What do animals feel?

0:00:01 (upbeat music)
0:00:04 Support for this show is brought to you by Nissan Kicks.
0:00:06 It’s never too late to try new things
0:00:09 and it’s never too late to reinvent yourself.
0:00:12 The all-new re-imagined Nissan Kicks
0:00:14 is the city-sized crossover vehicle
0:00:17 that’s been completely revamped for urban adventure.
0:00:20 From the design and styling to the performance,
0:00:21 all the way to features
0:00:23 like the Bose Personal Plus sound system,
0:00:26 you can get closer to everything you love about city life
0:00:29 in the all-new re-imagined Nissan Kicks.
0:00:34 Learn more at www.nisanusa.com/2025-Kicks.
0:00:36 Available feature,
0:00:39 Bose is a registered trademark of the Bose Corporation.
0:00:46 – Your own weight loss journey is personal.
0:00:47 Everyone’s diet is different,
0:00:49 everyone’s bodies are different,
0:00:50 and according to Noom,
0:00:53 there is no one-size-fits-all approach.
0:00:55 Noom wants to help you stay focused
0:00:57 on what’s important to you,
0:01:00 with their psychology and biology-based approach.
0:01:02 This program helps you understand the science
0:01:04 behind your eating choices
0:01:07 and helps you build new habits for a healthier lifestyle.
0:01:09 Stay focused on what’s important to you
0:01:13 with Noom’s psychology and biology-based approach.
0:01:16 Sign up for your free trial today at Noom.com.
0:01:24 – Can you ever really know what’s going on
0:01:26 inside the mind of another creature?
0:01:31 – In some cases, like other humans or dogs and cats,
0:01:35 we might be able to guess with a bit of confidence,
0:01:39 but what about octopuses or insects?
0:01:41 What about AI systems?
0:01:44 Will they ever be able to feel anything?
0:01:48 Despite all of our progress in science and technology,
0:01:50 we still have basically no idea
0:01:52 how to look inside the private experiences
0:01:54 of other creatures.
0:01:57 The question of what kinds of beings can feel things
0:02:00 and what those feelings are really like
0:02:01 remains one of the biggest mysteries
0:02:04 in both philosophy and science.
0:02:07 And maybe, at some point,
0:02:10 we’ll develop a big new theory of consciousness
0:02:14 that helps us really understand the inside of other minds.
0:02:18 But until then, we’re stuck making guesses
0:02:22 and judgment calls about what other creatures can feel
0:02:25 and about whether certain things can feel at all.
0:02:30 So, where do we draw the line
0:02:34 of what kinds of creatures might be sentient?
0:02:37 And how do we figure out our ethical obligations
0:02:39 to creatures that remain a mystery to us?
0:02:43 I’m O’Shawn Jarrow, sitting in for Sean Illing,
0:02:45 and this is the Gray Area.
0:02:54 My guest today is philosopher of science, Jonathan Birch.
0:02:56 He’s the principal investigator
0:02:58 on the Foundations of Animal Sentience Project
0:03:00 at the London School of Economics,
0:03:03 and author of the recently released book,
0:03:07 The Edge of Sentience, Risk and Precaution in Humans,
0:03:08 Other Animals and AI.
0:03:13 He also successfully convinced the UK government
0:03:17 to consider lobsters, octopuses, and crabs sentient
0:03:19 and therefore, deserving of legal protections,
0:03:22 which is a story that we’ll get into.
0:03:24 And it’s that work that earned him a place
0:03:26 on Vox’s Future Perfect 50 list,
0:03:30 a roundup of 50 of the most influential people
0:03:33 working to make the future a better place for everyone.
0:03:37 And in Birch’s case, for every sentient creature.
0:03:42 In this conversation, we explore everything that we do
0:03:45 and don’t know about sentience
0:03:47 and how to make decisions around it,
0:03:50 given all the uncertainty that we can’t yet escape.
0:03:54 Jonathan Birch, welcome to the Gray Area.
0:03:56 Thanks so much for coming on.
0:03:57 – Thanks for inviting me.
0:04:00 – So, one of the central ideas of your work
0:04:04 is this fuzzy idea of sentience.
0:04:06 And you focus on sentience across creatures,
0:04:08 from insects to animals,
0:04:11 to even potentially artificial intelligence.
0:04:14 And one of the challenges in that work
0:04:17 is defining sentience in the first place.
0:04:19 So, can you talk a little bit about how you’ve come
0:04:21 to define the term sentience?
0:04:25 – For me, it starts with thinking about pain
0:04:27 and thinking about questions like,
0:04:29 can an octopus feel pain?
0:04:31 Can a crab, can a shrimp?
0:04:35 And then realizing that actually pain is too narrow
0:04:40 for what really matters to us and that matters ethically.
0:04:43 Because other negative experiences matter as well,
0:04:47 like anxiety and boredom and frustration
0:04:49 that are not really forms of pain.
0:04:53 And then the positive side of mental life also matters.
0:04:57 Pleasure matters, joy, excitement.
0:04:59 And the advantage of the term sentience for me
0:05:02 is that it captures all of that.
0:05:04 It’s about the capacity to have
0:05:07 positive or negative feelings.
0:05:11 – The way that you define sentience
0:05:13 struck me as kind of basically the way
0:05:15 that I’ve thought about consciousness.
0:05:17 But in your book, you have this handy diagram
0:05:20 that shows how you see sentience and consciousness
0:05:22 as to some degree different.
0:05:24 So how do you understand the difference
0:05:27 between sentience and consciousness?
0:05:29 – The problem with the term consciousness, as I see it,
0:05:32 is that it can point to any other number of things.
0:05:34 Sometimes we are definitely using it
0:05:37 to refer to our immediate raw experience
0:05:39 of the present moment.
0:05:41 But sometimes when we’re talking about consciousness,
0:05:45 we’re thinking of things that are overlaid on top of that.
0:05:47 Herbert Feigel in the 1950s
0:05:49 talked about there being these three layers,
0:05:53 sentience, sapience and selfhood.
0:05:55 Where sapience is about the ability
0:05:59 to not just have those immediate raw experiences,
0:06:01 but to reflect on them.
0:06:03 And selfhood is something different again,
0:06:06 ’cause it’s about awareness of yourself
0:06:09 as this persistent subject of the experiences
0:06:13 that has a past and has a future.
0:06:15 And when we use the term consciousness,
0:06:18 we might be pointing to any of these three things
0:06:22 or maybe the package of those three things altogether.
0:06:25 – So sentience is maybe a bit of a simpler,
0:06:27 more primitive capacity for feeling
0:06:30 where consciousness may include these more complex layers?
0:06:31 – I think of it as the base layer.
0:06:34 Yeah, I think of it as the most elemental,
0:06:37 most basic, most evolutionarily ancient
0:06:40 part of human consciousness
0:06:41 that is very likely to be shared
0:06:43 with a wide range of other animals.
0:06:45 – I do a fair bit of reporting
0:06:48 on these kinds of questions of consciousness and sentience.
0:06:51 And everyone tends to agree that it’s a mystery, right?
0:06:53 And so a lot of emphasis goes on
0:06:56 trying to dispel the mystery.
0:06:58 And what I found really interesting about your approach
0:07:00 is that you seem to take the uncertainty
0:07:02 in the mystery as your starting point.
0:07:04 And rather than focusing on how do we solve this?
0:07:06 How do we dispel it?
0:07:07 You’re trying to help us think through
0:07:11 how to make practical decisions given that uncertainty.
0:07:13 I’m curious how you came to that approach.
0:07:14 – Yeah, the question for me
0:07:16 is how do we live with this uncertainty?
0:07:20 How do we manage risk better than we’re doing at present?
0:07:25 How can we use ideas from across science and philosophy
0:07:28 to help us make better decisions
0:07:29 when faced with those problems?
0:07:32 And in particular to help us err on the side of caution.
0:07:34 – Just to maybe make it explicit,
0:07:37 you mentioned the risk of uncertainty.
0:07:39 What is the risk here?
0:07:41 – Well, it depends on the particular case
0:07:42 we’re thinking about.
0:07:44 One of the cases that brought me to this topic
0:07:47 was the practice of dropping crabs and lobsters
0:07:49 into pans of boiling water.
0:07:52 And it seems like a clear case to me
0:07:55 where you don’t need certainty actually.
0:07:56 You don’t even need knowledge.
0:08:00 You don’t need high probability to see the risk.
0:08:04 And in fact, to do sensible common sense things
0:08:05 to reduce that risk.
0:08:07 – So the risk is the suffering we’re imposing
0:08:10 on these potentially other sentient creatures.
0:08:13 – That’s usually what looms largest for me, yeah.
0:08:15 The risk of doing things
0:08:17 that mean we end up living very badly
0:08:20 because we cause enormous amounts of suffering
0:08:22 to the creatures around us.
0:08:26 And you can think of that as a risk to the creatures
0:08:28 that end up suffering, but it’s also a risk to us.
0:08:31 A risk that our lives will be horrible
0:08:32 and destructive and absurd.
0:08:35 – I worry about my life being horrible
0:08:37 and destructive and absurd all the time.
0:08:39 So this is a handy way to think about it.
0:08:40 – We all should.
0:08:43 – I’d like to turn to your very practical work,
0:08:45 advising the UK government
0:08:49 on the Animal Welfare and Sentience Act of 2022.
0:08:50 The question was put to you
0:08:53 of whether they should consider certain invertebrates
0:08:55 like octopus and crabs and lobsters,
0:08:58 whether they should be included and protected in the bill.
0:09:01 Could you just give a little context on that story
0:09:02 and what led the government to come
0:09:05 and ask you to lead a research team on that question?
0:09:08 – Yeah, it was indirectly a result of Brexit,
0:09:11 the UK leaving the European Union,
0:09:15 because in doing that, we left the EU’s Lisbon Treaty
0:09:18 that has a line in it about respecting animals
0:09:20 as sentient beings.
0:09:22 And so Animal Welfare Organization said to the government,
0:09:25 are you going to import that into UK law?
0:09:27 And they said, no.
0:09:29 And they got a lot of bad press along the lines of,
0:09:32 well, don’t you think animals feel pain?
0:09:35 And so they promised new legislation
0:09:38 that would restore respect for sentient beings
0:09:40 back to UK law.
0:09:43 And they produced a draft of the bill
0:09:46 that included vertebrate animals.
0:09:48 You could say that’s progressive in a way
0:09:50 because fishes are in there, which is great,
0:09:52 but it generated a lot of criticism
0:09:55 because of the omission of invertebrates.
0:09:58 And so in that context, they commissioned a team led by me
0:10:01 to produce a review of the evidence of sentience
0:10:03 in two groups of invertebrates,
0:10:06 the cephalopods like octopuses
0:10:09 and the decopod crustaceans like crabs and lobsters.
0:10:11 I’d already been calling for applications
0:10:14 of the precautionary principle to questions of sentience
0:10:16 and had written about that.
0:10:19 And it already established at the LSE a project
0:10:22 called the Foundations of Animal Sentience Project
0:10:25 that aims to try to place the emerging science
0:10:29 of animal sentience on more secure foundations,
0:10:31 advance it, develop better methods,
0:10:33 and find new ways of putting the science to work
0:10:35 to design better policies,
0:10:37 laws and ways of caring for animals.
0:10:39 So in a way, I was in the right place at the right time.
0:10:43 I was pretty ideally situated to be leading a review like this.
0:10:46 – How do folks actually go about trying to answer
0:10:51 the question of whether a given animal is or is not sentient?
0:10:53 – Well, in lots of different ways.
0:10:55 And I think when we’re looking at animals
0:10:58 that are relatively close to us in evolutionary terms,
0:11:00 like other mammals,
0:11:02 neuroscience is a huge part of it
0:11:05 because we can look for similarities of brain mechanism.
0:11:08 But when thinking about crabs and lobsters,
0:11:09 what we’re not going to find
0:11:11 is exactly the same brain mechanisms
0:11:13 because we’re separated from them
0:11:16 by over 500 million years of evolution.
0:11:17 – That’s quite a bit.
0:11:19 – And so I think in that context,
0:11:23 you can ask big picture neurological questions.
0:11:27 Are there integrative brain regions, for example?
0:11:29 But the evidence is quite limited,
0:11:33 and so behavior ends up carrying a huge amount of weight.
0:11:36 Some of the strongest evidence comes from behaviors
0:11:41 that show the animal valuing pain relief when injured.
0:11:45 So for example, there was a study by Robin Crook
0:11:46 on octobuses, which is where you give the animal
0:11:49 a choice of two different chambers,
0:11:52 and you see which one it initially prefers.
0:11:56 And then you allow it to experience the effects
0:12:00 of a noxious stimulus, a nasty event.
0:12:03 And then in the other chamber that it initially dispreferred,
0:12:07 you allow it to experience the effects of an aesthetic
0:12:10 or a pain relieving drug.
0:12:12 And then you see whether its preferences reverse.
0:12:14 So now going forward,
0:12:17 it goes to that chamber where it had a good experience
0:12:20 rather than the one where it had a terrible experience.
0:12:22 So it’s a pattern of behavior.
0:12:26 In ourselves, this would be explained by feeling pain
0:12:28 and then getting relief from the pain.
0:12:30 And when we see it in other mammals,
0:12:32 we make that same inference.
0:12:34 – Are there any other categories?
0:12:36 ‘Cause we mentioned pain is one bucket of sentience,
0:12:38 but there’s much more to it.
0:12:39 Is there anything else that tends to play
0:12:42 a big role in the research?
0:12:42 – There’s much more to it.
0:12:44 And what I would like to see in the future
0:12:47 is animal sentience research moving beyond pain
0:12:50 and looking for other states that matter,
0:12:53 like joy for instance.
0:12:58 In practice though, by far the largest body of literature
0:13:01 exists for looking at markers of pain.
0:13:05 – I would love to read a paper that tries to assess
0:13:07 to what degree rats are experiencing joy
0:13:09 rather than pain, that would be lovely.
0:13:12 – I mean, studies of play behavior are very relevant here.
0:13:16 The studies of rats playing hide and seek for example,
0:13:18 where there must be something motivating
0:13:20 these play behaviors.
0:13:23 In the human case, we would call it joy, delight,
0:13:26 excitement, something like that.
0:13:29 And so it gets you taking seriously the possibility
0:13:32 there might be something like that in other animals too.
0:13:34 – I think the thing I’m actually left wondering is
0:13:39 what animals don’t show signs of sentience in these cases?
0:13:42 – Right, I mean, there’s many invertebrates
0:13:45 where you have an absence of evidence
0:13:48 ’cause no one has really looked.
0:13:53 So snails for example, there’s frustratingly little evidence.
0:13:58 Also bivalve mollusks, which people talk about a lot
0:14:00 ’cause they eat so many of them.
0:14:03 Very, very little evidence to base our judgments on.
0:14:05 And it’s hard to know what to infer from this.
0:14:07 There’s this slogan that absence of evidence
0:14:09 is not evidence of absence.
0:14:11 And it’s a little bit oversimplifying
0:14:13 ’cause you sort of think, well, you know,
0:14:18 when researchers find some indicators of pain,
0:14:20 they’ve got strong motivations to press on
0:14:22 because it could be a useful pain model
0:14:24 for biomedical research.
0:14:27 And this is exactly what we’ve seen in insects,
0:14:29 particularly Drosophila fruit flies,
0:14:31 that seeing some of those initial markers
0:14:33 has led scientists to think, well, let’s go for this.
0:14:38 And it turns out they’re surprisingly useful pain models.
0:14:39 – A pain model for humans?
0:14:40 – Right, exactly.
0:14:44 Yeah, that traditionally biomedical researchers have used rats
0:14:48 and there’s pressure to replace.
0:14:50 I don’t personally think that replacement here
0:14:53 should mean replacing mammals with invertebrates.
0:14:56 It’s not really the kind of replacement that I support,
0:14:59 but that is how a lot of scientists understand it.
0:15:03 And so they’re looking for ways to replace rats with flies.
0:15:04 – How do they decide
0:15:06 that the fly is a good pain model for humans?
0:15:08 – I mean, researchers have the ability
0:15:11 to manipulate the genetics of flies
0:15:16 at very, very fine grains using astonishing technologies.
0:15:22 So there was a recent paper that basically installed
0:15:26 in some flies sensitivity to chili heat.
0:15:30 Which of course in us, over a certain threshold,
0:15:32 this becomes painful.
0:15:34 So if you have one of the hottest chilies in the world,
0:15:37 you’re not gonna just carry on as normal.
0:15:38 – Certainly not.
0:15:40 – And they showed that the same behavior
0:15:41 can be produced in flies.
0:15:45 You can engineer them to be responsive to chili
0:15:48 and then you can dial up the amount of capsaicin
0:15:49 in the food they’re eating.
0:15:52 And there’ll come a point where they just stop eating
0:15:57 and withdraw from food, even though it leads them to starve.
0:16:00 And things like this that you’re leading researchers
0:16:03 to say, wow, the mechanisms here are mechanisms
0:16:07 we can use for testing out potential pain relieving drugs.
0:16:11 And the fruit flies are a standard model organism,
0:16:13 as they say in science.
0:16:16 So there’s countless numbers of them,
0:16:18 but traditionally they’ve been studied
0:16:20 for genetics primarily.
0:16:21 People haven’t been thinking of them
0:16:24 as model systems of cognitive functions
0:16:28 or of sentience or of pain or of sociality.
0:16:31 And they’re realizing to their surprise
0:16:33 that they’re very good models of all of these things.
0:16:35 And then your question is, well,
0:16:38 why is it such a good model of these things?
0:16:42 Could it be in fact that it possesses sentience of some kind?
0:16:46 – I don’t wanna go too far down this rabbit hole
0:16:48 ’cause I could spend hours asking you about this.
0:16:52 Let’s swing back to your research on the UK’s Act for a second.
0:16:54 You wound up recommending that the invertebrates
0:16:56 you looked at should be included.
0:16:59 And you mentioned this included, you know, octopuses,
0:17:01 which to me seems straightforward.
0:17:04 These seem very intelligent and playful.
0:17:06 I don’t need a lot of research to convince me of that.
0:17:09 But you recommended things like, you know, crabs and lobsters
0:17:11 and things where maybe people’s intuitions differ
0:17:14 a little bit in practical terms.
0:17:17 What changed for the life of a crab
0:17:20 after the UK did formally include them in the bill?
0:17:23 How does that wind up benefiting crabs?
0:17:26 – It’s a topic of ongoing discussion, basically,
0:17:27 ’cause what this new act does
0:17:30 is it creates a duty on policymakers
0:17:33 to consider the animal welfare consequences
0:17:36 of their decisions, including to crabs.
0:17:39 Now, we recommended, don’t just put crabs
0:17:41 in this particular act.
0:17:45 Also, amend the UK’s other animal welfare laws
0:17:48 to be consistent with the new act.
0:17:49 And this we’ve not yet seen.
0:17:52 So we’re really hoping that this will happen
0:17:53 and will happen in the near future.
0:17:56 And it’s something that definitely should happen.
0:17:58 ‘Cause in the meantime, we’ve got a rather confusing picture
0:18:01 where you have these other laws that say
0:18:04 animals should not be caused unnecessary suffering
0:18:07 when they’re killed and people should require training
0:18:09 if they’re going to slaughter animals.
0:18:12 And then you have this new law that says
0:18:14 for legal purposes, decapod crustaceans
0:18:16 are to be considered animals.
0:18:19 And as a philosopher, I’m always thinking,
0:18:20 well, read these two things together
0:18:22 and think about what they logically imply
0:18:23 when written together.
0:18:26 And lawyers don’t like that kind of argument.
0:18:28 Lawyers want a clear precedent
0:18:31 where there’s been some kind of test case
0:18:35 that has convicted someone for boiling a lobster alive
0:18:36 or something like that.
0:18:38 And that’s what we’ve not yet had.
0:18:41 So I’m hoping that lawmakers will act
0:18:44 to clarify that situation.
0:18:46 To me, it’s kind of clear.
0:18:47 How much clearer could it be
0:18:50 that this method causes unnecessary suffering
0:18:51 quite obviously.
0:18:56 And it’s illegal to do that to any animal,
0:18:58 including crabs.
0:19:02 But in practice, because it’s not explicitly ruled out,
0:19:07 it’s not quite good enough at the moment.
0:19:09 We wanna see this explicitly ruled out.
0:19:13 – So we’ll take incremental steps to get there.
0:19:15 – Yeah, in a way, I’m glad people take this issue
0:19:16 seriously at all.
0:19:19 I didn’t really expect that when I started working on it.
0:19:23 And so to have achieved any policy change that benefits
0:19:25 crabs and lobsters in any way,
0:19:27 I’ve gotta count that as a win.
0:19:40 – Support for the gray area comes from Mint Mobile.
0:19:42 There’s nothing like the satisfaction
0:19:44 of realizing you just got an incredible deal.
0:19:47 But those little victories have gotten harder
0:19:48 and harder to find.
0:19:50 Here’s the good news though.
0:19:52 Mint Mobile is resurrecting that incredible
0:19:54 “I got a deal” feeling.
0:19:56 Right now, when you make the switch to a Mint Mobile plan,
0:19:59 you’ll pay just $15 a month when you purchase
0:20:01 a new three month phone plan.
0:20:03 All Mint Mobile plans come with high speed data
0:20:05 and unlimited talk and text delivered
0:20:08 on the nation’s largest 5G network.
0:20:10 You can even keep your phone, your contacts,
0:20:11 and your number.
0:20:13 It doesn’t get much easier than that.
0:20:15 To get this new customer offer
0:20:17 and your new three month premium wireless plan
0:20:18 for just 15 bucks a month,
0:20:21 you can go to mintmobile.com/grayarea.
0:20:24 That’s mintmobile.com/grayarea.
0:20:27 You can cut your wireless bill to 15 bucks a month
0:20:29 at mintmobile.com/grayarea.
0:20:33 $45 upfront payment required equivalent to $15 a month.
0:20:36 New customers on first three month plan only.
0:20:39 Speed slower above 40 gigabytes on unlimited plan,
0:20:41 additional taxes, fees, and restrictions apply.
0:20:43 See Mint Mobile for details.
0:20:50 Support for the gray area comes from Cook Unity.
0:20:52 You know one way to eat chef prepared meals
0:20:53 in the comfort of your home?
0:20:56 You can spend years at culinary school,
0:20:58 work your way up the restaurant industry,
0:21:00 become a renowned chef on your own,
0:21:02 and then cook something for yourself.
0:21:04 Cook Unity delivers meals to your door
0:21:06 that are crafted by award winning chefs
0:21:10 and made with local farm fresh ingredients.
0:21:13 Cook Unity’s selection of over 350 meals
0:21:15 offers a variety of cuisines
0:21:17 and their menus are updated weekly.
0:21:18 So you’re sure to find something
0:21:21 to fit your taste and dietary needs.
0:21:22 One of our colleagues, Nisha,
0:21:24 tried Cook Unity for herself.
0:21:26 – Sometimes you’re just too tired to cook.
0:21:28 I’m a, I have a two and a half year old.
0:21:31 Sometimes you’re just exhausted at the end of the day.
0:21:33 And it’s very easy to default to take out.
0:21:35 So it was really nice to not have the mental load
0:21:37 of having it cook every day,
0:21:40 but having healthy home cooked meals
0:21:42 already prepared for you
0:21:45 and not having to go the takeout route.
0:21:48 – You can get the gift of delivering mouthwatering meals
0:21:49 crafted by local ingredients
0:21:52 and award winning chefs with Cook Unity.
0:21:55 You can go to cookunity.com/grayarea
0:21:58 or enter code grayarea before checkout
0:22:00 for 50% off your first week.
0:22:02 That’s 50% off your first week
0:22:04 by using code grayarea
0:22:07 or going to cookunity.com/grayarea.
0:22:14 – Support for the gray area comes from Shopify.
0:22:18 Viral marketing campaigns have gotten pretty wild lately.
0:22:19 Like in Russia,
0:22:22 one pizza chain offered 100 free pizzas a year
0:22:24 for 100 years to anyone
0:22:26 who got the company logo tattooed on their body.
0:22:29 Apparently 400 misguided souls did it,
0:22:33 which is a story that deserves its own podcast.
0:22:34 But if you want to grow your company
0:22:38 without resorting to a morally dubious viral scheme,
0:22:40 you might want to check out Shopify.
0:22:43 Shopify is an all-in-one digital commerce platform
0:22:46 that wants to help your business sell better than ever before.
0:22:49 Shopify says they can help you convert browsers
0:22:52 into buyers and sell more over time.
0:22:55 And their shop pay feature can boost conversions by 50%.
0:22:58 There’s a reason companies like Allbirds turn to Shopify
0:23:01 to sell more products to more customers.
0:23:03 Businesses that sell more sell with Shopify.
0:23:05 Want to upgrade your business
0:23:07 and get the same checkout Allbirds uses?
0:23:10 You can sign up for your $1 per month trial period
0:23:14 at Shopify.com/Vox, all lowercase.
0:23:18 That’s Shopify.com/Vox to upgrade your selling today.
0:23:20 Shopify.com/Vox.
0:23:28 (gentle music)
0:23:38 – Let’s move to another set of potential beings.
0:23:41 Your work on Sentience covers artificial intelligence.
0:23:44 And one of the things that I’ve been most interested
0:23:46 in watching as the past few years
0:23:48 have really thrust a lot of questions around AI
0:23:52 into the mainstream has been this unbundling
0:23:54 of consciousness and intelligence
0:23:56 or Sentience and intelligence.
0:23:59 We’re clearly getting better at creating
0:24:01 more intelligent systems that can achieve
0:24:05 and with competency perform certain tasks.
0:24:07 But it remains very unclear
0:24:09 if we’re getting any closer to Sentient ones.
0:24:12 So how do you understand the relationship
0:24:15 between Sentience and intelligence?
0:24:17 – I think it’s entirely possible
0:24:21 that we will get AI systems with very high levels
0:24:26 of intelligence and absolutely no Sentience at all.
0:24:27 That’s entirely possible.
0:24:31 And when you think about shrimps or snails, for example,
0:24:34 we can also conceive of how there can be Sentience
0:24:37 with perhaps not all that much intelligence.
0:24:40 – On another podcast, you had mentioned that
0:24:43 it might actually be easier to create AI systems
0:24:45 that are Sentient by modeling them
0:24:47 off of less intelligent systems
0:24:50 rather than just cranking up the intelligence dial
0:24:52 until it bursts through into Sentience.
0:24:54 Why is that?
0:24:55 – That could absolutely be the case.
0:24:59 I see it many possible pathways to Sentient AI.
0:25:01 One of which is through the emulation
0:25:03 of animal nervous systems.
0:25:06 There’s a long running project called Open Worm
0:25:09 that tries to recreate the nervous system
0:25:14 of a tiny worm called C. elegans in computer software.
0:25:16 There’s not a huge amount of funding going into this
0:25:19 because it’s not seen as very lucrative,
0:25:20 just very interesting.
0:25:23 And so even with those very simple nervous systems,
0:25:25 we’re not really at the stage where we can say
0:25:27 they’ve been emulated.
0:25:28 But you can see the pathway here.
0:25:31 You know, suppose we did get an emulation
0:25:33 of a worms nervous system.
0:25:35 I’m sure we would then move on to fruit flies.
0:25:39 If that worked, researchers would be going on to open mouse,
0:25:43 open fish and emulating animal brains
0:25:45 at ever greater levels of detail.
0:25:49 And then in relation to questions of Sentience,
0:25:51 we’ve got to take seriously the possibility
0:25:56 that Sentience does not require a biological substrate,
0:25:59 that the stuff you’re made of might not matter.
0:26:01 It might matter, but it might not.
0:26:03 And so it might be that if you recreate
0:26:08 the same functional organization in a different substrate,
0:26:11 so no neurons of a biological kind anymore,
0:26:13 just computer software,
0:26:15 maybe you would create Sentience as well.
0:26:18 – You’ve talked about this idea that you’ve called
0:26:20 the N equals one problem.
0:26:22 Can you explain what that is?
0:26:28 – Well, this is a term that began in origins of life studies,
0:26:31 where it’s people searching for extraterrestrial life
0:26:34 or studying life’s origin and asking,
0:26:37 well, we only have one case to draw on.
0:26:39 And if we only have one case,
0:26:43 how are we supposed to know what was essential to life
0:26:46 from what was a contingent feature
0:26:48 of how life was achieved on Earth?
0:26:53 And one might think we have an N equals one problem
0:26:55 with consciousness as well.
0:26:59 If you think it’s something that has only evolved once,
0:27:01 seems like you’re always gonna have problems
0:27:02 disentangling what’s essential to it
0:27:06 from what is contingent.
0:27:06 Luckily though,
0:27:09 I think we might be in an N greater than one situation
0:27:12 when it comes to Sentience and consciousness
0:27:14 because of the arthropods like flies and bees
0:27:17 and because of the cephalopods and crabs.
0:27:18 And because of the cephalopods
0:27:21 like octopuses, squid, cuttlefish,
0:27:24 we might even be in an N equals three situation,
0:27:28 in which case, studying those other cases,
0:27:32 octopuses, crabs, insects has tremendous value
0:27:35 for understanding the nature of Sentience
0:27:37 ’cause it can tell us,
0:27:39 it can start to give us some insight
0:27:43 into what might be essential to having it at all
0:27:45 versus what might be a quirk
0:27:48 of how it is achieved in humans.
0:27:50 – Just to make sure I have this right,
0:27:54 if we are in an N equals one scenario with Sentience,
0:27:56 that means that every sentient creature evolved
0:27:59 from the same sentient ancestor.
0:28:01 It’s one evolutionary lineage.
0:28:01 – That’s right.
0:28:05 – And so Sentience has only evolved once on Earth’s history
0:28:07 so it gives us one example to look at.
0:28:08 – Exactly.
0:28:10 – But if we’re not in an N equals one situation,
0:28:12 you mentioned N equals three
0:28:13 and there’s a fair bit of research
0:28:16 suggesting this could be the case or something like it,
0:28:19 then Sentience has evolved three separate times
0:28:22 in three separate kind of cases of form
0:28:24 and the architecture of a being.
0:28:26 – That’s fascinating to me,
0:28:28 the idea that Sentience could have involved
0:28:31 independently multiple times in different ways.
0:28:34 – Yeah, we know it’s true of eyes, for example,
0:28:37 when you look at the eyes of cephalopods,
0:28:40 you see a wonderful mixture of similarities and differences.
0:28:44 So we see convergent evolution, similar thing,
0:28:49 evolving independently to solve a similar problem
0:28:51 and Sentience could be just like that.
0:28:55 – The greater the number of N’s we have here,
0:28:59 the number of separate instances of Sentience evolving,
0:29:03 it strikes me as that lends more credence to the idea
0:29:06 that AI could develop its own independent route
0:29:09 to Sentience as well that might not look exactly
0:29:11 like what we’ve seen in the past.
0:29:14 – It’s also the way towards really knowing
0:29:16 whether it has or not as well
0:29:19 because at present, we’re just not in that situation.
0:29:22 We’re not in a good enough position
0:29:25 to be able to really know that we’ve created Sentience AI
0:29:28 even when we do, we’ll be faced
0:29:31 with horrible disorienting uncertainty.
0:29:33 But to me, the pathway towards better evidence
0:29:36 and maybe one day knowledge lies through
0:29:38 studying other animals.
0:29:41 And it lies through trying to get other N’s,
0:29:45 other independently evolved cases
0:29:48 so that we can develop theories
0:29:51 that genuinely disentangle the quirks
0:29:54 of human consciousness from what is needed
0:29:56 to be conscious at all.
0:30:01 – What kind of evidence would you find compelling
0:30:05 that tests for Sentience in AI systems?
0:30:08 – It’s something I’ve been thinking about a great deal
0:30:12 because when we’re looking at the surface linguistic behavior
0:30:14 of an AI system that has been trained
0:30:18 on over a trillion words of human training data,
0:30:22 we clearly gonna see very fluent talking
0:30:24 about feelings and emotions.
0:30:29 And we’re already seeing that.
0:30:33 And it’s really, I would say not evidence at all
0:30:36 that the system actually has those feelings
0:30:40 because it can be explained as a kind of skillful mimicry.
0:30:43 And if that mimicry serves the system’s objectives,
0:30:45 we should expect to see it.
0:30:48 We should expect our criteria to be gained
0:30:50 if the objectives are served by persuading
0:30:54 the human user of Sentience.
0:30:56 And so this is a huge problem and it points
0:31:01 to the need to look deeper in some way.
0:31:03 These systems are very substantially opaque.
0:31:07 It is really, really hard to infer anything
0:31:10 about what the processes are inside them.
0:31:12 And so I have a second line of research as well
0:31:15 that I’ve been developing with collaborators at Google
0:31:19 that is about trying to adapt some of these animal experiments.
0:31:22 Let’s see if we can translate them over to the AI case.
0:31:24 – These are looking for behavior changes?
0:31:27 – Yeah, looking for subtle behavior changes
0:31:32 that we hope would not be gaming
0:31:35 because they’re not part of the normal repertoire
0:31:37 in which humans express their feelings,
0:31:39 but are rather these very subtle things
0:31:41 that we’ve looked for in other animals
0:31:43 because they can’t talk about their feelings
0:31:45 in the first place.
0:31:47 – So it’s funny, we’re hitting the same problem in AI
0:31:49 that we are in animals and humans,
0:31:53 which is that in both cases, there’s a black box problem
0:31:54 where we don’t actually understand
0:31:55 the inner workings to some degree.
0:31:58 – The problems are so much worse in the AI case though
0:32:03 because when you’re faced with a pattern of behavior
0:32:07 in another animal like an octopus
0:32:10 that is well explained by there being a state
0:32:14 like pain there, that is the best explanation
0:32:15 for your data.
0:32:18 And it doesn’t have to compete with this other explanation
0:32:21 that maybe the octopus read a trillion words
0:32:23 about how humans express their feelings
0:32:27 and stands to benefit from gaming our criteria
0:32:29 and skillfully mimicking us.
0:32:33 We know the octopus is not doing that, that never arises.
0:32:36 In the AI case, those two explanations always compete
0:32:39 and the second one with current systems
0:32:41 seems to be rather more plausible.
0:32:43 And in addition to that,
0:32:46 the substrate is completely different as well.
0:32:48 So we face huge challenges
0:32:49 and I suppose what I’m trying to do
0:32:51 is maintain an attitude of humility
0:32:53 in the face of those challenges.
0:32:57 Now, let’s not be credulous about this,
0:32:59 but also let’s not give up the search
0:33:02 for developing higher quality kinds of test.
0:33:14 Support for the gray area comes from green light.
0:33:28 Anyway, it applies to more than fish.
0:33:30 It’s also a great lesson for parents
0:33:32 who want their kids to learn important skills
0:33:35 that will set them up for success later in life.
0:33:37 As we enter the gifting season,
0:33:39 now might be the perfect time
0:33:41 to give your kids money skills that will last
0:33:43 well beyond the holidays.
0:33:45 That’s where green light comes in.
0:33:46 Green light is a debit card
0:33:49 and money at made specifically with families in mind.
0:33:50 Send money to your kids,
0:33:52 track their spending and saving
0:33:54 and help them develop their financial skills
0:33:57 with games aimed at building the confidence they need
0:34:00 to make wiser decisions with their money.
0:34:02 My kid is a little too young for this.
0:34:04 We’re still rocking piggy banks,
0:34:06 but I’ve got a colleague here at Vox
0:34:08 who uses it with his two boys and he loves it.
0:34:10 You can sign up for green light today
0:34:12 at greenlight.com/grayarea.
0:34:14 That’s greenlight.com/grayarea
0:34:16 to try green light today.
0:34:18 Greenlight.com/grayarea.
0:34:24 Support for the show comes from Give Well.
0:34:27 When you make a charitable donation,
0:34:30 you want to know your money is being well spent.
0:34:32 For that, you might want to try Give Well.
0:34:34 Give Well is an independent nonprofit
0:34:36 that’s spent the last 17 years
0:34:39 researching charitable organizations.
0:34:40 And they only give recommendations
0:34:44 to the highest impact causes that they vetted thoroughly.
0:34:47 According to Give Well, over 125,000 donors
0:34:50 have used it to donate more than $2 billion.
0:34:52 Rigorous evidence suggests that these donations
0:34:55 could save over 200,000 lives.
0:34:58 And Give Well wants to help you make informed decisions
0:34:59 about high impact giving.
0:35:02 So all of their research and recommendations
0:35:04 are available on their site for free.
0:35:06 You can make tax deductible donations
0:35:08 and Give Well doesn’t take a cut.
0:35:09 If you’ve never used Give Well to donate,
0:35:12 you can have your donation matched up to $100
0:35:14 before the end of the year
0:35:16 or as long as matching funds last.
0:35:18 To claim your match, you can go to GiveWell.org
0:35:22 and pick podcast and enter the gray area at checkout.
0:35:24 Make sure they know that you heard about Give Well
0:35:26 from the gray area to get your donation matched.
0:35:30 Again, that’s GiveWell.org to donate or find out more.
0:35:37 Support for the gray area comes from Delete Me.
0:35:39 Delete Me allows you to discover and control
0:35:40 your digital footprint,
0:35:43 letting you see where things like home addresses,
0:35:44 phone numbers, and even email addresses
0:35:47 are floating around on data broker sites.
0:35:50 And that means Delete Me could be the perfect holiday gift
0:35:52 for a loved one looking to help safeguard
0:35:54 their own life online.
0:35:55 Delete Me can help anyone monitor
0:35:59 and remove the personal info they don’t want on the internet.
0:36:01 Claire White, our colleague here at Vox,
0:36:04 tried Delete Me for herself and even gifted it to a friend.
0:36:08 This year, I gave two of my friends a Delete Me subscription
0:36:09 and it’s been the perfect gift
0:36:12 ’cause it’s something that will last beyond the season.
0:36:14 Delete Me will continue to remove their information
0:36:16 from online and it’s something I’ve been raving about
0:36:19 so I know that they’re gonna love it as well.
0:36:21 – This holiday season, you can give your loved ones
0:36:25 the gift of privacy and peace of mind with Delete Me.
0:36:27 Now at a special discount for our listeners.
0:36:30 Today, you can get 20% off your Delete Me plan
0:36:33 when you go to joindeleteme.com/vox
0:36:35 and use promo code Vox at checkout.
0:36:39 The only way to get 20% off is to go to joindeleteme.com/vox
0:36:42 and enter code Vox at checkout.
0:36:46 That’s joindeleteme.com/vox code Vox.
0:36:52 (gentle music)
0:37:00 – One of the major aims of your recent book
0:37:02 is to propose a framework
0:37:04 for making these kinds of practical decisions
0:37:06 about potentially sentient creatures,
0:37:09 whether it’s an animal, whether it’s AI,
0:37:11 given this uncertainty.
0:37:12 Tell me about that framework.
0:37:15 – Well, it’s a precautionary framework.
0:37:18 One of the things I urge is a pragmatic shift
0:37:20 in how we think about the question.
0:37:23 From asking, is the system sentient
0:37:26 where uncertainty will always be with us?
0:37:30 To asking instead, is the system a sentient’s candidate?
0:37:32 Where the concept of a sentient’s candidate
0:37:36 is a concept that we’ve pragmatically engineered.
0:37:39 And what it says is that a system is a sentient’s candidate
0:37:41 when there’s a realistic possibility of sentient’s
0:37:44 that it would be irresponsible to ignore.
0:37:45 And when there’s an evidence base
0:37:49 that can inform the design and assessment of precautions.
0:37:52 And because we’ve constructed the concept like that,
0:37:56 we can use current evidence to make judgments.
0:37:59 The cost of doing that is that those judgments
0:38:02 are not purely scientific judgments anymore.
0:38:05 There’s an ethical element to the judgment as well,
0:38:07 because it’s about when a realistic possibility
0:38:10 becomes irresponsible to ignore.
0:38:13 And that’s implicitly a value judgment.
0:38:15 But by reconstructing the question in that way,
0:38:16 we make it answerable.
0:38:19 – So, presumably then,
0:38:21 given your recommendation to the UK government,
0:38:25 you would say that those invertebrates you looked at
0:38:27 are sentient’s candidates.
0:38:29 That there’s enough evidence to at least consider
0:38:31 the possibility of sentience.
0:38:35 Where would you stop with the current category
0:38:36 of sentient’s candidate?
0:38:39 What is not a sentient’s candidate in your current view?
0:38:43 – I’ve come to the view that insects really are,
0:38:44 which surprises me.
0:38:46 You know, it would have surprised past me
0:38:49 who hadn’t read so much of the literature about insects.
0:38:53 The evidence just clearly shows a realistic possibility
0:38:55 of sentience, it would be irresponsible to ignore.
0:38:59 But really, that’s currently where I stop.
0:39:01 So the cephalopod mollusks, the decapod crustaceans,
0:39:05 the insects, lot of evidence in those cases.
0:39:08 And I think in other invertebrates,
0:39:11 what we should say instead is that we lack the kind of
0:39:14 evidence that would be needed to effectively design
0:39:17 precautions to manage welfare risks.
0:39:21 And so the imperative there is to be getting more evidence.
0:39:24 And so in my book, I call these investigation priorities.
0:39:27 – So insects are sentience candidates.
0:39:30 Where does today’s generation of AI,
0:39:33 let’s LLMs in particular, so open AI is chatgy, BT,
0:39:36 anthropics, Claude, are these sentience candidates
0:39:37 in your view yet?
0:39:40 – I suggest that they’re investigation priorities,
0:39:43 which is already controversial because I’m saying that,
0:39:47 well, just as snails, we need more evidence.
0:39:49 Equally in AI, we need more evidence.
0:39:51 So I’m not being one of those people who just dismisses
0:39:56 the possibility of sentient AI as being a ridiculous one.
0:39:58 But I don’t think they’re sentience candidates
0:40:00 because we don’t have enough evidence.
0:40:02 – When you say that something is a sentience candidate,
0:40:05 it’s implying that we need to consider their welfare
0:40:08 and our behaviors and the decisions that we make.
0:40:09 – In public policy.
0:40:10 Yeah, I mean, in our personal lives,
0:40:13 we might want to be even more precautionary,
0:40:17 but I’m designing here a framework for setting policy.
0:40:19 – Right, ’cause I can imagine,
0:40:21 I think that the standard kind of line
0:40:23 that you get at this point is,
0:40:24 if you’re telling me I need to consider
0:40:26 the welfare of insects,
0:40:28 how can I take a step on the sidewalk?
0:40:30 And one of the ideas that’s central to your framework
0:40:33 is this idea of proportionality, which I really liked.
0:40:36 You talk about how the precautions that we take
0:40:39 should match the scale of the risk of suffering
0:40:41 that our actions kind of carry.
0:40:43 So how do you think about quantifying the risk
0:40:45 of suffering an action carries, right?
0:40:49 Does harming simpler creatures or insects
0:40:52 carry less risk than harming larger, more complex ones
0:40:54 like pigs or octopuses?
0:40:57 – Well, I’m opposed to trying to reduce it
0:41:00 or to a calculation and perhaps disagree
0:41:02 with some utilitarians on that point.
0:41:04 When you’re setting public policy,
0:41:08 cost-benefit analysis has its place,
0:41:10 but we’re not in that kind of situation here.
0:41:13 We’re weighing up very incommensurable things,
0:41:16 things that it’s very, very hard to compare.
0:41:18 And I think in that kind of situation,
0:41:21 you don’t want to be just making a calculation.
0:41:24 What you need to have is a democratic, inclusive process
0:41:27 through which different positions can be represented
0:41:32 and we can try to resolve our value conflicts democratically.
0:41:35 And so in the book, I advocate for citizens assemblies
0:41:39 as being the most promising way of doing this,
0:41:41 where you bring a random sample of the public
0:41:45 into an environment where they’re informed about the risks,
0:41:47 they’re informed about possible precautions,
0:41:50 and they’re given a series of tests to go through
0:41:53 to debate what they think would be proportionate
0:41:54 to those risks.
0:41:57 And things like, we’re all banned from walking now
0:41:59 because it might hurt insects.
0:42:02 I don’t see those as very likely to be judged proportionate
0:42:04 by such an exercise.
0:42:06 But other things we might do to help insects,
0:42:09 like banning certain kinds of pesticides,
0:42:12 I think might well be judged proportionate.
0:42:15 – Is this, this sounds to me almost like a form of jury duty.
0:42:17 You have a random selection of citizens brought together.
0:42:18 – Yeah.
0:42:20 – How do you, when I think about this on one hand,
0:42:21 I think it sounds lovely.
0:42:23 I like the idea of us all coming together
0:42:26 to debate the welfare of our fellow creatures.
0:42:28 It also strikes me as kind of optimistic,
0:42:32 to imagine us not only doing this, but doing it well.
0:42:35 And I’m curious how you think about balancing
0:42:38 the value of expertise in making these decisions
0:42:40 with democratic input.
0:42:44 – Yeah, I’m implicitly proposing a division of labor
0:42:47 where experts are supposed to make this judgment
0:42:51 of sentience candidature or candidacy.
0:42:53 Is the octopus a sentience candidate?
0:42:56 But then they’re not adjudicating
0:42:58 the questions of proportionality.
0:43:00 – So what to do about it?
0:43:02 – Yeah, then it would be a tyranny of expert values.
0:43:05 You’d have this question that calls for value judgments
0:43:06 about what to do.
0:43:08 And you’d be handing that over to the experts
0:43:12 and letting the experts dictate changes to our way of life.
0:43:15 That question of proportionality,
0:43:18 that should be handed over to the citizens assembly.
0:43:21 And I think it doesn’t require ordinary citizens
0:43:24 to adjudicate the scientific disagreement.
0:43:27 And that’s really crucial because if you’re asking
0:43:28 random members of the public to adjudicate
0:43:32 which brain regions they think are more important to sentience,
0:43:34 that’s gonna be a total disaster.
0:43:37 But the point is you give them questions
0:43:40 about what sorts of changes to our way of life
0:43:44 would be proportionate, would be permissible,
0:43:47 adequate, reasonably necessary and consistent
0:43:50 in relation to this risk that’s been identified.
0:43:52 And you ask them to debate those questions.
0:43:55 And I think that’s entirely feasible.
0:43:57 I’m very optimistic about citizens assemblies
0:44:00 as a mechanism for addressing that kind of question,
0:44:02 a question about our shared values.
0:44:05 – Do you see these as legally binding
0:44:07 or kind of making recommendations?
0:44:11 – I think they can only be making recommendations.
0:44:14 What I’m proposing is that on certain specific issues
0:44:17 where we think we need public input,
0:44:19 but we don’t wanna put them to a referendum
0:44:22 because we might need to revisit the issues
0:44:24 when new evidence comes to light
0:44:26 and you need a certain level of information
0:44:28 to understand what the issue is.
0:44:31 Citizens assemblies are great for those kinds of issues.
0:44:34 And because they’re very effective,
0:44:36 the recommendations they deliver
0:44:38 should be given weight by policymakers
0:44:40 and should be implemented.
0:44:43 They’re not substituting for parliamentary democracy,
0:44:46 but they’re feeding into it in a really valuable way.
0:44:51 – One thing that I can’t help but wonder about all of this,
0:44:54 humans are already incredibly cruel to animals
0:44:56 that most of us agree are very sentient,
0:44:58 I’m thinking of pigs or cows.
0:45:01 I think we’ve largely moved away from,
0:45:03 Descartes in the 1600s
0:45:06 where all animals were considered unfeeling machines.
0:45:09 Today we might disagree about how small
0:45:11 and simple down the chain we go
0:45:14 before we lose in consensus on sentience,
0:45:17 but agreeing that they’re sentient
0:45:18 doesn’t seem to have prevented us
0:45:21 from doing atrocious things to many animals.
0:45:24 So I’m curious if the goal is to help guide us
0:45:27 in making more ethical decisions,
0:45:29 how do you think that determining sentience
0:45:31 in other creatures will help?
0:45:37 – You’re totally right that recognizing animals as sentient
0:45:40 does not immediately lead to behavioral change
0:45:42 to treat them better.
0:45:45 And this is the tragedy of how we treat
0:45:48 lots of mammals like pigs and birds like chickens,
0:45:51 that we recognize them as sentient beings,
0:45:54 and yet we fail them very, very seriously.
0:45:57 I think there’s a lot of research to be done
0:46:01 about what kinds of information about sentience
0:46:03 might genuinely change people’s behavior.
0:46:07 And I’m very interested in doing that kind of research
0:46:12 going forward, but with cases like octopuses,
0:46:15 at least there’s quite an opportunity
0:46:16 in this particular case, I think,
0:46:20 because you don’t have really entrenched industries
0:46:22 already farming them.
0:46:24 Part of the problem we face with the pigs and chickens
0:46:28 and so on is that in opposing these practices,
0:46:31 the enemy is very, very powerful.
0:46:33 The arguments are really easy to state
0:46:38 and people do get them and they do see why this is wrong,
0:46:41 but then the enemy is so powerful
0:46:44 that actually changing this juggernaut,
0:46:47 this leviathan is a huge challenge.
0:46:51 By contrast with invertebrate farming,
0:46:54 we’re talking about practices sometimes
0:46:57 that could become entrenched like that in the future,
0:46:59 but are not yet entrenched.
0:47:04 Octopus farming is currently on quite small scales,
0:47:06 shrimp farming is much larger,
0:47:09 insect farming is much larger,
0:47:11 but they’re not as entrenched and powerful
0:47:14 as pig farming, poultry farming.
0:47:16 And so there seem to be real opportunities here
0:47:19 to effect positive change, or at least I hope so.
0:47:21 In the octopus farming case, for example,
0:47:24 we’ve actually seen bands implemented
0:47:27 in Washington State and in California.
0:47:31 And that’s a sign that progress is really possible
0:47:32 in these cases.
0:47:35 – There are talks of banning AI development.
0:47:37 The philosopher Thomas Metzinger is famously called
0:47:41 for a ban until 2050, that might be difficult operationally,
0:47:45 but I’m curious how you think about actions we can take today
0:47:47 at the early stages of these institutions
0:47:49 that might help in the long run.
0:47:51 – Yeah, huge problems.
0:47:55 I do think Metzinger’s proposal deserves to be taken seriously,
0:48:00 but also we need to be thinking about what can we do
0:48:05 that is more easily achieved than banning this stuff,
0:48:08 but then nonetheless makes a positive difference.
0:48:11 And in the book, I suggest there might be some lessons here
0:48:14 from the regulation of animal research
0:48:17 that you can’t just do what you like,
0:48:19 experimenting on animals.
0:48:22 In the UK, at least, there’s quite a strict framework
0:48:25 requiring you to get a license.
0:48:27 And it’s not a perfect framework by any means.
0:48:29 It has a lot of problems,
0:48:33 but it does show a possible compromise
0:48:35 between simply banning something altogether
0:48:39 and allowing it to happen in a completely unregulated way.
0:48:41 And the nature of that compromise
0:48:44 is that you expect the people doing this research
0:48:47 to be transparent about their plans,
0:48:49 to reveal their plans to a regulator.
0:48:53 Who is able to see them and assess the harms and benefits
0:48:54 and only give a license
0:48:57 if they think the benefits outweigh the harms.
0:49:01 And I’d like to see something like that in AI research
0:49:03 as well as in animal research.
0:49:04 – Well, it’s interesting
0:49:05 ’cause it brings us right back
0:49:06 to what you were talking about a little while ago,
0:49:10 which is, if we can’t trust the linguistic output,
0:49:12 we need the research on understanding,
0:49:14 well, how do we even assess harm and risk
0:49:16 in AI systems in the first place?
0:49:18 – As I say, it’s a huge problem coming down the road
0:49:20 for the whole of society.
0:49:24 I think there’ll be significant social divisions opening up
0:49:28 in the near future between people who are quite convinced
0:49:31 that their AI companions are sentient
0:49:33 and want rights for them
0:49:38 and others who simply find that ridiculous and absurd.
0:49:40 And I think that there’ll be a lot of tensions
0:49:42 between these two groups.
0:49:45 And in a way, the only way to really move forward
0:49:49 is to have better evidence than we do now.
0:49:52 And so there needs to be more research.
0:49:55 I’m always in this difficult position of,
0:49:57 I want more research, the tech companies might fund it,
0:49:59 I hope they will, I want them to fund it.
0:50:02 At the same time, it could be very problematic
0:50:04 for them as well.
0:50:06 And so I can’t make any promises in advance
0:50:08 that the outcomes of that research
0:50:11 will be advantageous to the tech companies.
0:50:14 So, but even though I’m in a difficult position there,
0:50:18 I feel like I still have to try and do something.
0:50:21 – Maybe by way of trying to wrap this all up,
0:50:24 you have been involved in these kinds of questions
0:50:25 for a number of years.
0:50:27 And you’ve mentioned a few times throughout the conversation
0:50:29 that you have seen a pace of change
0:50:31 that’s been kind of inspiring.
0:50:32 You’ve seen questions that previously
0:50:35 were not a part of the conversation now,
0:50:37 becoming part of the mainstream conversation.
0:50:41 So what have you seen in the last decade or two
0:50:43 in terms of the degree to which we are really beginning
0:50:44 to embrace these questions?
0:50:46 – I’ve seen some positive steps.
0:50:49 I think issues around crabs and lobsters and octopus
0:50:53 is taking far more seriously than they were 10 years ago.
0:50:56 For example, I really did not expect that California
0:51:00 would bring in an octopus farming ban
0:51:03 and in the legislation cite our work
0:51:07 as being a key factor driving it.
0:51:08 I mean, that was extraordinary.
0:51:11 So it just goes to show that it really pays off sometimes
0:51:14 to do impact driven work.
0:51:16 I think we’ve seen over the last couple of years
0:51:19 some changes in the conversations around AI as well.
0:51:23 The book is written in a very optimistic tone, I think,
0:51:26 because well, you’ve got to hope to make it a reality.
0:51:30 You’ve got to believe in the possibility of us
0:51:34 taking steps to manage risk better than we do.
0:51:36 And the book is full of proposals
0:51:38 about how we might do that.
0:51:42 And I think at least some of these will be adopted in the future.
0:51:49 – I would love to see it, I’m optimistic as well.
0:51:52 Jonathan Birch, thank you so much for coming on the show.
0:51:53 This was a pleasure.
0:51:54 – Thanks, Hachan.
0:52:07 – Once again, the book is The Edge of Sentience,
0:52:10 which is free to read on the Oxford Academic Platform.
0:52:13 We’ll include a link to that in the show notes.
0:52:14 And that’s it.
0:52:17 I hope you enjoyed the episode as much as I did.
0:52:20 I am still thinking about whether we’re in an N equals one
0:52:23 or an N equals three world,
0:52:25 and how the future of how we look for sentience
0:52:29 in AI systems could come down to animal research
0:52:30 that helps us figure out
0:52:35 whether all animals share the same sentient ancestor,
0:52:37 or whether sentience is something
0:52:40 that’s evolved a few separate times.
0:52:43 This episode was produced by Beth Morrissey
0:52:46 and hosted by me, O’Shan Jarrow.
0:52:50 My day job is as a staff writer with Future Perfect at Vox,
0:52:53 where I cover the latest ideas in the science
0:52:55 and philosophy of consciousness,
0:52:57 as well as political economy.
0:53:01 You can read my stuff at box.com/futureperfect.
0:53:04 Today’s episode was engineered by Patrick Boyd,
0:53:08 fact-checked by Anouk Dussot, edited by Jorge Just,
0:53:10 and Alex Overington wrote our theme music.
0:53:14 New episodes of The Gray Area drop on Mondays.
0:53:16 Listen and subscribe.
0:53:17 The show is part of Vox.
0:53:19 Support Vox’s journalism
0:53:22 by joining our membership program today.
0:53:25 Go to vox.com/members to sign up.
0:53:27 And if you decide to sign up because of the show,
0:53:28 let us know.
2