The Eigenvector Trap: Embodied Intelligence Beyond Disembodied Optimization

Inês Hipólito (Macquarie University, Humboldt University, Berlin, King's College London)

Abstract:

Cognitive science has systematically employed disembodiment as its methodological foundation, reaching its apex in modern AI systems. These Helmholtzian predictive machines extract data from context, learn statistical patterns, and generate predictions, but fall into an "Optimisation Trap" where accuracy trades off against adaptability. Unlike humans who learn through embodied interaction, AI systems rigidly classify inputs into known categories, exhibiting brittleness outside training distributions. Current approaches—more data, larger models, reinforcement learning—remain trapped in the same paradigm. This perpetuates a narrow conception of intelligence that privileges prediction over adaptation and serves commercial interests whilst failing to address fundamental limitations.

Bio:

Inês Hipólito is a lecturer of Philosophy of Artificial Intelligence at Macquarie University. Her research focuses on E-Cognition and how it can help understand and design augmented forms of cognition (e.g. brain-computer interfaces, neurotechnologies and smart environments). Inês was awarded the prestigious Talent Grant by the University of Amsterdam in 2021. She is also a co-PI at the consortium project “Neurourbanism as a novel approach to mental health” funded by the Berlin University Alliance. She also serves as a co-founder and vice-president of the International Society of the Philosophy of the Sciences of the Mind and an elected member of the Women in Philosophy Committee and the Committee in Diversity and Inclusivity at the Australasian Association of Philosophy. 

Zoom link available on request to Sean Bowden ([email protected])