Marcus Davis: Rethink Priorities — empirical research on neglected causes

by EA Global Transcripts10 min read9th Jul 20202 comments



Rethink Priorities is a research organization dedicated to conducting research on neglected causes. Over the past year, they’ve developed a track record of high-quality research in the areas of farmed animal welfare, wild animal welfare, and risks from nuclear weapons. Co-founder Marcus Davis discusses the organization’s approach to producing valuable research, with specific reference to their work on animal welfare.

We’ve lightly edited Marcus’s talk for clarity. You can also watch it on YouTube and read it on

The Talk

Hello, everyone. I’m here to talk to you today about Rethink Priorities, which is an organization I started with my colleague Peter Hurford in January 2018.


We're primarily a research organization dedicated to investigating neglected, tractable causes. What does that mean? We try to work on things that are high-value and also actually doable. We explicitly focus on those elements because I think they can influence the way funders and other animal organizations behave.


But that's still a very broad scope, so we're looking primarily at interventional research. Within effective altruism, there are different camps: people working on global health, people working on animal welfare, people working on existential risk. Instead of trying to compare these causes to one another, we are drilling down into the [sub-topics] within these areas. We’re gathering information on [sub-topics that people are aware of] but perhaps haven’t fully analyzed, as well as new [sub-topics] within these fields.

I'm going to talk to you today about two examples of our work that we've done in 2019.


The first is fish-stocking within the area of invertebrate sentience. (I will explain both of those terms for people who aren’t familiar with them.)


We investigated fish-stocking, and Saulius Šimčikas was the primary researcher. Fish-stocking is the process by which fish are raised in hatcheries to later be released into the wild for various reasons. Our work entailed rather straightforward information collection; there's a lot of information about fish-stocking in various countries and domains. But we had to gather that information into one place and analyze it to get a sense of what is happening and what can — and should — be done. Basically, it was part of a larger effort to count animals in various conditions, which can give you a sense of how you should prioritize.


There are a lot of fish raised this way — something like 30 to 100 billion in a given year. For context, maybe 150 billion farmed fish are raised every year, and 60 or 70 billion farmed land animals. The fact that this hasn't been well-studied is a considerable problem. It's an issue that we think could potentially be worth prioritizing. (There are a lot of caveats as well, but I don't have time to get into those.)

Importantly, when we were looking into this, we didn't find any existing organizations that are trying to reduce the possible harms from fish-stocking. And given that these fish are raised for economic reasons (like other farmed animals), there's no particular reason to believe that their welfare considerations are being addressed.


Why would you raise fish in a hatchery and then release them into the wild? It may sound very strange. But there are several reasons. The primary one is commercial fishing. Commercial fishermen want higher catches, perhaps because local fishing has already diluted the pool of existing fish. So they've come up with a workaround: they just raise fish themselves and then release them [so that they can be caught again after they’ve grown bigger]. This is perhaps on the order of tens of billions of fish per year, and the plurality of fish are likely raised this way.

Second, there are fish raised for recreational fishing in lakes, streams, and ponds that are perhaps over-fished or contain no fish at all. To increase the pleasure of fishermen, people raise fish to be released into these situations. I’ll discuss the conditions a bit more and show you what the process of releasing a fish this way looks like.

Finally, I should note that there are ecological reasons for raising fish in this way — for example, if there's an endangered species or an environment that’s being rehabilitated. These are perhaps the fish whose welfare we have the least reason to be concerned about. If the goal is to increase their number, as opposed to increasing the temporary stock or supply for fishermen, there are clearly different incentives.


What are the conditions of these fish? Fish can be stocked at a variety of ages. Pictured here are fish when they are quite young. The mean time spent in these hatcheries is uncertain, because a lot of different species are raised in these conditions. But it’s something like a few days to a few months, with a lot of uncertainty around the edges. Fish stock in recreational fisheries are usually a bit older, because if you're trying to directly improve a fisherman’s catch, there’s not much time for the fish to grow up. Fish are dropped into a lake and then are caught at a reasonable size — again, without any real consideration of the welfare of the fish.


If you want to see what these hatcheries look like, this is a still photo from a video on YouTube, which is not very common to find. Consider land animals that are farmed: There's not a lot of information publicly available showing the inside of those farms.


If you squint, you may see a sign at the top of this image. It suggests that each of these tanks may contain 30,000 fingerlings. Even in a facility not much larger than the size of this room, we're talking about the welfare of hundreds of thousands of individuals.

It’s quite a challenge to contain this problem if there are hundreds of thousands of fish, even in small hatcheries, raised this way. And again, it's not clear how much incentive hatcheries have to do anything about the fishes’ welfare. It's also the case that with commercial fish stocking, fish that are raised and then released into the wild might have a hard time learning to forage and to feed themselves, especially if they’re older. They've been fed in a tank their whole life. Imagine releasing an animal under those conditions. They may immediately starve to death.


This is an image that looks very terrifying, but it's actually quite normal. I took the image directly from an agency of the Canadian government's Twitter account. It shows a fish being released via helicopter into a lake. It doesn't look great. The survival rate is actually surprisingly high for something like this, but the process of being stocked, forced into a helicopter, and then released is probably not terrific.


So what, if anything, can we do about this? It's a little uncertain given that there are not a lot of people working in this area. That said, some obvious areas to go into would be learning more about what the conditions are like. You can do this through a variety of means: scientific research [on fish biology], or research on the individual countries and companies that are doing this. We can require better conditions for fish as we make welfare strides for farmed land animals. In general, we can gather more data on the welfare of a species and how each species needs to be treated to thrive.

Invertebrate sentience

I'm going to pivot hard to talk about invertebrate sentience, which is another project we worked on this year. I'm going to use the word “sentience” very vaguely, meaning there’s some internal process going on [in the animal], some subjective experience.


Some of you may be wondering: Why bother doing this at all for invertebrates? I’ll get into that, how we did what we did, and what it all means. And again, I’m skipping many limitations and caveats.


There are a lot of invertebrates. I can't emphasize that enough. The table to the left is a scale of mass, but by number there are approximately 10 to the 27th power animals. And there are approximately 10 to the 27th power invertebrates. So most animals are invertebrates. Basically, any plausible moral theory assumes some animals can suffer, and I would say, there's a strong possibility you should give invertebrates some moral credence. Many invertebrates express behaviors in cognitive complexity that people typically associate with dogs, mammals, and cats — animals we have a lot of experience with and already care about.


So what do we do to better understand this topic? We were inspired by some work done by Open Philanthropy. This is a table from Luke Muelhauser's look at consciousness. He considered a series of invertebrates and tried to analyze which sentience-indicating features these animals possess.

We looked at a rather wide variety of animals: 12 invertebrate groups and a few examples of things that are probably not sentient, namely plants, prokaryotes, and protists. In addition, we looked at humans, chickens, and cows, with humans obviously being the baseline case since we’re the most certain that we are sentient. And interestingly, we looked at things that could happen in humans unconsciously as a potential counterexample.


We looked at a variety of features. This [slide shows] a high-level categorization of those features. I'm not going to get into the details of all of them. Some are rather obvious anatomical evolutionary features. These are things that are biologically present in animals inherently or they're distant from humans on an evolutionary timescale. There are drug responses — things like how an animal responds to opiates or recreational drugs.

Of course, there's a huge caveat: Can you extrapolate how an animal responds to something like recreational drugs to whether they're actually experiencing something? Similar caveats apply to simply assessing whether they respond to opiates at all. There are a number of different things around mood, like whether they seem to become depressed if you give them the wrong stimulus, or whether they seem to be anxious. Again, there are a lot of these features. I don't have time to get into them all.

Perhaps the one that most interests people is cognitive sophistication. What types of impressive feats can animals do intellectually, whether it's solving mazes or puzzles, escaping from situations, or just displaying general intelligence? There are a lot of limitations here. Many of these features are possibly relevant. We primarily try to gather evidence from existing literature reviews and prominent studies, but we still had to make some exclusions and trade-offs between gathering certain information [based on] whether it fit into a certain category.

Cognitive features are limited to things that invertebrates could plausibly do, so it wouldn't make sense for me to include, say, writing an opera. There's no reason to think any given invertebrate, or even most humans, could do that. And there are a variety of data limitations around possessing these features at all.


This is the full list of features. I'm not going to get into this at all, but [note] the tiny URL in the corner. You can come back to this at your leisure.

What did our end product, after doing this type of analysis, look like?


We gathered this information into an interactive table. You can select the features and animals you want, and review our collection of the evidence. For ease of navigation, we rated the end result of each feature on a scale of [how likely the animal is to possess it]. Then, you can click through each of these links and drop down to the scientific evidence (and sometimes the notes) that we gathered about the topic.


[The slide above shows] how it works. Again, this is very useful if you want to come back to it later.


There’s an important limitation here. I mentioned that we gathered 12 different invertebrate taxa — but not species. Among the animals we gathered, some groups are much larger than others. Spiders are the largest; there are some 40,000 species of spiders. So if only certain species of those spiders display certain features, [it would be misleading to say that those features translate to all spiders]. Similarly, let’s say you asked, “Can mammals play chess?” The answer is yes, some mammals can play chess, but this doesn't tell you anything useful about dogs.


There are even more limitations, but I'm not going to get into some of these. Basically, drug responses, like a lot of things, come in degrees. It's hard, even within an individual study, to suggest that the behavior you observe is actually indicative of a given thing. So there's reason to be cautious about any of the results we see on a number of these things, where we're extrapolating not just from animal behavior, but considering what it means for similar human-like behavior under those conditions.

I want to highlight self-control, which is a particularly interesting feature. Some of you may be aware of the “marshmallow experiment.” This is a test we give to children, and sometimes birds and other animals, in which we offer the child or animal a marshmallow, and say, “If you eat this right now, you can have one, but if you wait 10 minutes, you can have two.” This is an incredibly difficult challenge for four-year-olds. There are great videos of gray parrots doing the same task — they seem to visually employ the same strategies [as four-year-olds] — and some of these tests have been performed on invertebrates, which I'll get to.

Finally, I want to flag the feature “flexible tool use.” Classical tool use would be the ability to use any given thing that's not part of your natural body. With flexible tool use, we set a higher, [albeit] still subjective, standard. The tool isn’t something in [the animal’s] evolutionary environment; it’s something that [the animal] could perhaps adapt, demonstrating its intellectual ability.


I want to get to a couple of specific examples: honey bees and ants. Honey bees display an array of impressive features. They're considered very social creatures, but specifically, they displayed a dose-dependent response to analgesics. That would be impressive if you're thinking, “What other reason, besides actually experiencing pain, would there be for having a dose-dependent response?” In contrast, there was no evidence at all for ants. It’s not that there was evidence to the contrary for them; it’s just that there have been no studies on this topic. This is another indication of the limitations of this type of approach.

Both honey bees and ants groom themselves in response to toxins. You can imagine evolutionary reasons for why this may be the case. Nevertheless, it seems like evidence that they are at least doing something to extend their own lives.

For bumblebees — which are not honeybees, but are genetically close — there's actually evidence that they can learn by observing things. This is, again, something that is often difficult for small children to accomplish, but bumblebees can visually watch something happen and then pick up on it after the fact. Also, as I mentioned regarding self-control, they can pick a larger delayed reward over a smaller present one. Again, there's no information like this for ants, [pointing to the limitations of the information available to us].


There are a ton of studies on octopuses. Perhaps one of the more interesting things is that there are no known analgesics for them — i.e., no known painkillers. On the other hand, they can regrow their own limbs. So what does it even mean to say that there are no known painkillers when you have a behavior like that? This is another example of the limitations of this type of approach.

Also, as the image showed earlier, they can take human-discarded coconut shells — something they've not interacted with on a long time scale — and use them as housing. This is a good indication of their general intelligence. And there are many more [anecdotal indicators of it]. [In labs,] they’ve escaped mazes, solved puzzles, and — in perhaps my personal favorite anecdote — deliberately thrown down their food on the floor if they don't think it's high enough quality.


So are invertebrates sentient? Maybe. It depends a lot on the invertebrate; the evidence varies a lot by species. The table on the left is an indication of the features we looked at, and which animal captured those features. This is not a degree-of-sentience scale.


Basically, you can make a case against invertebrates. I want to throw it out there. You could argue that we shouldn't care about their pains and pleasures, because any moral theory which recommends you do so is invalid. But overall, I think these types of arguments are generally too strong. We're going to release some opinion pieces on this topic later. There's not so much certainty to these arguments that you can give less than 1-5% credence to the idea that insects, octopuses, and some other invertebrates actually possess these features.

I'm already over time, so I'm going to stop there. Thank you for coming and feel free to fire questions at me.

Moderator: Thank you, Marcus. So when it comes to sentience — and particularly invertebrate sentience — people have a lot of different priors [prior studies] to consider. I'm wondering if you know what your team's priors are and how you account for that in your research methodology, if at all?

Marcus: One thing about our project is that we were trying to be as theory-neutral as possible. We weren't gathering evidence based on the theory that a given theory of consciousness is correct. One of the nice things about being able to select the features that you think are important in the interactive table [we created] is that you can [base your selections on] whatever your current priors are.

The goal of this project wasn't to reach a certain [level of] credence that invertebrates are sentient. But having looked at it, I think it was important to release our takes on it. We're going to explain in some depth why we came to the conclusions we did. There's wide divergence within the team about the correct level of credence to give to different invertebrates.

Moderator: Fabulous. That's actually all the time we have for questions. Thank you so much for your presentation and for being part of EA Global this year. Everyone give Marcus a round of applause, please.