Interaction Effect

post by Lloy2 · 2019-12-16T15:42:10.436Z · score: 7 (5 votes) · EA · GW · No comments

This is a question post.


    21 David_Moss
    11 abrahamrowe
    2 alexrjl
    1 ishi
No comments

Disclaimer: I'm no expert on any of this-- 18 years old. I have utmost respect for EA and much faith in the movement's ideas, there are just some questions weighing on my mind.

I'm not sure if 'interaction effect' is actually the correct phrase to use but I'll explain what I mean anyway.

When discussing an interaction effect in the context of EA, I'm describing how the actions/career paths with most expected impact are dependent on the existence of choices with less estimated impact. From my (fairly surface level) exploration of EA literature so far, including much of 80,000 Hours, I've gathered that this concept is overlooked by the community thus far, but I'm very open to being shown otherwise.

I'll illustrate what I'm calling the interaction effect with an example. Let's say someone goes into strategic AI research at the Future of Humanity Institute because this is proposed to be one of the most impactful career paths there is. In aiming for that career this person relied on the labour of several teachers. When the researcher is sick, they rely of the labour of doctors. They need to eat and so rely on the labour of people working in supermarkets. They sleep on a bed that only could have been bought through the labour of people working at a bed shop. You get my point. Every aspect of this AI research role is inextricable from other components of society.

My question is, if different roles of society are interdependent on each other, even when seemingly disconnected, then how can they rationally be ranked? How can a set of careers be deemed most important when they literally couldn't exist without (most of) all the other careers out there?

A response I can imagine reading is that even if most jobs are equally important by virtue of their interdependence, how skewed the ratio between how needed the work is vs how many people are actually working on it is a measure for priority. Like sure, entertainment is an important form of escapism in hard times and an agreeable source of information, but maybe there are too many artists and not enough people working in, say, biosecurity?

Anyway, hopefully if nothing else this could stimulate some discussion on a topic I've not seen addressed in EA. Cheers


answer by David_Moss · 2019-12-16T16:49:28.724Z · score: 21 (8 votes) · EA(p) · GW(p)

Let's say someone goes into strategic AI research at the Future of Humanity Institute because this is proposed to be one of the most impactful career paths there is. In aiming for that career this person relied on the labour of several teachers. When the researcher is sick, they rely of the labour of doctors...

This doesn't seem to pose so much of a problem if you are trying to rank what is most valuable on the margin. Suppose every human activity is dependent on having at least one doctor and at least one farmer producing food, such that these are completely necessary for any other job to take place. It doesn't follow that we couldn't determine which job it would be most valuable to have one additional person working in. For example, if we already have enough doctors or farmers, even if these jobs are entirely necessary, we could still say that it is more valuable for a further person to work in a different field.

I think you've basically captured this with your artist example, although it's worth noting explicitly that how important art is on average, is different from its value on the margin, i.e. we could think that art or being a doctor or whatever, is the single most valuable human activity (on average) and still think that it would be more important for a particular person to go and work in another activity.

comment by Lloy2 · 2019-12-16T16:56:05.166Z · score: 1 (1 votes) · EA(p) · GW(p)

Interesting, thank you.

answer by abrahamrowe · 2019-12-16T16:57:20.111Z · score: 11 (7 votes) · EA(p) · GW(p)

I'm not particularly EA, but I think they gist of the argument is - you should work where can you make the most marginal impact, not necessarily in a job that is the highest impact overall. So if you're choosing a career for impact, you might be one of only a few thousand people thinking about things in EA terms. If you want to have a large impact, then you ought to look at things that are large in scope and neglected, etc.

If somehow the EA community coordinated all resources, or was much much larger in size, the recommended careers would probably be different. In that case, obviously some people would need to be teachers, farmers, etc., and it would be important to encourage people capable of doing those things well to pursue those careers. But, given that there are relatively few people willing to change their careers for this sort of impact right now, the careers recommendations that are made in fields where a few people might have a larger impact.

This isn't a denial of interdependence. It's more of an implicit acknowledgement of the limits of the current size of the community.

Another factor is that many careers that EA careers are dependent upon are likely to be filled regardless. There are people who would like to be, or whose circumstances cause them to be, teachers, construction workers, farmers, truck drivers, etc. So while all those jobs probably have a (positive) impact, it's less urgent for someone who wants to have the greatest impact to pursue that as a career. While education might be important, I know that if I don't apply for a job at my local high school, another (even more) capable teacher probably will. Instead, on the margin, an EA might have a greater impact by pursuing something more neglected, or pursuing a career where they can earn money to donate to a charity that can hire more people for a neglected cause, etc.

The core idea is that because there is only a small community of people interested in having the greatest impact they can, then they should pursue careers that on the margin would be most likely to have the greatest impact. It doesn't necessarily mean that these careers are intrinsically or functionally "better" or higher ranked than others. They are prioritized by EA because few people are in EA, and fewer people are thinking about pursuing recommended careers.

comment by Lloy2 · 2019-12-16T17:47:30.862Z · score: 1 (1 votes) · EA(p) · GW(p)

Okay yeah, I think I get it now. Thanks.

answer by alexrjl · 2019-12-16T18:53:12.238Z · score: 2 (2 votes) · EA(p) · GW(p)

I think the other responses capture the most important response to your question, which is that we tend to look at the value of things on the margin. However, as you're clearly thinking intelligently about important ideas, I thought I'd point you in the direction of some further thinking.

Another, perhaps clearer case where this "thinking on the margin" happens is with charity evaluation. If, for example, there existed some very rare and fatal disease which cost only pennies to cure, it would be extremely cost effective for people to donate to an organisation providing cures, until that organisation had enough to cure everyone with the disease. After this point, the cost effectiveness of additional funding would dramatically drop. Usually this doesn't happen quite so dramatically, but it's still an important effect. It is this sort of reasoning which has prompted givewell, for example, to look at "room for additional funding", see here.

There's another way of looking at your question though, which is to re-phrase it as "how should we assign credit for good outcomes which required multiple actors?"

One approach to answering this version of the question is discussed in depth here. [EA · GW] I think you may enjoy it.

answer by ishi · 2019-12-16T18:02:13.794Z · score: 1 (3 votes) · EA(p) · GW(p)

I have heard some people wanted to have a 'high impact career' but instead they became a 'stay at home mom or dad'. They had to raise 1 or more children--who then went on to become noble prize winners. That to me is an 'interaction effect'.

No comments

Comments sorted by top scores.