AAP 2026 CONFERENCE

Recent developments in AI, particularly in affective computing, have brought renewed attention to the question of first‑person authority—the authority subjects ordinarily take themselves to have with respect to their own avowals about their mental states. Emotion Recognition Systems, in particular, are often presented as being able to infer what you are feeling, in some cases better than you do. At the same time, recent work on self-knowledge emphasises that much of our self-knowledge is inferential, technologically mediated, and fallible. Together, these developments give rise to a tension: if AI systems can accurately infer our mental states, might they become accurate enough to override our avowals and thereby undermine our authority?
I argue that the question rests on a mistaken but common assumption that first-person authority is grounded in a form of epistemic reliability or superiority that subjects enjoy over their mental states, an assumption often inherited from accounts of self-knowledge. Instead, I develop a capacities-based, non-epistemic, account of first-person authority. On this view, first-person authority is grounded in a subject’s distinctive set of capacities to relate to their mental states—through avowal, endorsement, and related capacities—in ways that are not available to others. Although first-person avowals are often accurate as a matter of contingent fact, that is not what makes them authoritative. The upshot is that AI does not pose a threat to first-person authority simpliciter; rather, it helps reveal the types of self-avowals that are vulnerable to challenge.

When existential and religious choices are made under uncertainty, complexity and entanglement are likely to follow (De Cruz, 2021, p. 2). This is especially true at the intersection of generative AI (GAI), philosophy, neuroscience, and theology. If, as some argue, AI has become a new kind of entity — an 'autosapiens' that is adaptive (it learns), amiable (it befriends), and arcane (it mystifies) — then the question of how we encounter and respond to it becomes urgent (Heimans & Timms, 2024). This commentary uses the concept of kairos — with its classical and theological resonances — to argue that philosophy, neuroscience, and theology not only share common ground in addressing this challenge, but that their genuine dialogue may yield important breakthroughs in understanding what it means to be human in an age of intelligent machines.
At the convergence of philosophy, neuroscience, and theology, this paper argues that the rise of generative AI constitutes a self-renewing kairos that calls for urgent interdisciplinary dialogue. As AI systems such as Claude increasingly occupy neurological and meaning-making roles once associated with gods and sages, sentience—the felt, conscious awareness that Claude lacks but humans possess—emerges as the decisive hinge of the entire human–AI encounter.

Much of the recent academic debate in the philosophy of AI revolves around a deceptively simple question: are AI tools only as good as their datasets? Existing discussions, however, tend to focus on improving the quality or quantity of training data, thereby underestimating a more subtle issue: how do AI agents handle outlier cases? This paper examines the ‘majoritarian drift’ – the tendency of AI agents to privilege positions with the largest training footprints, systematically marginalising underrepresented perspectives and edge cases. I approach this problem through the lens of exceptionalism, understood as a condition in which a phenomenon is sufficiently unusual to demand treatment outside standard frameworks. The paper proceeds in two parts. First, I demonstrate that majoritarian drift in ethical reasoning disproportionately favours utilitarian and aggregative approaches, while disadvantaging particularist, casuistic, and minority-tradition ethics. Second, I analyse analogous distortions in logical reasoning, focusing on how majoritarian drift impairs AI judgment in cases involving exceptions to general rules, with particular attention to legal clauses. The paper argues that exceptionalism reveals a structural limitation in how current AI architectures process normative and logical complexity.

This paper examines the role of fine arts in shaping temporal consciousness within a technology-mediated environment, particularly amid the rise of AI-generated art. Insights from Watsuji Tetsurô and Imamichi Tomonobu have already presented a reaction to the significance of temporality and aesthetic experience in their discussions on technology and the ethics of post-structural alterity. However, there remains a potential discussion in systematically articulating how artistic expression uniquely cultivates the virtue of temporal attentiveness in contrast to the technological production of AI art. This paper engages that discussion by analyzing the fine arts as bearers of temporal awareness in the contemporary context of AI art. From such discussion, there is a close examination of the concepts of artistic creation, expression, and ephemerality—further arguing that aesthetic experience enables an awareness of the transient, thereby restoring a sense of temporality diminished by technological abstraction. Such findings suggest that, unlike AI-generated art, human artistic activity embodies a lived temporal process essential to one’s aesthetic formation. The study brings further insight that preserving the expressive and temporal dimensions of art is crucial for sustaining an orientation wherein a critical framework is offered for evaluating the limitations and implications of AI art in contemporary society.

Philosophers endorsing Computational Functionalism (CF) have argued we should afford AI systems moral consideration in virtue of their possessing (or possibly possessing) conscious states. I argue we have no good reason to think CF is true, and good reason to think it isn’t. I distinguish three versions of Computational Functionalism and give arguments against each. Identity CF says the property of a physical system implementing the right computation is identical to the property of physical system being in a conscious state. I show that on plausible assumptions about computational implementation we have a straightforward deductive argument against Identity CF. I then consider more popular versions of CF which say that the property of a physical system implementing the right computation is either ‘sufficient for’, or ‘necessary and sufficient for’ (rather than identical to), a physical system being in a conscious state. Responding to Chalmer’s dancing and fading qualia arguments, I argue we have no good reason to think that a physical system could be conscious in virtue of implementing a computation, and the idea ought to strike us as a bizarre and implausible.

It has recently been suggested that large language models (LLMs) fine-tuned on the corpus of text from palliative care patients could be used to alleviate their distress by completing projects or relationships that would otherwise be cut short by their deaths. For example, a fine-tuned LLM could be used to complete the novel of a dying author. I contest the alleged benefits of this technology by drawing on the philosophy of Martin Heidegger. Heidegger claims that our anticipation of death is significant because of its ability to induce anxiety, which he characterises as a collapse in the meaningfulness of our self-interpretations. This experience is valuable because it enables us to live authentically, that is to say, in a way that understands that we are not necessarily defined by any of our meaning-giving self-interpretations. I argue that fine-tuned LLMs would disarm death of anxiety and the benefits of authenticity, above all the ability to live with greater flexibility and openness to the present. After considering the potential benefits that fine-tuned LLMs may nonetheless bring to palliative care, I conclude that they should not replace the work of human therapists capable of guiding the dying through these intense existential feelings. 

Talk of “agentic AI” can illuminate real changes in technical delegation, but it can also move agency-talk toward artificial systems while human and institutional actors recede from view. This paper argues that responsible AI governance requires neither machine personhood nor metaphysical quietism, but fitting individuation: naming AI systems enough to govern, contest, authorize, and repair their uses without personifying them beyond warrant. On this account, AI systems are not autonomous moral agents, but socio-technical deployments through which judgment, authority, risk, and responsibility flow. The central question is therefore not simply “Is the AI an agent?” but “Where must judgment, contestability, and answerability be located for this deployment to remain governable?” I propose three adequacy tests: identity credentials sufficient for governance, delegation-with-answerability, and responsibility-flow mapping. These tests distinguish legitimate technical delegation from agency laundering, explanation theatre, and nominal human oversight. The result is a modest metaphysical account of AI: thick enough to locate responsibility, light enough to avoid machine mystification, and practical enough to guide institutional governance.

Many think that we should respect humans and not AIs. This paper shows that this approach runs into trouble in “ambiguous spaces,” where we can’t tell whether someone is an AI. We can either extend respect to ambiguous agents, or withhold respect from them. Either approach comes with significant costs. We call this dilemma the contagion of disrespect. Extending respect ties our hands, and incentivizes people to deploy ambiguous AIs against us. Withholding respect risks blocking some humans from respect, and risks creating spirals of disrespect.

The use of Generative Artificial Intelligence (GenAI) by children demands critical examination over whether and how the technology affects their cognitive development. Given the growing empirical research showing the impacts of GenAI on human cognition, this paper aims to philosophically examine the threats that this technology may pose to children’s development as practical reasoners, given the importance of childhood and adolescence in developing this central capability. This paper first outlines and justifies the use of the Capabilities Approach and the particular focus on the central capability of practical reason. I then explore empirical research which shows that GenAI may put downward pressure on children’s ability to reach the necessary threshold for practical reason. I then argue that, due to the pressures it places on this capability, if children are unable to reach the necessary threshold for practical reason due to cognitive offloading and delegation to GenAI, they may be at risk of becoming dependent on GenAI tools and the corporations that control them. Such dependency would raise two important critiques of the adoption of GenAI in childhood and education: first, prudential critiques where the agent’s own interests are undermined; second, political critiques where unjust social forces are reinforced and exacerbated.

This talk investigates the following two questions: Q1. Under what conditions do human AI-based beliefs qualify as knowledge? Q2. Do the seemingly crazy errors that AI systems sometimes make pose a threat to human AI-based beliefs qualifying as knowledge? The discussion of Q1 and Q2 is set against the background of a stock of examples of AI errors, including adversarial examples drawn from the large literature on image classifiers and LLMs. Many of these errors strike humans as bizarre or crazy—e.g., LLMs ‘hallucinating’ references or an image classifier correctly classifying an image of a panda but switching the output to ‘gibbon’ after the original image is subjected to a humanly imperceptible manipulation of its pixel structure. The talk brings Q1 and Q2 into connection with mainstream epistemology—more specifically, modal epistemology. The key idea is that, in order for a belief output of a given method to qualify as knowledge in a given world w, the belief must not only be true in w; it must likewise be sufficiently modally robust. The talk discusses the prospects of AI-based knowledge, given modal conditions on knowledge and the wealth of adversarial examples that have surfaced in AI research.

According to scientific realists, the success of a scientific theory provides strong evidence that it is (approximately) true (Putnam, 1975). In response, antirealists argue that the theories we have are successful because they are survivors of a selection process where unsuccessful theories are rejected, so truth is not necessary to explain success (van Fraassen 1980). This paper argues that the training and testing process of artificial intelligence is structurally analogous to the selection process of scientific theories. Convolutional Neural Networks (CNN) achieve human-level performance in image classification through iterative training procedures that adjust weights and biases to minimise errors. 
Moreover, recent techniques in explainable AI (XAI) can approximate concept-level interpretations of the CNN’s structure. Some of these concepts align with human concepts, while others do not, even when predictive performance is comparable. The CNN is interpreted as encoding a structural representation of the data, analogous to how a scientific theory represents phenomena. To the extent that the AI classifier uses similar concepts to humans, we have support for realist interpretations of successful representation. Conversely, divergence from human concepts lends weight to antirealist interpretations.

A fast-moving debate has emerged over whether LLMs are bullshitters in any significant sense. This talk develops an account of LLM bullshit that, in contrast to the most influential existing accounts, is entirely output-based.
I begin with an overview of the best-known treatment of LLM bullshit, due to Hicks, Humphries, and Slater, along with some of the main critical reactions to their views. One response, from Gunkel and Coghlan, argues that Hicks et al.’s process-based account should be replaced by an output-based one. I take this response to be compelling, though it is notable that Gunkel and Coghlan do not attempt to develop a detailed output-based account.
To fill this gap, I review Florian Cova’s recent output-based account of bullshit, explain how it can be streamlined, and show how it can be applied to LLMs. The main upshots are: (i) some but not all LLM outputs are bullshit; (ii) LLMs engage in the activity of bullshitting sometimes but not always; and (iii) LLMs are bullshitters in only a rather weak sense.

The de facto situation regarding trustworthy AI is that the principles and supporting guidelines of are largely settled, from a pan-cultural perspective, and that if we build this trustworthy AI—all other things being equal—this will lead to greater AI adoption. 

There are some consequences that may be drawn by AI accelerationists from this. First, we don’t need to expend resources engaging with the communities impacted by AI to determine what makes AI trustworthy for them. Instead, it is a matter of building trustworthy AI and getting that AI in front of people to facilitate AI adoption. Second, on balance, this version of trustworthy AI constitutes a societal good: real trustworthy AI mitigates harms while delivering maximal benefits. Third, if we build trustworthy AI according to these assumptions, it is not rational for people to be sceptical of AI.

And, following from that, those who raise fears among the community regarding AI adoption are both doing a disservice to that community and are not acting in a rationally-justified manner.
In this paper I critique this common notion of trustworthy AI, discussing AI in the context of a plurality of world views, and critique the claims made above.

Critics rightfully identify that AI models are biased against marginalised groups. These biases deteriorate our shared hermeneutical resources—the narratives, frameworks, and concepts that structure how we understand the world and ourselves—by reflecting and exacerbating existing oppressive narratives. However, this is not the only way that AI models are sources of hermeneutical impoverishment. I propose that AI models warp our hermeneutical resources, not only by reinforcing existing problematic representations of identity groups, but by changing how these groups are represented. That is, AI models are conceptual engineers, capable of revising our social concepts.
When certain deep machine learning models perform predictions, they construct social concepts. Crucially, these algorithmic concepts differ from their human-constructed counterparts due to unavoidable trade-offs in model development. In constructing revised algorithmic concepts, AI models act as conceptual engineers. Once introduced, algorithmic concepts can take the place of our own concepts. Through these hermeneutical changes, AI models can also make a difference to our underlying social ontology: in redefining how we think of ourselves, they can redefine who we are. Finally, I offer upshots of attending to AI models as novel sources of epistemic and ontological harm.