Consciousness in Artificial Intelligence: Insights from the Science of Consciousness

Whether current or near-term AI systems could be conscious is a topic of scientific interest and increasing public concern. This report argues for, and exemplifies, a rigorous and empirically grounded approach to AI consciousness: assessing existing AI systems in detail, in light of our best-supported neuroscientific theories of consciousness. We survey several prominent scientific theories of consciousness, including recurrent processing theory, global workspace theory, higherorder theories, predictive processing, and attention schema theory. From these theories we derive ”indicator properties” of consciousness, elucidated in computational terms that allow us to assess AI systems for these properties. We use these indicator properties to assess several recent AI systems, and we discuss how future systems might implement them. Our analysis suggests that no current AI systems are conscious, but also suggests that there are no obvious technical barriers to building AI systems which satisfy these indicators.1

Introduction. In the last decade, striking progress in artificial intelligence (AI) has revived interest in deep and long-standing questions about AI, including the question of whether AI systems could be conscious. This report is about what we take to be the best scientific evidence for and against consciousness in current and near-term AI systems.

Because consciousness is philosophically puzzling, difficult to define and difficult to study empirically, expert opinions about consciousness—in general, and regarding AI systems—are highly divergent. However, we believe that it is possible to make progress on the topic of AI consciousness despite this divergence. There are scientific theories of consciousness that enjoy significant empirical support and are compatible with a range of views about the metaphysics of consciousness. Although these theories are based largely on research on humans, they make claims about properties and functions associated with consciousness that are applicable to AI systems. We claim that using the tools these theories offer us is the best method currently available for assessing whether AI systems are likely to be conscious. In this report, we explain this method in detail, identify the tools offered by leading scientific theories and show how they can be used.

We are publishing this report in part because we take seriously the possibility that conscious AI systems could be built in the relatively near term—within the next few decades. Furthermore, whether or not conscious AI is a realistic prospect in the near term, the rise of large language model-based systems which are capable of imitating human conversation is likely to cause many people to believe that some AI systems are conscious. These prospects raise profound moral and social questions, for society as a whole, for those who interact with AI systems, and for the companies and individuals developing and deploying AI systems. Humanity will be better equipped to navigate these changes if we are better informed about the science of consciousness and its implications for AI. Our aim is to promote understanding of these topics by providing a mainstream, interdisciplinary perspective, which illustrates the degree to which questions about AI consciousness are scientifically tractable, and which may be a basis for future research.

Method. 1. Computational functionalism: Implementing computations of a certain kind is necessary and sufficient for consciousness, so it is possible in principle for non-organic artificial systems to be conscious.

2. Scientific theories: Neuroscientific research has made progress in characterising functions that are associated with, and may be necessary or sufficient for, consciousness; these are described by scientific theories of consciousness.

3. Theory-heavy approach: A particularly promising method for investigating whether AI systems are likely to be conscious is assessing whether they meet functional or architectural conditions drawn from scientific theories, as opposed to looking for theory-neutral behavioural signatures.

These ideas inform our investigation in different ways. We adopt computational functionalism as a working hypothesis because this assumption makes it relatively straightforward to draw infer- 1.2.1 Computational functionalism Computational functionalism about consciousness is a claim about the kinds of properties of systems with which consciousness is correlated. According to functionalism about consciousness, it is necessary and sufficient for a system to be conscious that it has a certain functional organisation: that is, that it can enter a certain range of states, which stand in certain causal relations to each other and to the environment. Computational functionalism is a version of functionalism that further claims that the relevant functional organisation is computational.3 Systems that perform computations process information by implementing algorithms; computational functionalism claims that it is sufficient for a state to be conscious that it plays a role of the right kind in the implementation of the right kind of algorithm. For a system to implement a particular algorithm is for it to have a set of features at a certain level of abstraction: specifically, a range of possible information-carrying states, and particular dispositions to make transitions between these states. The algorithm implemented by a system is an abstract specification of the transitions between states, including inputs and outputs, which it is disposed to make. For example, a pocket calculator implements a particular algorithm for arithmetic because it generates transitions from key-presses to results on screen by going through particular sequences of internal states.

An important upshot of computational functionalism, then, is that whether a system is conscious or not depends on features that are more abstract than the lowest-level details of its physical make-up. The material substrate of a system does not matter for consciousness except insofar as the substrate affects which algorithms the system can implement. This means that consciousness is, in principle, multiply realisable: it can exist in multiple substrates, not just in biological brains. That said, computational functionalism does not entail that any substrate can be used to construct a conscious system (Block 1996). As Michel and Lau (2021) put it, “Swiss cheese cannot implement the relevant computations.” We tentatively assume that computers as we know them are in principle capable of implementing algorithms sufficient for consciousness, but we do not claim that this is certain.

1.2.3 Theory-heavy approach In section 1.2.1 we adopted computational functionalism, the thesis that implementing certain computational processes is necessary and sufficient for consciousness, as a working hypothesis, and in section 1.2.2 we noted that there are scientific theories that aim to describe correlations between computational processes and consciousness. Combining these two points yields a promising method for investigating consciousness in AI systems: we can observe whether they use computational processes which are similar to those described in scientific theories of consciousness, and adjust our assessment accordingly.

Discussion. There are risks on both sides of the debate over AI consciousness: risks associated with underattributing consciousness (i.e. failing to recognize it in AI systems that have it) and risks associated with over-attributing consciousness (i.e. ascribing it to systems that are not really conscious). Just as in other cases of uncertain consciousness, such as other animals (Birch 2018) and people with disorders of consciousness (Peterson et al. 2015; Johnson 2022), we must consider both types of risk.

As many authors have noted, as we develop increasingly sophisticated AI systems we will face increasingly difficult questions about their moral status (Bryson 2010, Gunkel 2012, Schwitzgebel & Garza 2015, 2020, Metzinger 2021, Shulman & Bostrom 2021). Philosophers disagree about the exact relationship between being conscious and moral status, but it is very plausible that any entity which is capable of conscious suffering deserves moral consideration. If we can reduce conscious suffering, other things being equal, we ought to do so. This means that if we fail to recognise the consciousness of conscious AI systems we may risk causing or allowing morally significant harms.

An analogy with non-human animals helps to illustrate the issue. Humans mistreat farmed animals in very large numbers, motivated by powerful economic incentives. Whether or not this mistreatment depends on a failure to attribute consciousness to these animals, it illustrates the potential problem. In the case of AI, there is likely to be considerable resistance to attributions of consciousness, partly because developers of AI may have powerful economic incentives to downplay concerns about welfare. So if we build AI systems that are capable of conscious suffering, it is likely that we will only be able to prevent them from suffering on a large scale if this capacity is clearly recognised and communicated by researchers. However, given the uncertainties about consciousness mentioned above, we may create conscious AI systems long before we recognise we have done so.

An important point in this context is that being conscious is not the same as being capable of conscious suffering. It is at least conceptually possible that there could be conscious systems that have no valenced or affective conscious experiences—that is, no experiences that feel good or bad to them (Carruthers 2018, Barlassina & Hayward 2019). If it is only valenced experiences that have special moral significance, then the key question for establishing the moral status of AI systems is whether they are capable of such experiences (that is, whether they are sentient, as this term is sometimes used). We have not discussed neuroscientific or philosophical theories of valence or otherwise investigated the prospects of specifically valenced conscious experience in AI systems. Theories of valenced conscious experience are less mature than theories of visual experience, suggesting an important priority for future work. However, we suspect that many possible conscious systems which are also agents will have valenced experiences since agents must evaluate options in order to select actions.

In short, the risk of under-attributing consciousness should be taken seriously. Failing to recognise conscious AI systems as such could lead us to cause unwarranted suffering to many conscious beings.

There is also a significant chance that we could over-attribute consciousness to AI systems—indeed, this already seems to be happening—and there are also risks associated with errors of this kind. Most straightforwardly, we could wrongly prioritise the perceived interests of AI systems when our efforts would better be directed at improving the lives of humans and non-human animals.

Conclusion. 4.3 Recommendations Several authors have made important practical recommendations concerning the possibility of AI consciousness. These include: Bryson’s (2010) case that conscious AI should be avoided; Metzinger’s (2021) call for a moratorium on work that could lead to conscious AI; Graziano’s (2017) arguments in favour of conscious AI; Schwitzgebel and Garza’s (2020) argument that we should only build particular AI systems if we can either be confident that they will be conscious, or be confident that they will not be; and the detailed recommendations of Bostrom and Shulman (2022). We have not assessed the arguments for these recommendations and do not consider them further here.

However, we do recommend support for research on the science of consciousness and its application to AI (as recommended in the AMCS open letter on this subject; AMCS 2023), and the use of the theory-heavy method in assessing consciousness in AI. One way that we can make progress in learning which AI systems are likely to be conscious is by developing and testing scientific theories of consciousness; theoretical refinements within the existing paradigms are valuable, as well as attempts to test predictions of competing theories, as in the ongoing Cogitate adversarial collaboration (Melloni et al. 2021). A second is by undertaking careful empirical research to extend theories of consciousness to non-human animals. This will help us to establish a more general account of the correlates of consciousness, based on evidence from a wider range of cases (Andrews & Birch 2023). And a third is research that refines theories of consciousness specifically in the context of AI. Research of this kind may involve theorising about AI implementations of mecha-

Limitations. This report is not the last word on AI consciousness. Far from it: one of our major aims is to spur further research. Further progress in the neuroscience of consciousness will contribute to our understanding of the conditions for consciousness in AI, and we have suggested that research on consciousness in non-human animals may be particularly valuable. Here, we highlight research questions that are relevant for understanding AI consciousness in particular.

• Examine other plausible theories of consciousness, not considered in this report, and use them to derive further indicators of consciousness; • Refine or revise the indicators which we have derived from the theories considered here; • Conduct assessments of other AI systems, or investigate different ways in which the indicators could be implemented.

Furthermore, our approach could be extended by developing a formal evaluation procedure for consciousness that could be applied to AI systems, although it is questionable whether this would be justified at present, in the context of significant uncertainty about computational functionalism and about particular scientific theories.