Bridging Consciousness and AI: Evaluating the Global Latent Workspace in Embodied Agents

Explore a 20-minute conference talk from Models of Consciousness Conferences that delves into the intersection of consciousness theories and artificial intelligence through the lens of the global latent workspace model. Learn how attention-based transformer networks have sparked renewed discussions about consciousness in artificial systems, particularly through the computational framework of global neuronal workspace theory (GNWT). Understand the innovative global latent workspace (GLW) model's architecture, which features a central latent representation connecting multiple specialized modules through transformer-like attention mechanisms. Examine practical applications through a case study of a virtual embodied reinforcement learning agent operating in the Obstacle Tower Challenge 3D environment, demonstrating how sensorimotor interactions may contribute to conscious-like experiences in artificial systems. Consider the philosophical implications of bridging GNWT with sensorimotor contingency theory and contemplate whether machine learning representations could potentially give rise to phenomenal spatial perception.