Population Ethics is a Sham

Crossposted from my blog.

For a field titled “Population Ethics”, there doesn’t seem to be a whole lot of… ethics going on. It would be more accurate to call the current discourse “Population Welfare” instead - the field is, after all, about deliberating among how good certain worlds are, given the welfare of the beings within them.

And, you know, this isn’t just a semantic name change. According to a recent EA forum post, a big problem of the field is that

There is no version of population ethics where you can just say, "Whatever feels intuitively right is what's right."

Well, I just so happen to believe in a Hedonistic theory of welfare, which I do intuitively find correct. So I kind of… just have an answer to Population Welfare, that aligns with all my intuitions. Specifically, Hedonism says that welfare only lives in instances of qualia. So, to evaluate a world’s welfare, I just do the same thing that I do to evaluate a single being’s welfare; I just imagine myself experiencing every single instance of qualia, one after another.

“But… isn’t that just the totalist position? Doesn’t that bite the bullet on the Repugnant Conclusion?”

No, it’s not, and no it doesn’t. Notice how I said “imagine experiencing the qualia” rather than “aggregate the welfare of all the qualia”. The Repugnant Conclusion asks whether a world with a few people living in bliss is more valuable than a world with lots of people living moderately happy lives. I, well, I’d rather just live a short life in bliss than a long, decent life, so you can probably guess which one I’d choose. No bullets bitten, but man, these berries taste amazing.

“Okay, but then your view is inconsistent, right?”

Look, intuitions are shaky. Intuitions are inconsistent. And intuitions are downright deceptive. But that’s kind of the point here. Theories of welfare are meant to serve their welfare agents, not the other way around. Who cares if a welfare function satisfies ‘transitivity’ or whatever - the welfare agent certainly doesn’t! Our intuitions about welfare^[1] are the closest thing (if not the only thing) we have to a ‘true’ welfare function, because we ourselves are welfare agents.

Oh, and I kind of lied. You know where the actual inconsistency comes from? It doesn’t even come from our intuitions. It comes from that one, all pervasive, assumption that everyone seems to use when talking about Population Welfare.

Please, Please Stop Representing Welfare as a Real Number!

Pause, for a moment, and just reflect on what you’re experiencing right now. You’re able to see bright, vivid colours all around you. You might be hearing faint birdsong in the background, as I am right now. Notice how your current physical position feels to you - the way you’re sitting down (or standing up), whether you feel slightly cool or warm, whether your muscles feel sore, or relaxed. Notice your emotional state, whether you feel content, whether your life feels purposeful right now, whether you’re envying someone else, experiencing vicarious embarrassment of something you did yesterday, or dreading an upcoming deadline.

Doesn’t your current experience feel unique? Even the mere act of reading a blog post online can be experienced in infinitely many ways.

And, even with this, we’re still only human. Think about the range of experience of nonhumans who can see infrared and ultraviolet, or can sense electric and magnetic fields, or whose sense of hearing and taste are much, much more amplified than our own. There are infinitely many experiences in the world we could only dream of, and infinitely more we literally can’t fathom.

Then the Population ‘Ethicist’ bursts in, chucks all that out the window, and says “Here’s a single number to represent all those complex things you’re feeling right now. Take it or leave it.”

And really, I want to emphasize just how much of an oversimplification this is. Think about, for example, the joy and purpose felt after completing a marathon, as opposed to the happiness experienced playing a video game in comfort. They both feel like good experiences, sure, but they seem fundamentally different from one another. The actual mechanism of welfare being experienced is different - emotional fulfillment, or more immediate physical comfort and pleasure. Doesn’t it seem like we’re ignoring everything that makes welfare special when we assert that both experiences are “10 units of welfare” or whatever?

And by the way, Population Welfarists don’t just assign scalar welfare quantities to instances of experience. No, they go one step further, and assign scalar welfare quantities to individual persons. They compress all the things that make personhood special^[2] - real relationships, the sense of continuity, and the notion of being part of something bigger than yourself - further into the scalar spectrum, as if there was more space left even after compressing the qualia.

So, yes. If we want to make the field of Population Welfare actually reflect the interests of welfare agents, we should worry about what those welfare agents actually have to say, rather than imposing our own model onto them!

What Do I Propose, Then?

The whole field is, as you could probably guess, extremely theoretical. We do need some framework of welfare to actually, you know, do things in Population Welfare. But, wait, yeah, the field is purely theoretical, right? We actually have the time and resources to flesh out this framework, rather than using the one mathematical structure we’re all familiar with!

And as it turns out, I happen to be someone who thinks that welfare actually does live in a mathematical structure. I just think that this structure is rich and infinitely more intricate than real numbers could ever hope to capture.

Now, as for this supposed structure that captures everything fully, I agree it’s most likely impossible to actually find and consider, even in theory; Integrated Information Theory, for example, is notoriously extremely computationally expensive (not to mention it doesn’t even provide a function measuring the welfare of the system)! But that doesn’t mean we can’t find a middle ground which is relatively easy to deliberate about, but also captures most of the core features of welfare.

There are many, many ways this could be done, and I propose that it be an important question in Population Welfare to explore how different mathematical models of welfare capture intuitively attributes of said welfare. For example, lexical thresholds can be captured by open balls in Banach spaces, or similarity between two welfare states can be captured by inner products in Hilbert spaces.

Let’s see what something like this might look like in practice. I proposed a theory of welfare in this post^[3], where welfare is represented as a tuple (M, S) of an n × n matrix and a set of vectors (in the post I set S = ℝⁿ). My idea was that S represented all the things a welfare agent wanted for themselves, and M was the actual experiential state of the agent. To evaluate how much M aligns with a vector v in S, you evaluate vᵀMv. The collection of all of these numbers is a summary of the valence of one’s experience.

Okay, cool. Notice that this framing of welfare actually includes more fine-grained details of what the welfare agent is actually experiencing. And it does stuff like actually differentiating between nonconscious entities and ‘valence zero’ entities, and having lexical thresholds (in the form of positive/negative definiteness).

Now, let’s say we have a bunch of welfare agents in some worlds, each with their own (M, S). How can we decide which world is better?

…

…

It’s… hard, isn’t it?

And that’s a good thing!

Having to actually compare worlds with nuance, exploring how some worlds are better in some specific aspects, and being unsure of a final verdict, is the sign of a good theory, rather than blindly assuming there exists some objective total order on all worlds. Some worlds should seem like they’re fundamentally incomparable with others; not better, not worse, just different.

What would a thought experiment using this theory of welfare look like? Let’s put a twist on the Repugnant Conclusion; Suppose I had two worlds, A and B.

• World A consists of 100 people. Each person’s M matrix has mostly high eigenvalues, but one slightly negative eigenvalue (e.g a 3 × 3 matrix having eigenvalues 100, 10, and -1). So most vᵀMv evaluations will be positive and high in valence, but few of them will be slightly negative. You could imagine people in this world experiencing intense joy, but with the capacity for mild suffering.
• World B consists of 1000 people. Each person’s M matrix has small but positive eigenvalues (e.g 3, 2, and 1). So all vᵀMv evaluations will be slightly positive, and you could imagine people in this world experiencing only mild satisfaction, but not anything actively bad.

Which one is better? Well, let’s just calculate the total welfare of both worlds, and―

No! Were you even listening? The whole point is that there isn’t just a simple aggregation function!^[4] We actually have to compare the worlds qualitatively, and delicately. Questions arise, like “is it worth it to tradeoff bliss for a guaranteed abolition of suffering”, or “how much does the valence of an experience really matter”, or “is it more ethical for a world’s suffering to be rare, or evenly shared”.

And yes, this is the place where the ethics actually lives.

I think the field should progress by discussing these questions, rather than merely discussing “which world is better”. Then, it will truly deserve its title of Population Ethics.

Now remember, my model still assumes a ton about welfare. It is, after all, just real numbers with a few extra steps. Imagine what sort of questions a more developed and abstract mathematical model could come up with!

Okay, But Is This Actually Useful?

What’s the point in having a framework for population ethics if it doesn’t just give us clear cut answers?

Well, for one, there aren’t clear cut answers in reality, either. As I explained earlier, some welfare states are fundamentally incomparable - they’re each good or bad in their own, subtle, way. Frameworks which capture more about welfare actually reflect this fact, rather than losing that subtlety by imposing a total ordering of worlds like real numbers do.

Secondly, does using real numbers even give us the answers? A commonly held view in ‘real number’ population ethics is to evaluate world welfare by add up all the individual welfare numbers, but this isn’t the consensus. Different frameworks can actually capture why people intuitively stray away from the totalist view.

And remember that the whole point of this is to actually implement it in practice. Mathematical models of welfare can be constructed to tailor for a specific intervention; you can try to have a particular welfare model evaluate the important consequences of said intervention. Models can also become very precise. It’s much easier to get people to imagine the exact same thought experiment when more distinction is laid out, as opposed to a wishy washy ‘1 unit of welfare’ label which can pretty much be interpreted however you want.

But, and this is important, even if you want to completely ditch the mathematical framework, that’s, fine too. Remember that the supposed contradictions arising from population ethics are a result of trying to squeeze the staggeringly complex structure of welfare into the cramped real numbers; they have nothing to do with purported views actually being contradictory. So yes, you are, indeed, free to just say “Whatever feels intuitively right is what’s right.” in Population Ethics. The mathematical frameworks are, after all, just trying to give those intuitions a more formal language to speak in.

1. ^{^}

   By this, I mean what we would experience if we were actually placed in the same position as a welfare agent, not what we believe we would experience in that position. We tend to underestimate the severity of extreme suffering, for example.

2. ^{^}

   Incidentally, I happen to think that personhood isn’t special, but my point here still stands even if we take a person to be a bunch of instances of qualia.

3. ^{^}

   Which I spent like, a minute thinking of, and for a completely different reason to population ethics (I wanted to reconcile Hedonism and Preferentialism, which I now realize is not even possible). So there’s probably heaps of flaws here, but again this is just an example of what something might look like in practice.

^{^}

You might have realised that when I discussed "experiencing each instance of qualia one after another", this was implicitly aggregating all those instances of qualia into a single welfare structure. But remember, this is an aggregation of the true mathematical structure underlying qualia. It's still a job for population ethics to figure out a scheme to aggregate these simpler models of welfare (e.g by trying to figure out what are the most important aspects of welfare that should be preserved in such a scheme).

Show all footnotes