Jonathan Birch

These kinds of things are underway as we speak, I think. And it puts us in a really difficult position, I think, epistemically. They're really difficult to know what to say about these cases. In the book, I talk about the gaming problem, which is, I think, a huge problem in this area, which is that we've got our lists of markers developed in good faith for assessing crabs, octopuses, and so on.

These kinds of things are underway as we speak, I think. And it puts us in a really difficult position, I think, epistemically. They're really difficult to know what to say about these cases. In the book, I talk about the gaming problem, which is, I think, a huge problem in this area, which is that we've got our lists of markers developed in good faith for assessing crabs, octopuses, and so on.

If we just test for those same markers consistently, in the large language model case? Well, there's always going to be two explanations competing. Now, one is that it produces these markers because it genuinely has the state in question. And the other explanation is, well, it produces these markers because it has decided that it serves its objectives to persuade us of its sentience.

If we just test for those same markers consistently, in the large language model case? Well, there's always going to be two explanations competing. Now, one is that it produces these markers because it genuinely has the state in question. And the other explanation is, well, it produces these markers because it has decided that it serves its objectives to persuade us of its sentience.

And it knows the list of criteria from its training data that humans use to judge that question. And a lot of I think by default, that second explanation starts off as more plausible. And when you have people even now being persuaded by their AI assistants that they're sentient, it's not that they've got genuine evidence that they are.

And it knows the list of criteria from its training data that humans use to judge that question. And a lot of I think by default, that second explanation starts off as more plausible. And when you have people even now being persuaded by their AI assistants that they're sentient, it's not that they've got genuine evidence that they are.

It's that the AI assistants have various goals relating to user satisfaction, prolonging interaction time. And in service of those goals, they superficially mimic the way a sentient human would behave. And now that is a huge epistemological problem that we don't face when we're dealing with an octopus or a crab.

It's that the AI assistants have various goals relating to user satisfaction, prolonging interaction time. And in service of those goals, they superficially mimic the way a sentient human would behave. And now that is a huge epistemological problem that we don't face when we're dealing with an octopus or a crab.

Right, yes. If you're totally naive, yeah, there's ways in which even a cat might deceive you. But I guess I don't think vets, sort of experts, are being deceived. But in the AI case, well, there are no experts, as it were. Right. There's no easy way to be sure you're dealing with the real thing rather than skillful mimicry. And no one has a solution to that problem right now.

Right, yes. If you're totally naive, yeah, there's ways in which even a cat might deceive you. But I guess I don't think vets, sort of experts, are being deceived. But in the AI case, well, there are no experts, as it were. Right. There's no easy way to be sure you're dealing with the real thing rather than skillful mimicry. And no one has a solution to that problem right now.

Yeah, I think that's what the whole Edge of Sentience book is about. This family of cases at the Edge of Sentience where they all have this science meets policy aspect, where they're trying to make policy based on an incredibly uncertain scientific picture. And hopefully one of the roles for philosophy here is to try and stabilise that relationship

Yeah, I think that's what the whole Edge of Sentience book is about. This family of cases at the Edge of Sentience where they all have this science meets policy aspect, where they're trying to make policy based on an incredibly uncertain scientific picture. And hopefully one of the roles for philosophy here is to try and stabilise that relationship

and say, well, here is how you can make sensible precautionary policy on the basis of uncertain science.

and say, well, here is how you can make sensible precautionary policy on the basis of uncertain science.

Well, I mean, I hope that my book is helpful. Good. I hope so. I mean, one has to hope this. And we will see. It's... It's a book that should be judged on its consequences in a way because it's making all kinds of proposals for how we could manage risk better and how we could be more precautionary.

Well, I mean, I hope that my book is helpful. Good. I hope so. I mean, one has to hope this. And we will see. It's... It's a book that should be judged on its consequences in a way because it's making all kinds of proposals for how we could manage risk better and how we could be more precautionary.

And the book succeeds if people take those proposals seriously and discuss them and think about how they might implement them in their own lives and organizations, institutions, policies.

And the book succeeds if people take those proposals seriously and discuss them and think about how they might implement them in their own lives and organizations, institutions, policies.

Yeah, there's a tendency sometimes for people to say, maybe we'll never know. But if you say, but maybe we'll never know, that can't be a license to do whatever you want. It can't be a license to drop the crabs into pans of boiling water and so on. There's got to be sensible precautionary steps we can agree on in the face of uncertainty. And the book is about trying to find these.

Yeah, there's a tendency sometimes for people to say, maybe we'll never know. But if you say, but maybe we'll never know, that can't be a license to do whatever you want. It can't be a license to drop the crabs into pans of boiling water and so on. There's got to be sensible precautionary steps we can agree on in the face of uncertainty. And the book is about trying to find these.