencing actually is - if it changed it a lot, it might well make it worth

e) all this seems to assume exponential growth of the EA sentiment given enough persuaders, but that seems ultra-optimistic. More likely we’ll quickly hit diminishing returns. The current members of 80K and GWWC are approximately the most enthusiastic and (given their Oxbridge/Ivy League locations) have some of the brightest futures of anyone in England and the US. Granted there’s still quite a lot of room to reach people, but basically the members we attract over the next couple of years will be the lowest hanging fruit. After that we should expect to have to put much more resources into gaining the same proportion of effective altruists, and the value of doing so relative to putting the money into AMF or whatever will diminish. Once it equals it, your argument will presumably no longer apply.

A complicated issue, anyway…

Glad you put this out here for people to think about. A quick addendum to this:

The haste consideration: resources for improving the world are vastly more valuable if you have those resources sooner.

This only applies if you actually use the resources in the most valuable ways available. The argument doesn’t show that if I spend $10K on buying insecticide treated nets this year, that’s much better than doing the same thing in two years. Without this caveat, there’s the whole muddle economists are in over opportunity cost arguments in favor of discounting.

“If you spend the next 2 years doing something at least as good as influencing people to become effective altruists, then these 2 years will plausibly be more valuable than all of the rest of your life.”

But you’ll be just as able to influence people in years 3 and 4. I don’t see how influencing now or later is fundamentally different from other “donate now or donate later”-type debates. Your reasoning doesn’t seem to hold unless it’s “influence now or never influence”.

Matt, that paragraph is clear, but it’s an explanation for why world (2) is improved at least as much as world (1). As you say, that follows from the definitions of worlds (1) and (2). I see nothing about why persuading one person to be an effective altruist in years 1 and 2 is more valuable than persuading several people during years 3-10.

I agree that it depends on how you fill in the details, and that with one way of filling in those details, (A) dominates. So it is indeed quite plausible that your next two years will be more valuable than everything else you do afterwards.

But there are other ways of filling in the details, and so it’s also plausible that your next two years, while very high-impact (because influencing can be very high-impact), are nevertheless not as impactful as the combined few years afterwards when you can influence more people.

Your argument seems to be (I could be misinterpreting), “Well, that first person you influenced is at least as good as you, so they’ll be making an even larger impact, and that’s all thanks to the influencing work you did in years 1 and 2.” But that can apply just as much to the people you influence later as well, who’ll themselves be doing great work (by assumption, at least as much as you), and that’ll be thanks to the work you did in years 3 and 4, or 5 and 6.

So yes: plausibly the next two years are really important, but plausibly so are later years. So I don’t see how it’s true that “resources for improving the world are vastly more valuable if you have those resources sooner.”

Thanks, Matt. This is a thought-provoking post and an important topic.

First, I’ll say that I think it may actually be fairly difficult to change someone else’s life course, especially to persuade someone who counterfactually would have done little to help others to go on and become more effective than you are. What’s probably more likely is that you’ll shift the course of someone who is already an altruist toward more effective charities and strategies. If you do this for enough people, maybe it adds up to being equivalent to creating a new effective altruist from scratch.

I – and probably others – have an intuitive revulsion to your argument at face value, because it feels like a pyramid scheme. These arguments about going meta can be tricky. You can do good, but maybe it would be better to convince others to do good, as you say. However, maybe it would be better to convince others to convince others to do good. (That seems to be the purpose of your writing this post.) How about convincing others to write articles to convince others to convince others to do good?

Still, even if going meta isn’t always the best thing to do, I think that empirically it can be a pretty good choice given the current state of things. There is lots of opportunity to expand the reach of ideas that we care about (effective altruism, or things like animal welfare in my case). If all but one person in the world were an effective altruist, then it would no longer pay off to do convincing. But we’re very far from that, and given how things look now, it does seem that convincing others can have huge payoff.

I’ll add one final point that applies to all of these donate-vs-invest decisions. A very strong reason to wait to donate (or, correspondingly, wait to convince others) is if you think you might become significantly wiser in the intervening years. If you find a cause that’s 1000 times better than the one you support now, then waiting and learning more is the much better strategy. Who knows – you might even decide that your current cause has negative value. (This can easily be the case when we get into uncertain waters like existential risk.) Of course, you could argue that if you inspire someone else to research these questions more effectively than you would do, then that’s even better. But at some point, the superiority of going meta has to stop…

Perhaps not surprisingly, I don’t find that answer satisfying. I’ll make one last brief comment here before chasing you up to continue this elsewhere.

“If you believe this and you actually spend the first two years influencing, then your early years are at least as valuable as all of your later years put together.” As I’ve said earlier, this is only true if you don’t subsequently influence anyone. Say you influence Person A in years 1 and 2, and Person B in years 3 and 4. Person B can do just as much good as Person A.

In reply to David:

Person B can’t do just as much good, because B is 2 years behind Person A.

In the next two years, person B persuades one more person. Person A, however, has persuaded someone else the two years before. Now that person, who persuaded A persuaded, persuades someone else.

So, the ‘A chain’ is always one step ahead of the B chain. So, the first 2 years are always more valuable than the next 2 years.

In reply to Brian:

It’s not obvious. There certainly seem to be students who are looking for a life project. They can suddenly flip from a non-altruistic course to an altruistic one.

Ben, Person B can be two years younger than Person A at the time of influencing, so their chain can extend an extra two years after Person B’s has finished. This sort of thing could happen if you’re targeting your influencing efforts on, say, philosophy uni students, where each new batch of potential altruists is the same age when you first meet them. More generally, it makes sense to persuade people to become career-long altruists early in their careers, so the influencing efforts of Persons A, B(, C, …) should converge on people of similar ages.

And even if you can only influence people of your own age, it is not the case that the first two years are more valuable than the rest of your life put together. The B chain is only a little shorter than the A chain, and the B and C chains combined would be larger. So I would agree in this case that there is some “haste consideration”, but it is nowhere near as strong as Matt’s original conclusion.

(Lesson: never say it’ll be your last comment in a thread….)

The point is that you persuade people to do ‘about as much good as you’, so we’ve assumed each person goes on to persuade one other person over the next 2yr, and then stops, irrespective of their age.

It’s true that if you keep influencing, the B chain is only a little shorter than the A chain. So it’s not true that the 2yr period is more valuable than the rest of your life if you keep influencing, but it is true that it’s more valuable than any later 2yr period.

To get that conclusion, Matt should have assumed you can persuade 2 people in the 2yr period. The weird thing about supposing that each person persuades one other, is that then you don’t get any compounding. If each person persuades 2 in the period, then the first chain is always a factor of 2 larger than the second. (and the 2nd is a factor of 2 larger than the 3rd etc.).

Then you can show that the first chain will always be larger than the sum all the others. So, then the first 2yr period is more valuable than all of the rest put together.

Firstly, Hello!

I think there’s a danger of getting caught up in the mathematics. I’m not sure we can assign value to any period of our lives in terms of an arbitrarily long causal chain stretching into the future. But even if we can, we shouldn’t take equal credit for every ethical action in the chain.

If we determine value in terms of counterfactuals there will be a damping factor that ensures that events near to your first 2-years of EA activity will be highly counterfactually dependent on your actions, but those further away will not be.

Assuming the chain leaks counterfactual dependence faster than can be accommodated by new generations of EAs, year 3 of your EA activity, in which you convert two further individuals, will in fact be more valuable than years 1 and 2, in which you convert only one.

Hello!

In realistic situations with a great deal of uncertainty, you’d certainly want to include some kind of damping factor like this. I’m not sure how rapid it would be. It seems to depend a lot on the situation.

If we suppose, however, that P2 would not have become an EA if it were not for P1, and P3 would not have become an EA if it were not for P2 becoming an EA, then it seems to be that P3 would not have become an EA if it were not for P1. (Apparently this pattern of reasoning, conditional transitivity, is not valid in general, although this is a matter of debate: http://analysis.oxfordjournals.org/content/70/2/286.extract) If that’s the case, then the counterfactual dependence genuinely transmits through the chain.

This is getting seriously abstract though!

Like you, I think that counterfactual dependence is transitive (although the article you mentioned is interesting!) All I need is a) for there to be degrees of dependence, and b) for ‘value’ or ‘credit’ to be assigned accordingly.

The chain you describe is fairly short, but as it becomes longer, the world in which one of the members would have become an EA regardless of your actions becomes closer and closer to the actual one (even if it is not the closest world).

You’re right, it’s getting a bit abstract! Maybe it’s better to put this way - I don’t think that the night you conceive a child borrows much value from their actions when they grow up. It certainly borrows less, if any, from the actions of your grandchildren and great-grandchildren.

In some technical sense it’s all counterfactually dependent and causally related, but that seems quite a departure from what we mean when we say ‘that was the most valuable/important time in my life’.

Matt’s argument was fairly general, but here’s one concrete model of how it could work. Suppose it takes exactly two years for you, or someone like you, to convince one other person to become an EA like yourself. After two years, you’ve created one EA. After another two years, you’ve together created two more EAs, giving a total of four. After two more years, you’re up to eight. And so on, exponentially. If you start two years later, you’ll grow exponentially too, but at any given time, you’ll always be half what you would have been.

Incidentally, this reminds me of exponential paradoxes in inflationary cosmology (http://felicifia.org/viewtopic.php?p=3950), although the problems aren’t exactly the same.

Perhaps we should model the growth of EAs not as exponential but as logistic, since the maximum number of EAs can’t be greater than the human population (at least until AI EAs come along). If we assume there’s a maximum possible number of EAs, then the value of creating new ones earlier is just the extra area under the curve you get by shifting the logistic curve left a little bit. Once the maximum is reached, you’ll just keep milking value from the existing EAs.

But this model seems to give too little credit to creating new EAs. For one thing, it’s not guaranteed that the world will eventually be filled with EAs (or that the maximum will be reached, even if the maximum is far smaller than the world population). Maybe creating EAs helps to increase the probability that the world reaches a stable max level of EAs, rather than letting the EA population fall to zero.

In general, I feel confused about the right way to model the spread our influence into the future. Once we have this, Matt’s question will become a lot easier.

(Yes, I know it’s overly simplistic to talk about EAs and non-EAs in this binary fashion and to talk about them dying out or remaining at a max level forever. Such a model would be more appropriate for something more concrete and more fragile, like membership of a particular religion.)

I chatted to Matt at the weekend, and I had misunderstood what he was claiming. He was claiming that in 2 years you persuade someone to do as much good as you do in the rest of your life. So, in those 2 years you create the value of your entire career, whatever activities it involves. So, even if you go on to recruit more people in subsequent 2 year periods, the person you persuaded in the first 2 year period ALSO recruits that many people. So, you can never overtake the value created in those first two years.

Hey Brian, we also think of movement growth in these terms - there’s a lot of work to be done!

Well I can tell you that there is a whole tonne of research on the topic of persuasion modelling within religious organisations ;)

MATT! I cannot believe you listened to that whole discussion me and Mark had regarding appropriate discount factors in discounted projections without mentioning that it was you who wrote the article that sparked the conversation! haha! We must skype.

one big point I will just raise regarding ‘persuasion modelling’ what have you: There is no amount of time that could be reeeally be equated to ‘converting’ someone. Any given individual, group or community can very quickly inoculate themselves to your efforts making all future efforts to that party fruitless.

Here’s a fun model concept I was playing with yesterday re giving efforts to Advocacy vs. to obtaining a Higher Salary

You have a time machine and have given $1,000 in your life time (in thousands) call it your net individual giving you decide to give half of your efforts to advocacy , your NIG falls to $500. but 50% of your life’s efforts yielded 5 EAs that coincidentally give exactly the same amount over the course of their remaining lives (above what they would have anyway) as you would have in your entire life! Net Impact = 500$+5$1000=$5,500 five times better!

but hang on. your $500 had to have come from a higher salary and the position that entailed that might have enabled high quality exposure to some seriously wealthy individuals. Considering the exponential distribution of wealth, twice the earnings might mean effective exposure to individuals 10x more wealthy! Lets say that Plan A guy and Plan B guy both get lucky once in their lifetime and persuade one person from their extended career network much more wealthy than them to become an EA…

Plan B: 5 equivelant people and one person 5 times as wealthy: $500 + 5 * $1000 + $5000 = $10,500

Plan A: no equivelant people but one person 50 times as wealthy: $1,000 + $50,000 = $51,000

Now Plan A has five times the net impact! now we can discount this (haste consideration it) and perhaps the ‘get lucky’ events are later in life than the equivelant events so then Plan B is five times in the lead again but then we try to take into account a ‘bill gates / warren buffet effect’ from people un-related to Plan A guy but were shocked into an Effectively Altruistic state by the surprise of discovering someone in such an immoral industry so noble of intentions! Plan A is Five times ahead.

but then on closer inspection Plan A guy has taken his ‘ends justify the means’ philosophy a bit too far, he became disillusioned and was actually causing 80% the value of his net impact in grief!…

and so on and so forth.

This reminds me of the “paradox of the indefinitely postponed splurge”: How long do you keep investing in expansion and building resources before starting to use those resources? The analogous question here is how long to dwell on creating new EAs before starting to do direct altruistic work.

I think that question will (at least partially) solve itself. Some proportion of 80K members are always going to feel that funding the causes immediately is better/feels better than investing in EAs.

I think building your movement is probably the best thing to do until its very well known, because more people multiplies the impact you have. However if you just build your movement people might criticize you of not doing anything so its probably neccesary to be taking some action always.

I agree with David’s comment: “Your argument seems to be (I could be misinterpreting), “Well, that first person you influenced is at least as good as you, so they’ll be making an even larger impact, and that’s all thanks to the influencing work you did in years 1 and 2.” But that can apply just as much to the people you influence later as well, who’ll themselves be doing great work (by assumption, at least as much as you), and that’ll be thanks to the work you did in years 3 and 4, or 5 and 6.”

I don’t quite understand the haste consideration…

+1 for Ruairi’s reply. Movements that only work on making themselves bigger can be hard to get people to rally behind. :)

That said, there’s still a question of where to draw the line for marginal use of resources – whether on “investment” or “consumption.”

I’m sure we could make a Warcraft III mod for this…

Good post Matt. I agree with others that we may be getting a bit too caught up in the exact way of modelling the situation, but it would be good to resolve that. Here are a few more points:

1) The consideration that at the start of a movement, movement building could easily be the most important thing to work on is pretty solid and uncontroversial. People often say: but your doing X alone won’t change much, you need to try to get thousands of people to start doing it. Or consider whether the founders of Google should have spent all their time coding or instead spent some of it hiring people to code, then also hiring people to do human resources (i.e. hiring to hire).

Sure it is slightly recursive, but there is nothing paradoxical about the basic structure, and on examination it is clearly not a pyramid scheme. Pyramid schemes are attempts to get benefits from people in the levels below you, who get benefits from those in the levels below them, which fails for many of the people because the population is finite. This is not what is happening here, as there is actually no private benefit passing at all, just a group of people working together.

2) The argument is about growing an organisation or movement in a lasting way. I agree with Ruairi that If one merely tried to get people involved to get people involved etc, you wouldn’t get much overall sustained growth (even if there was a promised future point at which they start doing first order work). It would be much more effective at convincing people to join if a large part of the organisation’s time was spent on first order work (maybe half?). This is true for 80,000 Hours and would have been true for other organisations such as Google or various movements.

3) I think the haste part of the argument (as opposed to the growth in general part) is sensitive to questions about how the rate of getting people to join drops off (e.g. is there an ultimate S-curve and what is the probability distribution of whether we reach that point). e.g. for simplicity, if there are only 4 people interested, then one might be able to convince them all early on, or later. It is also dependent on how effective the volunteering or giving opportunities are at different times and on different margins (i.e. if certain number of hours or dollars have been put in, how important is the next unit).