Figure 6.

The probabilistic graphical model used to simulate an agent capable of attentional control during a focused-attention perceptual task. Hidden states at each level have a single factor with two levels, the perceptual state can be either ‘standard’ or ‘deviant’, the attentional state can be ‘focused’ or ‘distracted’, and the meta-awareness state can be ‘high’ or ‘low’. Higher-level states determine the likelihood precision at the level below. The agent is given the instruction to pay attention to a visual oddball stimulus; this is modelled as a preference in C⁽²⁾ to observe the ‘focused’ outcome at the second level. Action here is only defined on the second level, with two possible actions u⁽²⁾, ‘stay’ and ‘switch’, which condition the attentional transition matrix B⁽²⁾ as shown.