Memory for dangers past: threat contexts produce more consistent learning than do non-threatening contexts

In earlier work we showed that individuals learn the spatial regularities within contexts and use this knowledge to guide detection of threatening targets embedded in these contexts. While it is highly adaptive for humans to use contextual learning to detect threats, it is equally adaptive for individuals to flexibly readjust behaviour when contexts once associated with threatening stimuli begin to be associated with benign stimuli, and vice versa. Here, we presented face targets varying in salience (threatening or non-threatening) in new or old spatial configurations (contexts) and changed the target salience (threatening to non-threatening and vice versa) halfway through the experiment to examine if contextual learning changes with the change in target salience. Detection of threatening targets was faster in old than new configurations and this learning persisted even after the target changed to non-threatening. However, the same pattern was not seen when the targets changed from non-threatening to threatening. Overall, our findings show that threat detection is driven not only by stimulus properties as theorised traditionally but also by the learning of contexts in which threatening stimuli appear, highlighting the importance of top-down factors in threat detection. Further, learning of contexts associated with threatening targets is robust and speeds detection of non-threatening targets subsequently presented in the same context.