In the forthcoming post, I will argue why wild animal welfare (WAW) MIGHT be a more pressing cause area than x-risk prevention based on total utilitarianism, even after taking into account longtermism. I will show how you can make the same calculation yourself (checking if you agree) and then outline a rationale for how you could apply this estimation.
The scenario of humans outnumbering (wild) animals is improbable, primarily due to the vast number of wildlife and the potential for their spread to space, the perspective of total utilitarianism suggests that human suffering may be deemed negligible. Additionally, because sentient beings originating from Earth can live at least 1034 biological human life-years if we achieve interstellar travel, working on anything that doesn't cause permanent effects is also negligible. Consequently, our paramount concern should be decreasing x-risk or increasing net utilities “permanently”. If one is comparatively easier than the other, then that’s the superior option.
I will outline a formula that can be used to compare x-risk prevention to other cause areas. Subsequently, I will present a variety of arguments to consider for what these numbers might be, so you can make your own estimations. I will argue why it might be possible to create “permanent” changes in net utilities by working on WAW.
My estimation showed that WAW is more than twice as effective as x-risk prevention. Due to the high uncertainty of this number, even if you happen to concur with my estimation, I don’t see it as strong enough evidence to warrant a career change from one area to the other. However, I do think that this could mean that the areas are closer in impact than many longtermist might think, and therefore if you’re a better fit for one of them you should go for that one. Also, you should donate to whichever one is the most funding-constrained.
I will also delve into the practical applications of these calculations, should you arrive at a different conclusion than me.
(Note that I’m not in any way qualified to make such an analysis. The analysis as a whole and the different parts of the analysis could certainly be made better. Many of the ideas and made-up numbers I’m presenting are my own and therefore may differ greatly from what a more knowledgeable or professional in the field would think. To save time, I will not repeat this during every part of my analysis. I mainly wanted to present the idea of how we can compare x-risk prevention to WAW and other cause areas, in the hope that other people could use the idea (more than the actual estimations) for cause prioritization. I also want this post criticized for my personal use in cause prioritization. So please leave your questions, comments and most of all criticisms of my theory.)
(Note regarding s-risk: this analysis is based on comparing working on WAW to working on x-risk. While it does consider s-risks, it does not compare working on s-risks to other causes. However, the analysis could relatively easily be extended/changed to directly compare s-risks to x-risks. To save time and for simplifications, I didn’t do this, however, if anyone is interested, I will be happy to discuss or maybe even write a follow-up post regarding this.)
While I have for a long time been related to EA, I only recently started to consider which cause area is the most effective. As a longtermist and total utilitarian (for the most part), finding the cause that increases utilities (no matter the time or type of sentient being) the most time/cost-effectively is my goal. I got an idea based on a number of things I read, that against the belief of most EAs, WAW might be a more pressing cause area than x-risk prevention. Since this question is quite important for my impact, I thought instead of speculating, I would make my own analysis (aka speculation with numbers), see if I change my mind, and let the analysis be criticized (see appendix 1: for which questions about my theory I want answered).
I suspect my theory is likely incorrect due to the following reasons:
- I don’t have any credentials that make me qualified to make such an analysis.
- This is my first post.
- Due to personal reasons, I’m biased towards working on WAW.
- This theory is based on the principles of total utilitarianism, and may therefore not overlap with your views.
- The theory is mainly based on the assumption that working on WAW now could potentially cause permanent changes to how we treat (wild) animals in the future (read why I think this might be the case in my section “Tractability of the cause itself (WAW)”).
- This is all highly speculative.
Two framework ideas for the theory
The following arguments which will be presented are based on the theory that human welfare is negligible compared to wild animal welfare. You can read “abrahamrowe” post if you want a further explanation. However, here I will just give (my own) short explanation: There are 1015 wild animals alive currently (more according to some sources but not super relevant for my theory). While humans could theoretically exceed this number if we achieve interstellar travel, this is only assuming there will be no wildlife in space.
My theory is also based on the idea that any scenario where an x-risk occurs, the utilities experienced (positive and negative, and by both humans and animals) before the x-risk are negligible. Let me explain. Nick Bostrom estimates that the accessible universe can support at least 1034 biological human life-years. If humans were to avoid an x-risk until then, it undoubtedly would be because humans had reached a stage where x-risks practically were impossible to occur (existential security). E.g. because we had an AI in place that could prevent any unintentional (or intentional) x-risk from occurring, or because multiple civilizations in space would live so independently that catastrophes would be unable to affect the whole human species. If we were to reach this level of security, it would likely occur within a fraction of this timespan. Therefore anything that doesn't impact utilities “permanently'' (until all sentient creatures will die off by uncontrollable courses e.g. after the 1034 years) or increase the likelihood that we will reach this level of security, is negligible in comparison (which I guess is also kind of similar to the general argument for longtermism).
Formula for comparing impact
So if you believe that the two previously mentioned statements are true, then we basically have two options for making the world better: decreasing the likelihood of x-risk or changing the world in such a way that net utilities will be more positive "permanently". I will now present a formula that can be used to compare these two options.
First of all, if you think it's 100 % guaranteed that the future will be net positive, then whether you should work on WAW or x-risks depends on two factors: how likely you think it is that an x-risk will occur at some point in the future and how tractable you think the two cause areas are. (Note: I will later in the post argue for what these numbers might be). Based on these two variables, we can use the following formula to calculate which cause is most effective:
(Note: I’m not a mathematician, so please let me know if there’s something wrong with any of my formulas. Also I’m sure there’s better variables which could have been picked.)
So x<1 means that WAW is more effective and x>1 means that x-risk prevention is more effective.
For example, let's say the likelihood of an x-risk happening at any point in the future is 50 %. Then this would mean that according to total utilitarianism, completely eliminating the chance that x-risk will ever occur, would be equivalent in value to doubling net utilities “permanently”.
So if you believe this to be the case, the determining factor for which area you should work on is tractability. Since this is quite unlikely for the EA movement to achieve, a more realistic way of saying it is: that decreasing the chance of x-risk by 1 percentage point will be equivalent to increasing net utilities by 1 %. However, if we think the likelihood of an x-risk ever happening is 75 %, then decreasing the chance of x-risk by 1 percentage point will be equivalent to increasing net utilities by 3 %.
While I have never personally seen it presented like this, I do see obvious reasons why most people would think that the former is easier than the latter (which would be aligned with current longtermist EA work). However, I will later present why I think this might not be the case.
If you instead think the future will have net negative utilities, then working on reducing the likelihood of x-risk would be negative. So if you think this is the case, then working on the problem that would help ensure a positive future most effectively would be your best bet. I personally don't think this will be the case, however, I do think there is a risk that it will be. And if you think like me, this is still an argument for working on WAW instead of x-risks.
To make our "simple formula" more precise, we can include the possibility of a net negative future into our formula, creating our main formula:
(If you do not understand unegative I will try to explain it with an example: if we quantify the net positive utilities of the world, in case the world will be net positive, with the number 1, and then choose that unegative is equal to 2. Then this will mean that in case we end up with a world that is net negative, it will on average be twice as bad as if it were to become good, or the net utilities will be equal to -2)
The exact numbers are of course unknown, so you can enter your own estimations into the formula. E.g. this is how they will look for me:
So based on my estimations, working on x-risk instead of WAW is 0,382 times more effective. Or dividing the number “1” by 0,382, we get the number of times more effective WAW is to x-risk: 2,615.
I will, for the rest of this post, argue why I chose these numbers. I’m not trying to argue that these numbers are correct, I just want to present some arguments that pull these numbers in different directions, and then I think it’s up to you to decide how the numbers should look, based on the strength of each argument.
What is the probability of an x-risk happening at any point in the future
I don’t think we have any current useful estimations for the likelihood of an x-risk happening in the long-term future (see Appendix 2: for my reasoning). I think for many of you, your own estimations might be your best bet. I will now try to give my own estimation, using mainly made-up numbers: My personal favorite estimation (mainly just because that’s the one I’m most familiar with) is the one by Toby Ord ("1 in 6" during the next century). If we assume that it will take us one millennium to reach existential security, and the risk will decrease by 20 % each century, we can calculate the probability of an x-risk ever occurring like this:
So based on my calculations, the probability of an x-risk ever occurring is ≈ 55 %. (Note: I will be happy to hear some of your guys' estimations. And if you want to play around with any of my estimations, here’s a link).
Net negative future
Many EAs think that the future will be good. However, the idea that the future might be bad is far from new. For example, Brian Tomasik wrote in 2013 about how we might be too optimistic about the future. He also believes there’s a 70 % probability of space colonization causing an increase in total suffering (ignoring happiness). And as mentioned before, the risk for a net negative future might be a reason against working on reducing x-risk (or more specifically human extinction), because this will simply ensure that the negative future will occur. Many of the reasons are related to s-risk. Once again, we only care about scenarios where the suffering is “permanent”, any others are negligible.
I will now give my estimation (again using made-up numbers) for how likely a number of different scenarios are and how bad they would be in comparison to a net positive future:
(“others” especially refer to once we are unable to imagine)
(See Appendix 3: for some thoughts on how to do the calculation)
These numbers reflect my own beliefs based on my limited knowledge, and will most likely differ greatly from yours (see Appendix 4: for why I chose these numbers).
Many longtermist may put a lot higher weight on “artificial sentience”, which could easily make the case for working directly on this cause area instead. Also, I think for a number of you, your calculations will be so that when you multiply “p(negative)” by “u(negative)” it will equal more than 1, meaning that you believe that the future will on average be net negative. Therefore the mean expected outcome of working on x-risk reduction would be negative.
I think we can divide tractability into three parts: “tractability of the cause itself”, “tractability based on personal fit” and “tractability as the means of neglectedness”. All three, but especially the last one, are dependent on time. And personal fit obviously depends on the person. This is just to say that the tractability may vary greatly over time and from person to person. I will first give arguments to consider for each, and then give my own estimation.
Note that for calculating tractability I will specifically use AI-alignment as my example. However, since this is properly the most tractable x-risk-related cause compared to its scale, this will make a stronger case for working on x-risk than other x-risk-related causes e.g. preventing catastrophic pandemics. I will take into account that AI-related x-risk only makes up part of x-risk during my calculation of “tractability of the cause itself”.
Tractability of the cause itself
To clarify, we’re in itself, not interested in the tractability of the cause area in general. While this is important for our evaluation, what we’re truly interested in is the tractability of the cause area achieving one of the following: reducing the probability of an x-risk happening at any point in the future by 1 percentage point or increasing the net utilities of the future by 1 % “permanently”. I will talk about each separately.
Tractability of the cause itself (x-risk reduction)
There are several ways you can work on reducing x-risk, in the area of AI alignment you can do e.g. technical research on how to create safe AI systems, strategy research into the particular risks AI might pose, and policy research into ways in which companies and governments could mitigate these risks. And in different areas, there are similarly many options. Here’s a quote from 80.000 hours, on working on AI alignment: “So I think it’s moderately tractable, though I’m highly uncertain — again, assessments of the tractability of making AI safe vary enormously.”
As explained earlier, the higher the likelihood of an x-risk, the more important the cause area is, according to our equation. However, this might make the cause area less tractable. Because if the likelihood is high, it might not be enough for us to prevent an x-risk once. Preventing an x-risk once might only save us temporarily. Or in other words, decreasing the probability of an x-risk ever occurring by 1 percentage point, will require more resources. Although this might not have such a big impact on the tractability, since if the cause of the x-risk in these scenarios are similar, e.g. both are related to AI, then the work used to prevent the first one might help prevent the next one.
Tractability of the cause itself (WAW)
(Note that most of my ideas in this section are inspired by the post by saulius “Wild animal welfare in the far future”.  However, not stating our values/beliefs to be homogeneous.)
And now, finally, for the elephant in the room - is it possible to increase net utilities permanently? Truthfully… I have no idea, but I think, based on my own educated guess, that the likelihood and scale are enough to seriously consider the idea. I will now present a possible scenario where this could happen:
Based on WAW research on earth, we realize that there's no possible way (with the use of any interventions) to create environments where wild animals will have net positive lives. Therefore we will never bring animals to space (e.g. do to an international agreement). Or similarly, we learn that certain species or even entire classes of animals are practically unable to experience net positive lives. Therefore we insure to prevent these species from colonizing other planets.
I think the probability of spreading wildlife to space is pretty high. It’s quite likely that at least some people will think spreading wildlife is the morally right thing to do. This could e.g. be done by directed panspermia (deliberate transport of microorganisms into space to be used as introduced species on other astronomical objects), terraforming or other ways. Especially because nature, by the majority of people, is seen as positive. There’s also the chance of accidental spread. For example, spacecraft sent from Earth for scientific or colonization purposes might be contaminated with microorganisms. These microorganisms might evolve into sentient beings. There is an international law to prevent this, though it is not well enforced.
Working on WAW could help us to achieve one or both of the following:
- Decrease the amount of animals in space
- Increase the welfare of the animals in space (e.g. by using interventions or intentionally making artificial ecosystems that ensure better welfare).
While most people might be against the spread of wildlife to space, the choice might not be decided by the collected human species. It could be decided by an individual in power or if certain technologies could become widespread enough, it could become possible for many individuals to do by themself.
You might think that this would only be a temporary problem since these scenarios are reversible. However, technologies able to eradicate entire biospheres likely would be limited to large organizations or nonexistent for security reasons. Meaning there has to be widespread agreement to cause this reversal. But since it’s way easier to convince someone to avoid creating unnatural life, which might be unenjoyable, than it is to convince someone to destroy life that’s already existing, it seems that preventing the spread of wildlife could cause long-term changes.
Here is a list of ways we could work on this issue (directly copied from the post by saulius):
“To reduce the probability of humans spreading of wildlife in a way that causes a lot of suffering, we could:
- Directly argue about caring about WAW if humans ever spread wildlife beyond Earth
- Lobby to expand the application of an existing international law that tries to protect other planets from being contaminated with Earth life by spacecrafts to planets outside of our solar system.
- Continue building EA and WAW communities to ensure that there will be people in the future who care about WAW.
- Spread the general concern for WAW (e.g., through WAW documentaries, outreach to academia).”
You might argue that spreading WAW values is a quite indirect and potentially inefficient way of preventing a future possible problem. Which I do somewhat agree with. However, the same argument can be used against working on AI alignment, since current machine learning doesn't bring us closer to AGI, meaning that the importance of current research is questionable. Even though I still think a lot of it will be useful.
In terms of the tractability of the cause area itself in general, I think it is less tractable than many other EA causes, nevertheless I think it’s more tractable than a lot of EAs seem to think. Several interventions seem doable (which in itself might have limited long-term effect, but could help the movement grow), for instance: wild animal vaccination, helping wild animals affected by weather events, helping wild animals in urban environments, animal contraception, and developing wild animal welfare assessment methods. There are tools, such as the reference system and adaptive management, which make the uncertainty of these interventions not intractable.
Neglectedness (as a part of tractability)
Most people reading this have properly heard numerous times that neglectedness can be used to assess how effective a cause area is. But you may not have realized that this is simply an indicator for how tractable the area is.
Both causes are obviously, at least in my opinion, extremely neglected compared to their scale. For AI alignment the numbers are according to 80.000 hours as follows: Around $50 million was spent in 2020 and around an estimated 400 people (90% CI = 200-1.000) working directly on the problem.
I have been unable to find estimations for the amount of money or FTE (“full-time equivalent“) that is allocated to WAW. I will now try to make a very rough estimation for the latter. According to ACE, they know fewer than five charities that focus on improving wild animal welfare. And they currently only recommend one of them. This organisation is Wild Animal Initiative and they have 19 employees.  However some organizations work partly on WAW. I will try to take this into account.
I do expect the numbers to increase proportionally more for AI-alignment because it’s about a developing technology, so the area will properly create more news. I also think WAW is less likely to be solved by non-EAs because everyone cares about creating a safe future (even though most care less than they should), but very few care about natural suffering. And I personally think it’s easier to convince someone of AI-alignment than WAW. However, I do also think that some people working on the issue of factory farming, could be convinced of switching to WAW, especially when we come closer to mass producing cultured meat. However, for simplification, I have not considered the future growth of the two movements in my calculation.
Calculating “t” (tractability)
I will calculate t by multiplying the three areas: “tractability of the cause itself * “tractability based on personal fit” * “tractability as the means of neglectedness”. So first we need to decide the numbers.
See my estimation (and explanation for chosen variables) for “tractability of the cause itself” or “x(cause)” below:
Explanation for chosen variables:
- “p(risk)”: According to Tobi Ord's estimates, the existential risk associated with unaligned AGI accounts for 60 % of the total risk next century. And since AGI “only” accounts for a part of x-risks, this reduces the scale of the cause area.
- “t(risk)”: I do think both of them are less tractable than many other cause areas. Based on having more public and professional support and less resistance, I do think x-risk prevention is a more tractable area in terms of the “cause itself”. Therefore I put this value to 2.
- “p(WAW)”: Based on my own (conservative) guess. Because based on my framework, if animals never leave to space, working on WAW will only cause temporary change, meaning that it would be negligible according to my framework. Therefore the effect is only counted if this is the case.
- “b”: Based on my own guess.
For our example, I will simply give the number 1 for personal fit, meaning that the personal fit is equal between the course areas. Remember that based on the way I made the formula, “t” is the number of times more tractable x-risk is than WAW. So if you set this value to 2, then this means your personal fit is twice as good for x-risk. And if you set the value to 0,5, then this means your personal fit is twice as good for WAW. When you set this number for yourself, remember how important personal fit is. Generally, the top 10% of the most prolific elite can be credited with around 50% of all contributions.  So you might have to make this number differ further from 1 than you might think.
According to 80000hours, “If you care about an output K times more than society at large, then (all else equal) you should expect investing in gaining that output to be K times more effective than making other investments.” Meaning that if all else equal, one area being twice as neglected, would make it twice as effective to work on. For our example, we can divide the number of WAW workers by the number of AI-alignment workers:
So: t = 3,2*1*0,235 ≈ 0,75. Or in other words: WAW is about 1,34 times more tractable than x-risk.
Possible conclusions from the calculation (discussion)
Based on my estimations, working on WAW is more than twice as effective. However, due to the uncertainty of these numbers, I feel unable to even say with 51% certainty that WAW is more effective than x-risk prevention. Which currently is what I conclude from this data. Based on my beliefs on the future and the tractability of the causes, I can’t say with any level of certainty which cause is the most effective. Therefore if you agree with this statement, aiming for the area where you have the best personal fit is the most important.
Also due to the high uncertainty of the future, it might be so that the most effective cause area will be unaligned with our current estimations and or even by a “flow-through effect” (indirect effects of a charity's work). This is not to say we shouldn't try to make the estimations, but that in case of uncertainty, working on where you have the best personal fit, should properly be the deciding factor.
If you think there is a high risk for a net negative future or a non-negligible risk for an extremely negative future, then working on x-risk prevention might be net negative. Due to the uncertainty again, I don't think trying to increase the likelihood of human extinction would be a good idea, but trying to ensure a net positive future would be a better option than working on preventing x-risk. The uncertainty I have on whether or not the future will be net good or bad is definitely a motivation for me to prioritize WAW over x-risks. However, if you think artificial sentience is a likely s-risk, I think this could easily be the most effective cause area.
I think if other EAs do their own estimations, a lot of them will end up with x>1 (x-risk being more effective than WAW). I think most likely this will be due to them thinking WAW is less tractable than me. After all, it’s not often people talk about making “permanent” changes to utilities. I believe the spread of animals (or microorganisms that later might develop and become sentient) to space is very likely. And if it happens, it’s likely difficult to undo. I also think it’s the most likely s-risk and therefore WAW is potentially the most effective cause area so far. However, if you disagree with any of these statements, your value “t” will properly be higher (x-risk is more tractable in comparison to WAW) than mine. And so, it’s likely that another cause area is the most effective. I think, based on the cause areas I’m currently aware of, that x-risk, s-risk (especially related to artificial sentience), meta and WAW are at the top of cause areas if you're a total utilitarian and a longtermist. But I certainly could be wrong about this.
My recommendations on how to use this information
If you agree with me that we can’t say for certain which of the two cause areas are most effective:
- Work in the area where you have the best personal fit.
- If you’re doing earning to give - donate to whichever cause area is the most funding constraint currently. Historically WAW has been both funding and talent-constrained (my impression is more funding-constrained), whereas x-risks have been mainly talent-constrained, however, this seems to have changed recently.
If you think x-risk or WAW is a more effective cause area than the other:
- If you haven't started your career path or it can be changed with minimal impact, then change based on your findings.
- In general, I don’t recommend using this data to make career changes unless you're highly certain that working on one of the cause areas is more effective than the other (e.g. >80 % certainty).
- Remember that these numbers change over time, so due to this uncertainty, personal fit may still be the most important.
If you think the mean future is net negative:
- Work on WAW if you think the spread of wildlife is likely and AI sentience is unlikely.
- Work on AI sentience if you think it’s somewhat or highly likely.
- Work on s-risk prevention in general if you think multiple s-risk are of approximately equal importance when taking into account probability and significance.
Appendix 1: Things that I want answered about my theory
In general, I will be happy to hear any thoughts on my theory. Here is a list of ideas:
- Do you agree with my theory?
- Are any of my estimations (severely) improbable?
- Are there any technical or logical mistakes in my theory (e.g. incorrect formula or illogical arguments)?
- Are there any poorly made formulations that make my theory hard to understand?
- How does your calculation compare to mine?
- (Since I would prefer working on WAW, do you think it’s solely my personal bias that tries to rationalize an illogical, selfish decision?)
Appendix 2: Why I don’t consider any estimation for x-risk in the longterm future as strong evidens
There have been several estimations and surveys on the likelihood of x-risks in the near term future, e.g. Toby Ord’s estimation of "1 in 6" in the next century, and the Global Catastrophic Risks Survey (Sandberg and Bostrom 2008) found a median estimation of 19% probability among mainly academic experts on various global catastrophic risks, that the human species will go extinct before the end of this century.
Although this might be useful information in some instances, it’s not that helpful in our calculation. What we’re interested in is the probability of the “long longterm future”. Here are the only two estimations I could find of the next 5 centuries or longer:
- Humanity has a 95% probability of being extinct in 7,800,000 years, (J. Richard Gott, III 1993)
- In 1996, John A. Leslie estimated a 30% risk over the next five centuries (equivalent to around 9% per century, on average) 
Some quick thoughts on the two estimations (note: I’m certainly not qualified in commenting on the estimations, and I have not used extensive time doing research):
- Both of them are more than 20 years old, so they might not reflect more recent information.
- Both, but especially the first one is largely based on the controversial Doomsday argument, while I do like the argument, I think it’s too speculative and holds some logical flaws e.g. observer selection effect.
- The first one is partly based on the idea that we’re unlikely to colonize the Galaxy. While I do think there is a significant chance that humans will never reach interstellar travel, I think presumably it will eventually happen. Read this post to learn why I think so. 
- (I, with my minimal time using to write this post, was unable to find out how Leslie came to the exact number of 30)
Appendix 3: Thoughts on how to do the calculation on “net negative future”
- Remember these numbers are made to be used for the formula. So it’s not about the probability that this future will happen in total, it’s instead about the probability, that if there will be no x-risk, there will be x probability for any of these scenarios to happen.
- Note that if you're using this calculation to compare working on an area that could be both an x-risk and a s-risk (e.g. AI misalignment), this should be reflected in your calculations. This could for example be done by instead of adding this specific number together with the others, you would instead minus it (because you’re both working on decreasing the likelihood of this specific negative future and increasing the chance that there will be a future at all). (Note that I did not do this in my calculation for the sake of simplification and because it would have minimal impact on the result).
- p(total risk) gets added together instead of multiplying the probability of it not happening (which was what we did in the calculation of x-risks) because these risks can coincide.
Appendix 4: Thoughts on why I chose these numbers for “net negative future”
- I may have put “b” to be lower than what you may expect, because of my high thoughts of how positive the future could be. I also think that if done right, there is a potential chance that we could have “the opposite of s-risks” in the future, e.g. positive spread of wildlife to space or artificial sentience (however I don’t think any EAs should attempt to create this (yet), due to the high risk of doing the opposite).
- I put the “p” for “AI misalignment” as being pretty low, because most scenarios of AI misalignment, will according to my previous statements, be negligible (talking about s-risk, not x-risk). It would in this case only be relevant if AI keeps us and/or animals alive “permanently” to have net negative lifes (which most likely would require traveling outside of the solar system). I also put “b” pretty low, because I think most likely (but not guaranteed) the impact will be minimal to animals.
- The probability of “Spread of wildlife to space” was set pretty high, because I think it’s quite likely that some people will think that using our power to spread wildlife is the morally right thing to do. Especially because nature, by the majority of people, is seen as positive. There’s also the chance of accidental spread. For example, spacecraft sent from Earth for scientific or colonization purposes might be contaminated with microorganisms.