Jonathan Birch

Rapid progress in artificial intelligence (AI) capabilities has drawn fresh attention to the prospect of consciousness in AI. There is an urgent need for rigorous methods to assess AI systems for consciousness, but significant uncertainty about relevant issues in consciousness science. We present a method for assessing AI systems for consciousness that involves exploring what follows from existing or future neuroscientific theories of consciousness. Indicators derived from such theories can be used to inform credences about whether particular AI systems are conscious. This method allows us to make meaningful progress because some influential theories of consciousness, notably including computational functionalist theories, have implications for AI that can be investigated empirically.

Motivational trade-off behaviours, where an organism behaves as if flexibly weighing up an opportunity for reward against a risk of injury, are often regarded as evidence that the organism has valenced experiences like pain. This type of evidence has been influential in shifting opinion regarding crabs and insects. Critics note that (i) the precise links between trade-offs and consciousness are not fully known; (ii) simple trade-offs are evinced by the nematode worm Caenorhabditis elegans, mediated by a mechanism plausibly too simple to support conscious experience; (iii) pain can sometimes interfere with rather than support making trade-offs rationally. However, rather than undermining trade-off evidence in general, such cases show that the nature of the trade-off, and its underlying neural substrate, matter. We investigate precisely how.

Can octopuses feel pain and pleasure? What about crabs, shrimps, insects, or spiders? How do we tell whether a person unresponsive after severe brain injury might be suffering? When does a fetus in the womb start to have conscious experiences? Could there even be rudimentary feelings in miniature models of the human brain, grown from human stem cells? And what about AI? These are questions about the edge of sentience, and they are subject to enormous, disorienting uncertainty. The stakes are immense, and neglecting the risks can have terrible costs. We need to err on the side of caution, yet it's often far from clear what 'erring on the side of caution' should mean in practice. When are we going too far? When are we not doing enough? The Edge of Sentience presents a comprehensive precautionary framework designed to help us reach ethically sound, evidence-based decisions despite our uncertainty.

Could emotions be a uniquely human phenomenon? One prominent theory in emotion science, Lisa Feldman Barrett’s Theory of Constructed Emotion (TCE), suggests they might be. The source of the sceptical challenge is that TCE links emotions to abstract concepts tracking socio-normative expectations, and other animals are unlikely to have such concepts. Barrett’s own response to the sceptical challenge is to relativize emotion to the perspective of an interpreter, but this is unpromising. A more promising response may be to amend the theory, dropping the commitment to the abstract nature of emotion concepts and allowing that, like olfactory concepts, they have disjunctive sensory groundings. Even if other animals were emotionless, this would not imply they lack morally significant interests. Unconceptualized valenced experiences are a sufficient basis for morally significant interests, and such experiences may occur even in the absence of discrete, constructed emotions.

Some patients, following brain injury, do not outwardly respond to spoken commands, yet show patterns of brain activity that indicate responsiveness. This is 'cognitive-motor dissociation' (CMD). Recent research has used machine learning to diagnose CMD from electroencephalogram recordings. These techniques have high false discovery rates, raising a serious problem of inductive risk. It is no solution to communicate the false discovery rates directly to the patient's family, because this information may confuse, alarm and mislead. Instead, we need a procedure for generating case-specific probabilistic assessments that can be communicated clearly. This article constructs a possible procedure with three key elements: (1) A shift from categorical 'responding or not' assessments to degrees of evidence; (2) The use of patient-centred priors to convert degrees of evidence to probabilistic assessments; and (3) The use of standardised probability yardsticks to convey those assessments as clearly as possible.

‘Sentience’ sometimes refers to the capacity for any type of subjective experience, and sometimes to the capacity to have subjective experiences with a positive or negative valence, such as pain or pleasure. We review recent controversies regarding sentience in fish and invertebrates and consider the deep methodological challenge posed by these cases. We then present two ways of responding to the challenge. In a policy-making context, precautionary thinking can help us treat animals appropriately despite continuing uncertainty about their sentience. In a scientific context, we can draw inspiration from the science of human consciousness to disentangle conscious and unconscious perception (especially vision) in animals. Developing better ways to disentangle conscious and unconscious affect is a key priority for future research.

When two agents engage in a joint action, such as rowing together, they exercise joint know-how. But what is the relationship between the joint know-how of the two agents and the know-how each agent possesses individually? I construct an “active mutual enablement” (AME) account of this relationship, according to which joint know-how arises when each agent knows how to predict, monitor, and make failure-averting adjustments in response to the behaviour of the other agent, while actively enabling the other to make such adjustments. I defend the AME account from three objections, and I then use this account as the platform for an examination of the reducibility (or otherwise) of joint know-how to joint propositional knowledge. A summative account of joint propositional knowledge is incompatible with the reduction of joint know-how to joint propositional knowledge, whereas a distributive account is not (although serious difficulties for any such reduction remain). I close by highlighting some open questions the AME account brings into view concerning the evolutionary origin and scaling up of joint know-how.

Hamilton introduced two conceptions of social fitness, which he called neighbour-modulated fitness and inclusive fitness. Although he regarded them as formally equivalent, a re-analysis of his own argument for their equivalence brings out two important assumptions on which it rests: weak additivity and actor's control. When weak additivity breaks down, neitherfi tness concept is appropriate in its original form. When actor's control breaks down, neighbour-modulated fitness may be appropriate, but inclusive fitness is not. Yet I argue that, despite its more limited domain of application, inclusive fitness provides a distinctively valuable perspective on social evolution.

Martha Nussbaum’s Justice for Animals calls upon humanity to secure for all sentient beings the central capabilities they need to flourish. This essay review critically examines the ethical and scientific foundations of Nussbaum’s position. On the ethical side, we explore the tension between a robust defence of animal rights and political liberalism, which requires tolerance of a range of reasonable views. On the scientific side, we reflect on how our uncertainty regarding the distribution of sentience in the natural world leaves us uncertain about how many claimants of justice there are and about the relative strength of their claims. We bring out some further problems with Nussbaum’s views on animal killing and pest control. We also reflect on how animal welfare science might be done differently if guided by Nussbaum’s framework.

Peter Carruthers has recently argued for a surprising conditional: if a global workspace theory of phenomenal consciousness is both correct and fully reductive, then there are no substantive facts to discover about phenomenal consciousness in non-human animals. I present two problems for this conditional. First, it rests on an odd double-standard about the ordinary concept of phenomenal consciousness: its intuitive non-gradability is taken to be unchallengeable by future scientific developments, whereas its intuitive determinacy is predicted to fall by the wayside. Second, it relies on dismissing, prematurely, the live empirical possibility that phenomenal consciousness may be linked to a core global broadcast mechanism that is (determinately) shared by a wide range of animals. Future developments in the science of consciousness may lead us to reconsider the non-gradability of phenomenal consciousness, but they are unlikely to lead us to accept that there are no facts to discover outside the paradigm case of a healthy adult human.

We introduce the virtual special issue on content in signalling systems. The issue explores the uses and limits of ideas from evolutionary game theory and information theory for explaining the content of biological signals. We explain the basic idea of the Lewis-Skyrms sender-receiver framework, and we highlight three key themes of the issue: (i) the challenge of accounting for deception, misinformation and false content, (ii) the relevance of partial or total common interest to the evolution of meaningful signals, and (iii) how the sender-receiver framework relates to teleosemantics.

Altruistic deception (or the telling of “white lies”) is common in humans. Does it also exist in non-human animals? On some definitions of deception, altruistic deception is impossible by definition, whereas others make it too easy by counting useful-but-ambiguous information as deceptive. I argue for a definition that makes altruistic deception possible in principle without trivializing it. On my proposal, deception requires the strategic exploitation of a receiver by a sender, where “exploitation” implies that the sender elicits a behaviour in the receiver that is beneficial in a different type of situation and is expressed only because the signal raises the probability, from the receiver's standpoint, of that type of situation. I then offer an example of a real signal that is deceptive in this sense, and yet potentially altruistic (and certainly cooperative): the purr call of the pied babbler. Fledglings associate purr calls with food, and adults exploit this learned association, in the absence of food, to lead fledglings away from predators following an alarm call. I conclude by considering why altruistic deception is apparently so rare in non-human animals.

Animal welfare scientists face an acute version of the problem of inductive risk, since they must choose whether or not to affirm attributions of mental states to animals in advisory contexts, knowing that their decisions hold significant consequences for animal welfare. In such contexts, the burden of proof should be sensitive to the moral consequences of error, but a framework for setting appropriate burdens of proof is lacking. Through reflection on two cases—the case of pain, and the case of cognitive enrichment—I arrive at a tentative general framework based on the principle of expected welfare maximization. I then discuss the limitations of this framework and the important questions it leaves open.

In debates about animal sentience, the precautionary principle is often invoked. The idea is that when the evidence of sentience is inconclusive, we should “give the animal the benefit of the doubt” or “err on the side of caution” in formulating animal protection legislation. Yet there remains confusion as to whether it is appropriate to apply the precautionary principle in this context, and, if so, what “applying the precautionary principle” means in practice regarding the burden of proof for animal sentience. Here I construct a version of the precautionary principle tailored to the question of animal sentience together with a practical framework for implementing it. I explain and defend the key features of this framework, argue that it is well-aligned current practice in animal welfare science, and consider and reject a number of influential counterarguments to the use of precautionary reasoning in this area.

Various results show the ‘formal equivalence’ of kin and group selectionist methodologies, but this does not preclude there being a real and useful distinction between kin and group selection processes. I distinguish individual and population-centred approaches to drawing such a distinction, and I proceed to develop the latter. On the account I advance, the differences between kin and group selection are differences of degree in the structural properties of populations. A spatial metaphor (‘K-G space’) provides a useful framework for thinking about these differences: kin and group selection may be conceptualized as large, overlapping regions of K-G space. I then consider some implications of the account, defend it from possible objections, and further argue that the structural features characteristic of both kin and group selection may recur at multiple levels of biological organization.

The notion that natural selection is a process of fitness maximization gets a bad press in population genetics, yet in other areas of biology the view that organisms behave as if attempting to maximize their fitness remains widespread. Here I critically appraise the prospects for reconciliation. I first distinguish four varieties of fitness maximization. I then examine two recent developments that may appear to vindicate at least one of these varieties. The first is the ‘new’ interpretation of Fisher’s fundamental theorem of natural selection, on which the theorem is exactly true for any evolving population that satisfies some minimal assumptions. The second is the Formal Darwinism project, which forges links between gene frequency change and optimal strategy choice. In both cases, I argue that the results fail to establish a biologically significant maximization principle. I conclude that it may be a mistake to look for universal maximization principles justified by theory alone. A more promising approach may be to find maximization principles that apply conditionally and to show that the conditions were satisfied in the evolution of particular traits.

In an incendiary 2010 Nature article, M. A. Nowak, C. E. Tarnita and E. O. Wilson present a savage critique of the best known and most widely used framework for the study of social evolution, W. D. Hamilton’s theory of kin selection. Over a hundred biologists have since rallied to the theory’s defence, but Nowak et al. maintain that their arguments ‘stand unrefuted’. Here I consider the most contentious claim Nowak et al. defend: that Hamilton’s rule, the core explanatory principle of kin selection theory, ‘almost never holds’. I first distinguish two versions of Hamilton’s rule in contemporary theory: a special version (HRS) that requires restrictive assumptions, and a general version (HRG) that does not. I then show that Nowak et al. are most charitably construed as arguing that HRS almost never holds, while HRG buys its generality at the expense of explanatory power. While their arguments against HRS are fairly uncontroversial, their arguments against HRG are more contentious, yet these have been largely overlooked in the ensuing furore. I consider the arguments for and against the explanatory value of HRG, with a view to assessing what exactly is at stake in the debate. I suggest that the debate hinges on issues concerning the causal interpretability of regression coefficients, and concerning the explanatory function Hamilton’s rule is intended to serve.

From mitochondria to meerkats, the natural world is full of spectacular examples of social behaviour. In the early 1960s Bill Hamilton changed the way we think about how such behaviour evolves. He introduced three key innovations - now known as Hamilton's rule, kin selection, and inclusive fitness - which have been enormously influential, but which remain the subject of fierce controversy. Hamilton's pioneering work kick-started a research program now known as social evolution theory. This is a book about the philosophical foundations and future prospects of that program. Part I, "Foundations", is a careful exposition and defence of Hamilton's ideas, with a few modifications along the way. In Part II, "Extensions", Jonathan Birch shows how these ideas can be applied to phenomena including cooperation in micro-organisms, cooperation among the cells of a multicellular organism, and culturally evolved cooperation in the earliest human societies. Birch argues that real progress can be made in understanding microbial evolution, evolutionary transitions, and human evolution by viewing them through the lens of social evolution theory, provided the theory is interpreted with care and adapted where necessary. The Philosophy of Social Evolution places social evolution theory on a firm philosophical footing and sets out exciting new directions for further work.

Cooperation is rife in the microbial world, yet our best current theories of the evolution of cooperation were developed with multicellular animals in mind. Hamilton’s theory of inclusive fitness is an important case in point: applying the theory in a microbial setting is far from straightforward, as social evolution in microbes has a number of distinctive features that the theory was never intended to capture. In this article, I focus on the conceptual challenges posed by the project of extending Hamilton’s theory to accommodate the effects of gene mobility. I begin by outlining the basics of the theory of inclusive fitness, emphasizing the role that the concept of relatedness is intended to play. I then provide a brief history of this concept, showing how, over the past fifty years, it has departed from the intuitive notion of genealogical kinship to encompass a range of generalized measures of genetic similarity. I proceed to argue that gene mobility forces a further revision of the concept. The reason in short is that, when the genes implicated in producing social behaviour are mobile, we cannot talk of an organism’s genotype simpliciter; we can talk only of an organism’s genotype at a particular stage in its life cycle. We must therefore ask: with respect to which stage(s) in the life cycle should relatedness be evaluated? For instance: is it genetic similarity at the time of social interaction that matters to the evolution of social behaviour, or is it genetic similarity at the time of reproduction? I argue that, strictly speaking, it is neither of these: what really matters to the evolution of social behaviour is diachronic genetic similarity between the producers of fitness benefits at the time they produce them and the recipients of those benefits at the end of their life-cycle. I close by discussing the implications of this result. The main payoff is that it makes room for a possible new mechanism for the evolution of altruism in microbes that does not require correlated interaction among bearers of the genes for altruism. The importance of this mechanism in nature remains an open empirical question.

We face two urgent challenges concerning consciousness and AI. Challenge One is that millions of users will soon misattribute human-like consciousness to AI friends, partners, and assistants on the basis of mimicry and role-play, and we don’t know how to prevent this. Challenge Two is that profoundly alien forms of consciousness might genuinely be achieved in AI, but our theoretical understanding of consciousness is too immature to provide confident answers one way or the other. Centrism about AI consciousness is the position that we must take both challenges seriously. The two challenges interact in ways that make this difficult. Steps to address Challenge One might undermine attempts to address Challenge Two by portraying the idea of conscious AI as impossible or inherently unlikely. Conversely, attempts to address Challenge Two might lead to higher levels of misattribution from ordinary users. This “manifesto” attempts to construct mutually consistent strategies for addressing both challenges.

In this report, we argue that there is a realistic possibility that some AI systems will be conscious and/or robustly agentic in the near future. That means that the prospect of AI welfare and moral patienthood — of AI systems with their own interests and moral significance — is no longer an issue only for sci-fi or the distant future. It is an issue for the near future, and AI companies and other actors have a responsibility to start taking it seriously. We also recommend three early steps that AI companies and other actors can take: They can (1) acknowledge that AI welfare is an important and difficult issue (and ensure that language model outputs do the same), (2) start assessing AI systems for evidence of consciousness and robust agency, and (3) prepare policies and procedures for treating AI systems with an appropriate level of moral concern. To be clear, our argument in this report is not that AI systems definitely are — or will be — conscious, robustly agentic, or otherwise morally significant. Instead, our argument is that there is substantial uncertainty about these possibilities, and so we need to improve our understanding of AI welfare and our ability to make wise decisions about this issue. Otherwise there is a significant risk that we will mishandle decisions about AI welfare, mistakenly harming AI systems that matter morally and/or mistakenly caring for AI systems that do not.