Evaluating Global Workspace Markers in Contemporary LLM Systems
Izak Tait, Benjamin Rode, Joshua Bensemann preprint
The authors operationalize Global Workspace Theory into six testable markers and evaluate whether large language models (GPT, Claude, Gemini, DeepSeek) instantiate workspace-like control structures. They find at most partial evidence for workspace dynamics at the base-model level, with stronger support when systems incorporate tool-calling and memory interfaces. Five ensemble architectures designed to satisfy the markers show substantially stronger marker satisfaction. The authors argue that systems satisfying workspace markers warrant precautionary treatment in welfare and governance contexts, not because workspace organization proves consciousness, but because it strengthens attributions of agency-relevant capacities and shifts evidential burdens regarding consciousness-relevant processing.