Models of Consciousness 7, University of Copenhagen -- 16/10/26

On the Natural Origins and Function of Self-Consciousness: Lessons from Studies of Animal Tool Use

Many Kantian and post-Kantian philosophers hold that consciousness requires self-consciousness, standardly understood as a conceptual capacity to think about oneself as oneself. Against this, some argue that non-human animals are plausibly conscious despite possessing no SELF concept. To maintain the Kantian idea, discussion has shifted to an ‘implicit’, nonconceptual awareness of self, sometimes termed ‘pre-reflective self-consciousness’. But this notion is not easy to characterise, nor to connect to the information-processing views of mind that dominate cognitive science. Drawing on studies of animal tool use, I argue that some birds’ innovative construction of non-species-typical tools to solve novel problems is best explained by positing self-conscious representational states. I assume that self-conscious states are self-representational states. And I argue that their contents represent the self by tracking identity of body—represented as locus of perception and action—across two kinds of mental state required for explaining the tool use. States tied to occurrent perceptions and desires in the present moment must be combined with states representing merely possible states of affairs. Self-representation makes possible this combination, I argue. My argument suggests that a limited kind of self-consciousness might arise in nature to allow animals to solve novel practical problems requiring thought about a limited future. My work elucidates a possible meaning for pre-reflective self-consciousness, grounded in scientific studies of animal behaviour, and with a ready connection to representationalist, information-processing accounts of mind. Moreover, it suggests a scientifically respectable way to maintain the core Kantian idea that consciousness requires self-consciousness.

The Philosophy of Brain-Computer Interfaces, UMass Boston -- 6/6/26

Spooky (Mental) Action at a Distance

Active Brain-Computer Interfaces (BCIs) allow users to intentionally control some mechanical devices in thought, e.g., robotic arms. Some think that BCI-mediated actions should force us to rethink action theory. BCI-mediated actions, they say, are ‘disembodied’, and so form a new and novel category of action. This attitude seems to be grounded in the impressiveness of how BCIs enable thinks to ‘directly’ influence the world outside of their own body. On the other hand, one might wonder what principled difference there is between moving a robotic arm with a BCI and moving the same arm with a handheld controller (say). Why is a BCI-mediated action an entirely different kind of action from any quotidian instance of tool use? Clearly, one holding the revisionist view will appeal to the ‘directness’ of control. In this talk I will argue that, on the most widely-accepted philosophical view of agency (the ‘causal theory’ or ‘standard view’), BCI-mediated actions just are ordinary instances of tool use, contra Steinert et al. (2019). This highlights a clash between out intuitions about the specialness of phenomena like BCI-mediated action and the dominant theoretical frameworks for understanding mind and action. I will end with some remarks about how we might navigate this clash.

UMass Fellow's Talk -- 30/11/25

Self-Consciosuness in Animal and Artificial Agents

In previous work I argued from empirical work that the innovative tool use of Goffin’s cockatoos likely requires a form of self-representation. Explaining their ability to use novel tools requires that they combine in thought representations tied to the here-and-now with representations of potential ways the world will be. I argue that nonconceptual self-representational content, indexing both kinds of representation to ‘this body’, would explain this combination. Thus, I conclude, some animals are likely self-conscious in a minimal way. From another perspective, the cockatoos are learning how to solve a problem by exploring their environment. Cognitive scientists apply the formalism of reinforcement learning from machine learning/artificial intelligence research to cases of animal learning. One such application investigate a group of neural circuits called the dopamine reward system. This system is implicated in animal learning, and its firing patterns are well described by a particular reinforcement learning algorithm, the temporal difference (TD) algorithm. This algorithm functions by combining information indexed to successive time points. This bears a functional similarity to my posited role for self-representational content. So, the question arises: would an artificial system that solves learning problems using the TD algorithm likewise minimally self-conscious? I claim they are not, for two reasons. First, the neuroscientific claim that the dopamine reward system implements the TD algorithm is at a lower level of analysis than my claim about the cockatoos’ thoughts. Without an account of how neurally implemented computations relate to psychological states with intentional content, it is unclear what inferences can be made between these levels. Second, the problem the cockatoos solve, as I present it, is essentially connected to their embodied capacities for action, whereas the problem the TD algorithm solves is not essentially connected to embodied capacities for action. The dopamine reward system may well be an information processing sub-system essential for embodied learning. But a full explanation of the information processing must refer to motor control systems, too.