The Edge of Sentience is an unusual book, integrating philosophy of mind, ethics, world Oxford University Press 2004 religion, philosophy of science, policy, and consciousness science. Written by Jonathan Birch Philosophy, London School of Economics, London UK, a philosopher whose research has impacted UK animal welfare policy, the book’s main goal is to offer advice about making policy decisions under conditions of uncertainty.

This is not a book that focuses merely on animal ethics, though, but one that situates animal ethics within a broader context: How do we make ethical and policy decisions about systems whose consciousness is uncertain? To answer the question Birch focuses on one dimension of consciosuness, namely sentience, which refers to valanced feelings such as the comfort of a gentle touch or the pain of a pinprick.

Birch looks at the possibility of sentience in five different systems: humans with disorders of consciousness, human fetuses, human neural organoids, nonhuman animals, and (yes) AI. In each case, the first question has to do with the reasonableness of ascribing sentience to the subject—the human in a coma, the fetus in a womb, the organoid in a dish, the octopus in the sea, and the LLM app running on a phone. For systems that qualify, we then must decide how to treat them. Birch takes as his ethical starting point that we have a duty to avoid causing gratuitous suffering (Framework Principle 1) and that precaution is warranted for those beings who have a realistic possibility of being sentient (Framework Principle 2).

Birch relies on these principles to defend twenty-six specific policy proposals in these five areas that are listed at the beginning of the book, though he defends them in only a modest way. Here is a taste of the kind of proposals he makes:

• “A patient with a prolonged disorder of consciousness should not be assumed incapable of experience when an important clinical decision is made.” (2)

• “Direct fetal anaesthesia and pain relief should be considered whenever therapeutic fetal surgery is performed.” (3)

• “It would be proportionate to ban octopus farming.” (5)

• “There should be a licensing scheme for companies attempting to create artificial sentience candidates.” (6).

The proposals are not presented as conclusions—Birch says he doesn’t want to be the “sentience tsar” (18)—but rather as policy recommendations for discussion.

If Birch isn’t going to decide what the policies should be, who should? This question is answered by the final framework principle, according to which democratic processes such as citizens panels should be called upon to assess the evidence and arguments and draw conclusions. Birch argues that citizen panels will avoid the “tyranny of expert values” and a well-constructed panel will be better placed to make decisions about what to do when values conflict. I admire, but can’t share, this optimism.

Before getting to any of these specific policy recommendations, Birch sets the stage in the first two sections of the book. In the first section, we get a high-level opiniated introduction to the philosophy of mind, Western ethical theories and world religious ethical commitments, and the science of consciousness. For philosophers working in these fields, there will be points of disagreement and quibbles about how views are presented, and objections about what gets left aside. But the point of these first three chapters isn’t to defend any view but to construct a “zone of reasonable disagreement.” In the ethical/religious context, we find that sentience is morally significant across several ethical and moral frameworks. In philosophy of mind/consciousness science, most existing theories are somewhere within the zone of reasonable disagreement.

The aim of this book isn’t to review the arguments for various theories of consciousness or ethical theories. Rather, the overarching aim is practical: Given what we do and don’t know about consciousness, how can our governments best make laws about how to handle these edge cases? How can precautionary thinking play out practically in these five cases?

Answering these questions is the task of the second section of the book—to propose a way of getting from our uncertain science, philosophy of mind, and ethics to actionable policies. The first step is to extract a meta-consensus from the zone of reasonable disagreement about the distribution of sentience in the natural world that is grounded in the systems’ neurobiology. This meta-consensus ranges from sentience being so narrowly distributed that it requires “distinctively primate neural mechanism” and only exists in primates, to sentience being “achieved in at least a minimal form by integrative subcortical mechanisms crucially involving the midbrain…[and] by other brain mechanisms performing relevantly analogous functions (such as the central complex in insects)” (117–18). Why is this the meta-consensus? Because the claims all have some evidence to support them, and he thinks everyone engaged in current research should agree that there are challenges to views that don’t take these brain regions as important for sentience. What about positions that might see such brain regions as involved, but not necessary for sentience? Less popular theories, such as integrated information theory or biopsychism, fall into the zone of reasonable disagreement, and are appropriately discussed by philosophers and scientists. But they are not going to be helpful when making policy decisions, Birch tells us (though he thinks advocates of such theories can still sign on to his meta-consensus). What he takes to be the shared consensus is supported by enough evidence that the views should play a role in determining what are the “sentence candidates” (systems that have a realistic possibility of being sentient given existing evidence, and where welfare considerations are in play) or “investigative priorities” (systems with lower evidential basis than sentience candidate, but where further investigation may lead to being promoted to sentience candidate). Some sentience candidates include chimpanzees and bees, twelve-week-old human fetuses, and neural organoids with a functioning brainstem. Some investigative priorities include insect larvae, spiders, snails, worms, and LLMs.

Birch expects the meta-consensus to change as more evidence comes to light. How it might change isn’t really discussed. If you already wonder what counts as enough evidence, and why particular neural structures are emphasized while remaining open to multiple realizability of function, you might also worry about the type of evidence that may lead to change in the meta-consensus. Anthropomorphic evidence leads us to focus on brain structures—primate brains support consciousness because we take human consciousness to be the gold standard. The analogy to human brains grows weaker as move away from primates and mammals, yet the behavioral analogies may remain just as strong. When does the evidential basis change from morphological similarity to functional similarity? Should we privilege brain structures over functions? Over behavior? If we include behavior, as Birch does when he gets to discussions of some invertebrate animals, then the range of sentience candidates would become very large.

In the last section of the book, we get three chapters focusing on sentience in AI. Birch argues that while we should worry about the possibility of AI sentience, we should worry more about people being fooled into thinking AI is sentient beyond the evidence, so we should develop policy that gets ahead of the current technology. LLMs, he claims, already qualify as investigative priorities, and the way to investigate them is not by looking at behavioral markers or functions, because, as he and I argued in a coauthored piece (Andrews and Birch Reference Andrews and Birch2023), LLMs are designed to game these markers. Rather, we need to look for deep computational markers taken from current computational theories of consciousness. Unfortunately, as Birch has affirmed in his earlier work (Birch Reference Birch2022), we don’t yet have secure theories of consciousness that can serve as consciousness markers. Nonetheless, Birch proposes that we should be looking for self-emerging global workspaces or perceptual monitoring systems in LLMs, and if we find them, the system should be upgraded to a sentience candidate—a system that has a realistic possibility of being conscious.

In the chapters on AIs and animals, the neurobiological meta-consensus becomes less relevant, as it doesn’t limit the range of investigative priorities. When Birch turns toward the edge of the edge of sentience, and considers insect larvae and nematode worms, systems without the most minimal brain structures presented in the meta-consensus, here too he sees the possibility of sentience. Not because he expects to find the brain structures of the meta-consensus, but because we can investigate the systems’ behavior. He excludes other living organisms as even investigative priorities, including plants and unicellular organisms, because he thinks there is no behavioral evidence, but also because they lack brain mechanisms analogous to the vertebrate midbrain. Given the discussion of the worms, behavior is doing the heavy lifting, leading me to wonder whether the meta-consensus would be more productively presented in behavioral terms than in neurobiological ones.

Who should read this book? I take it the readership is meant to be quite broad—including policy makers and the public as well as professional philosophers and scientists. Anyone can download the book for free, as it is published open access. Bioethicists can search for the discussions of disorders of consciousness and abortion and consider the proposal that fetal pain needs to be communicated to people making decisions about their own fetuses. Animal rights activists can look for positive proposals about veganism and fail to find any. Political philosophers can evaluate the efficacy of the citizen panel structure. For those first approaching the fields of consciousness, AI safety, or animal pain, they will find an introduction to the research landscape. This is a useful book. It supports important conversations about how to decide who are the most vulnerable among us, and how to think about our obligations to them. It recognizes conflicts of values, and heroically tries to find a way to bring people together, working through our disagreements in a philosophically rich, evidence-based fashion. There is much here to debate, but debate is the goal, and with this book Birch is egging us on to take these conversations public.