1Department of Engineering, University of Cambridge, Cambridge, UK
2Regional Primate Research Center, University of Washington, Seattle, USA
A decision is a commitment to a proposition or plan of action, based on evidence, prior knowledge and expected costs and benefits. In many instances, decisions are based on a stream of evidence in time. In this case, there are, in effect, two decisions: when to terminate the process and a choice. The link between choice and termination time is often apparent in the tradeoff between speed and accuracy. This statement masks a subtle but important distinction: Termination of the decision process is an internal event that must then give rise to an overt behaviour. This gap between decision- and reaction-time, termed ‘non-decision’ time, is poorly understood. We hypothesized that the motor system might exploit this gap in time by dissociating action initiation from action specification.
In the current study, we investigated whether humans can process sensory information available after a decision has been reached and prior to movement initiation, and how any post initiation processing (PIP) is incorporated into an ongoing action. Most choice-reaction time studies require subjects to make rapid, ballistic movements to indicate their choice. For example, in the direction-of-motion paradigm, monkeys and humans typically register their decisions by making saccadic eye movements to a left or right choice target. Saccadic eye movements are truly ballistic; once initiated, they are fully specified. Thus there is no opportunity to revise a decision. Here, human subjects were asked to judge the direction of moving random dots (left and rightward motion in which we control the coherence, and hence difficulty) and to reach to one of the two targets (left and right) based on their decision. The advantage of this setup was that it uses arm movements to indicate decisions which, unlike saccadic eye movements, can be modified during execution.
Crucially, in our experiment, on movement initiation the random dots were extinguished. We found that the visual information arriving between the decision and movement initiation could update the decision online and was reflected in altered hand trajectories. We examined the probability and timing of such updating of decisions as a function of both the motion coherence and whether the initial decision was correct or incorrect. We extend the theoretical framework developed for the initial decision, that of accumulation of noisy evidence to a decision ‘bound’ (drift diffusion model) to incorporate updating of the decision after an initial decision has been made.