Author Information
Muhammad Waqas Halim, International Edtech Lab, CanadaAbstract
What becomes of cognitive labor unique to human beings such as critical judgment, skepticism, and embodied reasoning when multimodal AI systems replace them? We theorize this systematic displacement of human heuristics by multimodal agentic AI, which we call Synthetic Humanism. It is centered around proactive agents which operate through "Joint Embedding Spaces" and stimulate human sensory intent across text, image, and more importantly, logic. Drawing on Tversky and Kahneman (1974) and Gigerenzer (2007), we argue that autonomous agents have the tendency to dominate the conditions of human heuristic possibilities and they consequently produce epistemic decoupling, systemic collapse of authorial intent, and mechanized erasure of human heuristics. We extend the concept of Foucault's (1977) algorithmic panopticon, and propose the condition of inferential cross-contamination, which is central to multimodal Agentic AI. The idea is that problems fundamental to AI such as hallucination can potentially embed themselves into premises across modalities. This process blurs the sources of errors and multiplies their impact. Empirical results (N=24 interviews; 150 AI outputs) present strong indication of loss of ability to audit the source-logic after cross-modal delegation was introduced (64%). Also, 73% treated hallucinated premises in cross-modal interaction as logically coherent. To deal with this limitation, we introduce Heuristic Sovereignty: institutional design principles that conserve the uncomputable dimensions of human cognition such as judgment under uncertainty, human ability to question and doubt, and embodied intuition, which cannot be reduced to pattern recognition (Ishiguro, 2021). Finally, we provide recommendations for authentic investigation in an era of synthetic agency.
Comments
Powered by WP LinkPress