Southern California Epistemology Network (SCEN) Workshop

11:00 - 11:50 Milana Kostić (USC), Conditionalizing on Conditionals
11:50 - 12:40 Annalisa Coliva (UCI) & Alessio Tacca (Milan), Can We Trust AI?
A Philosophical Framework for Responsible Trust in Artificial Intelligence

12:40 - 1:30 lunch break (a light lunch will be served)
1:30 - 2:20 Dan Kornman (UCSB), An Explanatory Theory of Evidential Status
2:20 - 3:10 June Lee-Rogers (USC), Awareness Growth from an Objective Bayesian Eye
3:10 - 4:00 Sebastian Liu (UCSB), Modestly Akratic

Milana Kostić (USC), Conditionalizing on Conditionals
Modeling the way learning conditional information impacts credences of epistemic agents has represented a major challenge for Bayesian epistemologists. Proposals for modeling conditional update presented in the literature have mostly employed non-standard update rules, since it is widely believed that the standard Bayesian update rule (conditionalization) cannot be used for these purposes due to Lewis’s triviality results (Lewis, 1976). In this paper I argue, against a near-consensus in the literature, that conditionalizing on propositions expressed by conditional sentencesis, in fact, the most promising approach to modeling learning conditional information. My starting point is a well-known contextualist strategy for circumventing Lewis’s triviality results (Bacon, 2015, van Fraassen, 1976). However, the contextua ist theory on its own is too underconstrained to be used for the purposes of modeling learning conditional information. I show that supplementing the contextualist framework with a simple yet surprisingly powerful principle, the Factual Independence Thesis, suffices to generate correct predictions about the effects of learning conditional information. Lastly, I present a tenability result for the Factual Independence Thesis.

Annalisa Coliva (UCI) & Alessio Tacca (Università Vita Salute, San Raffaele), Can We Trust AI? Philosophical Framework for Responsible Trust in Artificial Intelligence
Discussions of AI trustworthiness are widespread yet marked by persistent conceptual confusion.This presentation, drawing on Coliva (2025, 2026) and Coliva & Tacca (ms.), proposes a rigorous philosophical framework by distinguishing three questions that the literature routinely conflates: whether trust in AI is conceptually legitimate, whether AI systems are trustworthy, and whether a user's trust in AI is responsible.
On the first question, we argue that trust in AI is philosophically well-motivated once trust is understood following Nguyen (2022) and Coliva (2025) - as a default stance of unquestioning reliance rather than a morally loaded interpersonal relation. On the second, we establish that AI trustworthiness is always task relative, context-dependent, and defeasible, identifying eight structural categories of epistemic defeater including hallucination, systematic bias, distributional shift, adversarial vulnerability, and miscalibration.

On the third question, we argue that responsible trust requires genuine defeater sensitivity at the task, domain, and system levels - going beyond mere outcome alignment. We further identify deskilling and automation bias as the two primary systemic threats to responsible trust, and conclude with implications for AI governance, institutional design, and education in epistemic vigilance.

Dan Kornman (UCSB), An Explanatory Theory of Evidential Status
I'll introduce and defend a new, explanatory theory of evidential status, E=X. I show how E=X avoids the shortcomings of other theories, including Williamson's theory that one's evidence consists of all and only the propositions that one knows (E=K). In a longer version of the paper, I show how E=X yields an account of undercutting defeat, how it sheds light on the Benacerraf-Field challenge for mathematical Platonism and the evolutionary arguments against moral realism, and how it interacts with memory and testimony and the question of whether false propositions can count as evidence.

June Lee-Rogers (USC), Awareness Growth from an Objective Bayesian Eye
Learning often involves something other than acquiring new evidence. Agents also undergo awareness growth when they become aware of possibilities they previously were not. Standard Bayesian norms defined over a fixed algebra of propositions are silent about such changes. The literature has largely concluded that rational credences in awareness growth are not as tightly constrained as in evidential learning. However, this pessimism is largely due to the subjectivist setting most of the literature assumes. I show that switching to an objective Bayesian framework allows us to better understand the debates surrounding awareness growth. First, I develop a formal structure that dissolves several confusions in the literature about what awareness growth is, allowing us to model it properly. Second, I extend the notion of an ur-prior to address awareness along with evidence, by defining it over a maximally fine-grained algebra. Third, in light of the ur-prior framework thus constructed, I examine the debates over the rational constraints governing awareness growth.

Sebastian Liu (UCSB), Modestly Akratic
In addition to having uncertainty about what the world is like, I have uncertainty about whether my degrees of belief are rational. I'm 3/4 confident that it will rain tomorrow. Does my evidence warrant this exact level of confidence? I'm not sure. In this sense I'm modest: uncertain whether my degrees of belief are rational. Such modesty seems compatible with perfect rationality. But there is a problem: typically, when I should be modest, I should also be epistemically akratic: having one degree of belief in some proposition, while expecting that the rational degree of belief to have in that proposition is something different. Yet having akratic attitudes seems paradigmatically irrational, for it seems incoherent to be akratic. My primary aim in this paper is to defend an old position, that epistemic akrasia can be rational, in a new way, by explaining why even if I should be epistemically akratic, typically when I comply with the
demands of rationality in these instances, it'll appear to me that I'm irrational. This is because even if I should have akratic combinations of attitudes, for principled reasons, for any particular combination of akratic attitudes, I'm required to be uncertain whether that combination is the combination I ought to have.