Eleni Angelou

The paper critiques the Khalidi-Stich triggering model of innateness, arguing that it fails to account for paradigm cases of innate cognitive abilities, such as causal perception. This triggering model implicitly relies on a high degree of what I call trigger specificity, a requirement that the development of causal perception does not meet. In response to this shortcoming, I propose that trigger specificity should be understood as a spectrum rather than a binary property. This revision not only resolves the problem posed by causal perception but also helps articulate a more nuanced conception of innateness tailored specifically to the explanatory goals of cognitive science.

Artificial Intelligence (AI) safety is an interdisciplinary field undergoing transformation to become its own paradigm. By adopting a Kuhnian approach to this transformation, I seek to describe the features that make AI safety pre-paradigmatic along with the conditions that are required for the field to transition to the paradigmatic stage. I argue that the nature of the alignment problem–the key research question of the field–is such that it calls for its own paradigm as the research necessary to solve it goes beyond the typical desiderata in machine learning work. It incorporates a series of metascientific projects as well as breakthroughs in normative and applied ethics. The faster and better these projects are executed, the more likely that the field will come to resolve its puzzles, which will benefit society at large.