The bidirectional interactions between decisions and actions: a computational model of embodied choice

Recent evidence challenges the traditional serial (decide-then-act) view of decision making, showing that decision and action processes can run in parallel. Here, we present a formal model in which decisions and actions are considered and optimized jointly, and apply it to various dynamical choice tasks. A prominent aspect of this model is an analysis of how action processes influence decision processes and not only vice-versa as often proposed.

HFSP Young Investigator Grant holder Giovanni Pezzulo and colleagues
authored on Mon, 20 July 2015

Perceptual decisions have been often described as a process of accumulating sensory evidence until reaching a threshold level of certainty. This evidence accumulation model, however, neglects the contribution of action and motor processes to the choice that is made. Can the actions we make influence our perceptual choices? Recent studies that track the changing dynamics of actions during perceptual decisions are increasingly revealing that the action dynamics of our bodies causally influence our decision processes. This evidence is in agreement with embodied theories of cognition, which argue that action processes are not "peripheral" but part of and parcelled with central cognition.

Figure: Left - Computer mouse trajectories recorded during a decision-making task, when human participants were asked to press one of two buttons (located to the top-left or top-right) with a mouse to indicate their decision. The trajectory begins at a start position (bottom center) equidistant from both targets. The blue lines indicate the trajectories; the red line indicates the mean trajectory. Center and right - mouse trajectories simulated by various computational models that make different assumptions on the underlying cognitive processing and action dynamics.

Our paper compares a series of decision-making models, including traditional serial (decide-then-act) models, parallel (continuous flow) models in which the partial results of decision processes are fed to the action system, and embodied models in which not only decision and action proceed in parallel, but the latter can influence the former. The model illustrates how actions, encompassing behavioural strategies such as preparation and commitment, can bias decision-making processes in ways that optimize the ecological choices of animals behaving in natural environments. This work thus combines two disconnected research streams, decision-making and action control, and provides support for embodied theories of cognition.

More generally, this study contributes to shedding light on decision dynamics in ecologically valid contexts, such as those that animals face during foraging - a topic that is at the core of the HFSP Young Investigator Grant, Beyond simple choices: computational and neuronal mechanisms for complex spatial behaviors, awarded to Matthijs van der Meer, Giovanni Pezzulo and Caleb Kemere.


Embodied Choice: How action influences perceptual decision making. Nathan Lepora and Giovanni Pezzulo. (2015) PLOS Computational Biology 11(4): e1004110

Other References

The mechanics of embodiment: a dialog on embodiment and computational modeling. Pezzulo G, Barsalou LW, Cangelosi A, Fischer MH, McRae K and Spivey MJ (2011) Front. Psychology 2:5. doi: 10.3389/fpsyg.2011.00005.

PLOS Computational Biology article link

Pubmed link