Can you provide more insights on how attention allocation in AI is modeled using the global workspace theory?