Fig 1.
Schematic illustration of alternative ways to link decision and action systems: (A) serial models, (B) parallel models, and (C) embodied choice models.
Note that here decision, choice, and action are shown as separate and modular systems for illustrative purposes only.
Table 1.
Differences between serial and parallel models of decision-making.
Table 2.
Differences between parallel (continuous flow) and embodied choice models of decision-making.
Fig 2.
An action is made to either target T1 or target T2 to indicate the decision. This action has a trajectory beginning at a start position equidistant from both targets. The red line indicated the mean trajectory. Experiment: 10 trajectories randomly chosen from the data in ref. [16].
Fig 3.
Simulated 2AFC task with action dynamics (trajectory from Model 3).
For simplicity, movement along the trajectory is at constant velocity. The trajectory depends on the interaction between the decision and action systems, driven by information for the alternatives. An action focus is used to link the decision and action systems, which here drifts from target T2 to T1 with the accumulated information for each alternative.
Fig 4.
Accumulated information and motor response during the decision process.
The top plot shows the accumulated information z(t) plotted against time, with decision bounds also shown and the times of crossing (dashed lines). The bottom plots show the trajectory of the focus on the line between the two targets. Model 1 (action initiation after decision) and Model 2 (action initiation and changes of mind) begin when the information first crosses threshold, with Model 2 changing its mind. Model 3 (action preparation) and Model 4 (preparation and commitment) begin at time zero, with the focus moving continually throughout the decision process.
Fig 5.
Individual and average trajectories.
Models 1–4: 10 trajectories randomly taken from the analysis for Figs 3, 4. Trajectories were selected to complete at target 1 to the left. The decision bound b was chosen for Models 1–3 to have a mean response time of 1.5 sec; whereas (1 sec for model 4 to be in range for this model). The average trajectory (red plot) in each case is taken over 150 trajectories, and computed parallel to the x-axis.
Fig 6.
Trajectory dependence on noise.
The ‘normal’ average trajectories are taken from Fig. 5. A second average trajectory was computed with increased noise σ = 4 (red curve).
Fig 7.
Comparison of 4 decision making with action models and a Baseline Model without action.
Each model is characterized by a speed-accuracy curve obtained from simulating (100000 samples) over a range of decision boundaries, with mean decision error plotted against mean response time. The minimal time-to-target Ttarget for action completion is also shown (dotted line). Note that curves for Model 3 (green line) and Model 4 (red line) do not overlap, but span different response times.