Sampling from Internal and External Sensations Guides Decision-Making

Throughout our day, we continually make decisions to behave effectively in our environment. Imagine you want to cross a busy road. To ultimately decide when to cross, you accumulate sensory evidence about the vehicles on the road (e.g., “What is that object, where is it, and how is it moving?”). Only when sufficient evidence is accumulated to ensure safe passage is the decision to cross the road made (Fig. 1).

Figure 1.

Sampling from external and internal sensations while deciding when to cross a busy road. To ultimately decide whether it is safe to cross, one can sample from external (i.e., from perception) or internal (i.e., memory) sensations. For example, the features, position, motion, and speed of the blue car are gathered from visual perception (on the left). In contrast, these properties of the red car are sampled from working memory (on the right). External and internal sensations are sampled to gradually accumulate evidence until a decision threshold is reached, and the decision is made that it is safe to cross the road.

Sequential-sampling models state that decisions are made in an accumulation-to-bound fashion: evidence gradually accumulates until a decision threshold is reached (Ratcliff and McKoon, 2008). When this threshold is reached, a decision is formed (i.e., “I can safely cross now”) and behavior follows (i.e., crossing the road). Given that decision-making is omnipresent in everyday behavior, there is a strong motivation to identify neural markers that sensitively index evidence accumulation. In order to truly reflect evidence accumulation, such a neural decision variable must not correlate with other ongoing processes such as the preparation of motor responses or sensory processing (O’Connell et al., 2012).

Neural decision variables were first identified in nonhuman primates, where single-unit recordings in the lateral intraparietal cortex sensitively track evidence accumulation when judging the motion direction of a cloud of dots (Kiani et al., 2008; Kiani and Shadlen, 2009). In humans, scalp electroencephalography (EEG) potentials closely track evidence accumulation. Specifically, the centroparietal positivity (CPP) component observed over the parietal midline gradually emerges and peaks around response times (O’Connell et al., 2012; Kelly and O’Connell, 2013). The CPP is reportedly independent of sensory processing and motor preparation, allowing for the robust and isolated tracking of evidence accumulation.

There has been some debate about whether the CPP differs from the classical event-evoked P300 (or P3B) EEG component. While some suggest that the CPP reveals a more gradual process than the P300 (O’Connell et al., 2012), others claim that the components are identical (Twomey et al., 2015). Nonetheless, whether the CPP and P300 are equivalent or distinct, it is clear that parietal potentials in scalp EEG sensitively track evidence accumulation in the human brain.

Decision-making has predominantly been studied in situations where evidence is sampled from sensory information (external sensations). In one popular example alluded to above, participants judge whether a cloud of moving dots is predominantly moving left- or rightward. The strength of sensory information is manipulated by having relatively more or fewer dots move in the same direction (i.e., motion coherence; Kelly and O’Connell, 2013). In such studies, participants sample external sensations (i.e., motion), and the evidence accumulation process can be tracked using the CPP component of the EEG recording.

Although sampling external sensations is invaluable to the guidance of behavior, this may not tell the whole story. Recent advances have highlighted the role of working memory in guiding behavior (Heuer et al., 2020). Returning to our example of crossing the road (Fig. 1), deciding when to cross may be based not only on information about the blue car that is in view but also on the properties of the red car that is held in memory. From this memory-for-action perspective, it might be possible to sample information from internal sensations (i.e., working memory) to form decisions that lead to meaningful human behavior. Until recently, however, neural evidence for such a proposal has remained elusive.

In a recent article in The Journal of Neuroscience, van Ede and Nobre (2024) used EEG to investigate whether the CPP reflects the sampling of internal sensations. To this end, healthy human participants memorized the orientations of two bars that differed in color. After a brief delay, a colored probe indicated which memorized bar should be reported. Participants then recreated the orientation of the indicated bar. In this experiment, participants could not rely on external sensations but instead relied on internal sensations in working memory to complete the task. In contrast to previous work wherein memory contents were compared with a displayed test stimulus, here the orientation of one bar was probed only by presenting its color. This allowed for a relatively pure assessment of internal sampling that is uncontaminated with external processing.

Upon the onset of the colored probe, when evidence accumulation could begin, a reliable CPP component was observed. The amplitude of the CPP increased gradually and peaked just prior to the onset of the response. This suggests that the CPP reflected evidence accumulation rather than being evoked exclusively by the probe onset.

Taking this a step further, if the CPP indeed reflects evidence accumulation, it should be modulated by memory load. Specifically, evidence accumulation should take longer when sampling information from more internal sensations. This holds especially when considering that the strength of individual memory representations may decrease at higher memory loads (Bays and Husain, 2008; but see Luck and Vogel, 2013). A task similar to the previous one was used, but now in some trials, participants were cued to encode only one of the two colored bars. Participants reported orientations faster and more accurately when holding one bar in memory compared with when they had to remember two bars. The CPP was also modulated by load: upon the onset of the probe, CPP amplitude was greater and peaked later in the two-bar condition than in the one-bar condition. This indicated that more evidence needs to be accumulated when sampling from two instead of one item from working memory. Together, these data generally support the notion that information is sampled from internal sensations to guide decision-making, and this process can be tracked in humans using the CPP.

The results from van Ede and Nobre (2024) indicate that sampling not only from external (i.e., perception) but also from internal (i.e., working memory; Fig. 1) sensations can be sensitively tracked in the human brain. The role of the CPP has now been extended, as it may reflect gradual evidence accumulation when sampling both externally and internally (O’Connell et al., 2012; Kelly and O’Connell, 2013; van Ede and Nobre, 2024). This suggests that internal and external information sampling, at least in part, rely on similar neural mechanisms.

While the data from van Ede and Nobre (2024) indicate that the CPP reflects internal sampling, alternative interpretations have not been ruled out fully. For example, to determine whether the “internal-CPP” can be considered a decision variable that purely reflects evidence accumulation, the effects of motor preparation must be ruled out more directly. The authors do report some evidence that the observed effects are independent of motor preparation: the two-load conditions differed in CPP but had equal motor preparation demands. However, due to the task design, bar orientation was inherently linked with the response hand. Therefore, motor preparation occurred earlier in the one-load (during encoding) than in the two-load condition (upon the probe onset). This makes it difficult to compare the two conditions in this regard. An ideal internal sampling task should exclude the need of an overt response to fully rule out motor preparation (O’Connell et al., 2012), but this may be difficult to achieve.

Another caveat precludes one from definitively concluding that the CPP reflects internal sampling. As no decision-related variables were manipulated explicitly, the authors instead rely on previous studies investigating external sampling (i.e., reverse inference). Such studies systematically manipulated the strength of evidence to investigate decision-making directly and demonstrated the CPP to reflect evidence accumulation independent of potential alternative effects (O’Connell et al., 2012; Kelly and O’Connell, 2013). A similar approach should be applied to directly test whether the CPP indeed reflects internal sampling by systematically manipulating the strength of memory representations. An “internal-CPP” account would predict that less precise memory representations should require more extensive evidence accumulation (and vice versa). We note that the authors did take a first step toward such an approach by manipulating memory load, as increased memory load may decrease the strength of individual memory representations (Bays and Husain, 2008; but see Luck and Vogel, 2013). Nevertheless, to conclude that the CPP reflects internal sampling, a more systematic manipulation of memory strength is necessary.

Together, van Ede and Nobre (2024) report neural evidence for the sampling of internal sensations to guide decision-making to ultimately drive behavior. Although the exact nature and the extent of an “internal-CPP” should be elucidated in future work, this study provides an important first step in understanding how internal sensations may be sampled to shape decision-making and human behavior.