Learning occurs throughout the lifespan and is often an error-ridden process. As new learners make errors, error-detection plays a key role in updating incorrect associations in memory (Luft, 2014). Error detection is the recognition that an action conflicts with what is true relative to internal or external criteria (Ohlsson, 1996; Postma, 2000). Detection of errant behavior can be either internally (i.e., self-monitoring) or externally (i.e., feedback) driven. External error detection via feedback is critical when learners are acquiring new information or are unable to monitor the accuracy of their own responses (McCandliss et al., 2002; Pashler et al., 2005). For example, when learners are acquiring new phonological contrasts (e.g., Japanese speakers learning an English /r/ - /l/ distinction; McCandliss et al., 2002) or in certain cases of cognitive and/or linguistic deficit such as aphasia, Alzheimer’s, and aging (Schreiber et al., 2011; Nitta et al., 2017; Mandal et al., 2020). Thus often, feedback is not only helpful, but critical to the process of learning.

Learning conditions can influence how feedback is detected, processed, and utilized to update memory. Feedback timing (immediate vs. delayed) is one such condition of relevance to the current work. Immediate feedback is hypothesized to recruit dopamine-dependent striatal circuits which code prediction errors and send reward signals to the anterior cingulate cortex (ACC) (Holroyd and Coles, 2002; Nieuwenhuis et al., 2004). When feedback provision is delayed, (≥ 3,500 ms) the fast-acting dopamine-mediated learning is disrupted and processing shifts to the medial temporal lobe (MTL) which supports binding information that is separated by time (Foerde and Shohamy, 2011; Peterburs et al., 2016; Arbel et al., 2017).

Event-related potentials (ERPs) collected using electroencephalography (EEG) have been leveraged to elucidate the differences in the processing of feedback during learning. The feedback-related negativity (FRN) is a frontocentral negativity that peaks 250-300 ms after the provision of feedback (Gehring et al., 1995; Miltner et al., 1997).¹ The FRN is hypothesized to capture immediate feedback processing within the fronto-striatal circuit (Holroyd and Coles, 2002; Nieuwenhuis et al., 2004). The amplitude of the FRN is sensitive to feedback valence (negative > positive) (Gehring et al., 1995; Miltner et al., 1997) and feedback timing (immediate > delayed). Reductions in the FRN amplitude in response to delayed feedback is consistent with the hypothesis that delays in feedback timing shift processing away from fronto-striatal circuits (Weinberg et al., 2012; Peterburs et al., 2016; Weismüller and Bellebaum, 2016; Arbel et al., 2017; Kim and Arbel, 2019).

The N170 (Bentin et al., 1996) has been used to evaluate the processing of delayed feedback. The N170 is larger for delayed relative to immediate feedback (Arbel et al., 2017; Kim and Arbel, 2019; Höltje and Mecklinger, 2020; Albrecht et al., 2023) and in the context of feedback-based tasks has been hypothesized to reflect activity in the medial temporal lobe (MTL) (Arbel et al., 2017; Kim and Arbel, 2019; Höltje and Mecklinger, 2018, 2020; Albrecht et al., 2023). This is supported by neuroimaging studies that find that the MTL activation is heighted by delayed feedback (Foerde and Shohamy, 2011; Lighthall et al., 2018) and double dissociations in which individuals with MTL damage have been found to learn from immediate but not delayed feedback with the opposite pattern observed in individuals with basal ganglia damage due to Parkinson’s Disease (Foerde et al., 2013). Outside of the context of feedback-based learning, the N170 has been hypothesized to reflect activity in the MTL (Grippo et al., 1996; Baker and Holroyd, 2013) as well as the adjacent fusiform gyrus (Iidaka et al., 2006; Rossion and Jacques, 2011; Gao et al., 2019). Grippo et al. (1996) associated the N170 with the MTL when they observed a reduction in the amplitude of the N170 with increasing memory load in patients with temporal lobe epilepsy. More recently, Baker and Holroyd (2013) used source localization algorithms to localize the N170 to the MTL during a spatial navigation task. Similar activation during the processing of complex objects is localized to the fusiform gyrus (Iidaka et al., 2006; Rossion and Jacques, 2011). The N170, however, is not restricted to the visual domain and has been found to be elicited and modulated by the timing of auditory feedback; further supporting the notion that in the context of feedback-based learning, the N170 reflects cognitive processes that are not specific to the visual domain (Kim and Arbel, 2019). Albrecht et al. (2023) suggests that it is possible there are two potentially overlapping components that reflect activity in the MTL and fusiform gyrus, respectively. Our decision to use the N170 to gain insight into processes that are potentially supported by the MTL is motivated by (1) research finding that delayed feedback is associated with greater MTL activation, (2) previous studies identifying the MTL as a generator of the N170, and (3) research indicating that the N170 can be elicited by auditory feedback, and thus, likely does not reflect visual processing in the fusiform gyrus when elicited by feedback. In the current study, a larger amplitude N170 in the delayed relative to immediate feedback condition will support the claim that the N170 may reflect processing in the MTL although this cannot be definitively stated in the absence of source localization data which is beyond the scope of the current study.

Aging is associated with changes in striatal and MTL functioning and may influence older adults’ ability to learn successfully from certain types of feedback. Characterizing how age-related changes in neural functioning influences learning in older adults is key to supporting learning across the lifespan and may be particularly useful in rehabilitation contexts. Rehabilitation services, such as speech-language therapy, are primarily provided to adults over the age of 60 (American Speech-Language Hearing Association, 2019), many of whom have experienced an acquired brain injury (e.g., stroke, traumatic brain injury, tumor resection). Research in rehabilitation continually aims to improve the effectiveness of interventions and retention of treatment gains which may be informed by understanding how age-related changes in neural functioning interact dynamically with acquired neurologic damage.

In older adults, memory decline has been found to be greater in declarative memory tasks that require recruitment of the MTL relative to non-declarative tasks that require the recruitment of the striatal circuit (Hoyer and Verhaeghen, 2006). This is consistent with findings of a more accelerated loss of volume in the MTL relative to the striatum in later life (Raz et al., 2005, 2010; Walhovd et al., 2011). Yet, age-related changes in the striatum are also identified and have been evaluated in the context of reward processing (Mell et al., 2005; Bäckman et al., 2006; Eppinger et al., 2008, 2013; Braver, 2012; Chowdhury et al., 2013; Samanez-Larkin et al., 2014). Older adults show reduced neural response to reward signals such as feedback and differences in the processing of negative and positive rewards relative to younger adults (Holroyd and Coles, 2002; Nieuwenhuis et al., 2002; Eppinger et al., 2008, 2013). Importantly, age-related differences in reward processing have been found to influence learning (Eppinger et al., 2013). Evaluating learning in older adults from immediate and delayed feedback may elucidate how age-related changes in the striatum and MTL effect feedback-based learning.

Only one study (Lighthall et al., 2018) to our knowledge, has evaluated learning from immediate and delayed feedback in older adults to determine whether this manipulation alters learning outcomes and neural activity. Lighthall et al. (2018) evaluated probabilistic learning and recognition memory in immediate (1,000 ms after response) and delayed (7,000 ms after response) feedback conditions. They compared younger (M_years = 26.3) and older adults (M_years = 68.7) to better understand the effects of healthy aging on feedback processing. Behavioral accuracy analyses demonstrated that older adults showed lower rates of optimal response selection relative to young adults, but that both groups showed learning under both timing conditions. Region of interest analyses of functional magnetic resonance imaging data (fMRI) collected during learning demonstrated greater activity in the striatum relative to the hippocampus with immediate feedback in the young adult group, and the opposite pattern with delayed feedback. In contrast, in the older adult group, feedback timing did not lead to significant differences in regional activation during learning. Additional analyses focused on the nucleus accumbens, a region important for dopaminergic reward learning, identified enhanced activation for both groups under conditions of immediate feedback relative to delayed feedback. The authors concluded that findings provide evidence for age-related change in hippocampal mechanisms of learning more so than in striatal mechanisms.

The current study will characterize age-related differences in category learning with immediate and delayed feedback. During learning, individual’s electrophysiological response to feedback will be captured using ERPs. ERPs provide high-temporal resolution of the processing of feedback under different timing conditions. Category learning is central to human cognition and broadly defined as the ability to organize environmental information into meaningful groups based on patterns (Ashby and O’Brien, 2005). Abstract representations of a concept or “prototypes” can aid in categorization. In A/B prototype category learning, category exemplars are derived from an “A” prototype and “B” prototype. “A” category members share relatively more features with prototype “A” while “B” category members share relatively more features with prototype “B.” For example, an email may be categorized as phishing because it shares relatively more features with a scam email (from a financial institution, implies urgency, requests personal information, contains typos) relative to an official business email. To our knowledge, no studies have evaluated Prototype A/B category learning under immediate and delayed feedback. The ubiquity of category learning and the novelty of this investigation further supports the impact of the current work.

Consistent with previous research (Lighthall et al., 2018) we predict that learning outcomes will be equivalent under both immediate and delayed feedback timing conditions but be associated with different signatures of neural activity. In younger adults, we expect to see a larger FRN in response to immediate feedback relative to delayed feedback and a larger N170 in response to delayed relative to immediate feedback. As is characteristic of the FRN, we expect a larger magnitude FRN to negative relative to positive feedback (Sambrook and Goslin, 2015) in both the immediate and delayed conditions. Yet, we recognize that a disruption of feedback processing in the striatum may also be evidenced by the absence of a valence effect in the delayed condition (valence by timing interaction) indicating atypical extraction of outcome-related information conveyed by feedback. We expect to see the same pattern in the FRN for older adults. However, a larger magnitude N170 in response to delayed relative to immediate feedback may be absent in older adults with this affect potentially related to age-related changes in the MTL. We do not expect to see a valence effect in the N170 because the N170 is hypothesized to reflect the binding of temporally discontiguous events and not the extraction of outcome information conveyed via feedback. The findings of this study aim to further characterize consequences of age-related neural changes on feedback-based learning, setting foundations for future research examining how age-related changes in neural functioning may interact with learning after acquired neurologic injury.

The current study aimed to evaluate how altering the timing of feedback influenced feedback processing and learning for younger and older adults. Behavioral data revealed that at the group-level participants learned the category structure as evidenced by an increase in %BResp when an exemplar shared more features with prototype B. A small subset of participants (~25%) did not show evidence of a linearly increasing relationship between %BResp and number of features shared with prototype B. Slope data allowed for the reduction of %BResp to a single score to evaluate response as a function of distance from the prototype. Slope score analysis revealed that older and younger adults had comparable learning in the delayed feedback condition but not the immediate feedback condition. Older adults performed worse in the immediate relative to delayed condition. Delaying feedback may warrant further exploration as a potential means to improve learning in older adults and potentially as a means to equalize performance across younger and older learners.

Electrophysiological data may further clarify behavioral findings. Younger adults showed the expected pattern in the FRN data. There was a larger FRN to negative compared to positive feedback in the immediate but not the delayed condition. These findings are consistent with fast-acting, dopamine-driven processing of immediate feedback that shifts to other circuits when feedback is delayed. Importantly, older adults showed a reduced FRN effect when learning from immediate feedback, suggesting a disruption of feedback processing in the striatum regardless of feedback timing. Age-related limits on older adults’ ability to effectively recruit the striatum to process immediate feedback may explain why younger adults outperformed older adults when feedback was immediate and dependent upon fast-acting subcortical reward processing (Nieuwenhuis et al., 2002; Eppinger et al., 2008).

The N170 amplitude was larger for both groups in the delayed relative to the immediate feedback condition. Again, this supports the notion that delaying feedback alters the neural processing of feedback and that this is captured by the N170. One potential neural generator is the MTL (Arbel et al., 2017; Kim and Arbel, 2019; Höltje and Mecklinger, 2020) which supports the integration of temporally discontiguous information. Specifically, research finds that the hippocampus contains “time cells” that encode key events that are separated by a temporal gap (MacDonald et al., 2011, 2013; Kitamura et al., 2015). In the context of the current task, the response (e.g., This animal lives in a cave) and feedback (e.g., incorrect) are temporally discontiguous events in the delayed condition (i.e., separated by a 6,000 ms time gap). What is known about the MTL and the demands of this task, make the MTL a potential candidate for the current task. However, further research is necessary to investigate the neural generator in this context and potential alternative sources (e.g., the fusiform gyrus).

Older adults learned better in the delayed relative to immediate condition suggesting that delaying feedback may be a fruitful way to support feedback-based learning in older adults. The locus of the gain in learning for older adults remains to be elucidated. One potential explanation is that delayed feedback allowed older adults to rely on the MTL to update cue-response contingencies when separated by a delay. The MTL may be better suited to support Prototype A/B learning in the setting of age-related neural changes compared to the striatum. While previous research has suggested a steeper age-related decline in the functioning of the MTL relative to the striatum in older adults, there is also a body of work which characterizes age-related dysfunction in reward processing within the striatal system with aging (Mell et al., 2005; Bäckman et al., 2006; Eppinger et al., 2008, 2013; Braver, 2012; Chowdhury et al., 2013; Samanez-Larkin et al., 2014). Changes in reward processing may also disrupt learning within the current task. A series of studies (Holroyd and Coles, 2002; Nieuwenhuis et al., 2002; Eppinger et al., 2008) have evaluated age-related reduction in feedback processing using the FRN. Eppinger et al. (2008) found that even when controlling for performance differences, older adults showed an atypical pattern in the processing of negative and positive feedback. The current findings are consistent with an age-related decline in dopaminergic processing reflected in symmetric processing of negative and positive feedback in older adults. Considering the current results within this context, age-related changes in the dopaminergic reward system may be more detrimental to performance on a Prototype A/B learning task than age-related changes in the MTL.

Alternatively, it could be that the delaying of feedback altered attentional recruitment in a manner that was advantageous for older adults. Post-hoc analyses revealed an interaction in which at P7 older, but not younger adults, showed a larger P100 for immediate relative to delayed feedback. Thus, immediate but not delayed feedback was associated with increased attentional allocation. One potential interpretation of this finding is that when older adults are unable to effectively rely on striatal mechanisms to process immediate feedback, more attention is allocated to the feedback signal.

Differences in demands on cognitive processing speed across the immediate and delayed feedback tasks may also contribute to age-related differences. Aging has long been associated with a gradual reduction in cognitive processing speed (Cerella and Hale, 1994). Delaying feedback may give adults sufficient time to process incoming stimuli and the resultant feedback despite reduced rates of processing. Potential explanations as to why older adults seemingly benefitted from delays in feedback timing are not mutually exclusive and warrant further evaluation to understand neural underpinnings.

The findings of the current study do not align with Lighthall et al. (2018) in which age-related decline was more evident in the hippocampal relative to striatal regions. Differences across studies may come from variations in the timing of immediate feedback. Immediate feedback was presented at 500 ms in the current study and at 1000 ms in Lighthall et al. (2018). Worthy et al. (2013) suggests that timing differences as small as 500 ms can affect learning by influencing the intracellular chemical concentrations at the time of the dopaminergic reward signal. In support of this hypothesis, Worthy et al. (2013) found that during an information integration category learning task accuracy was highest when feedback was presented at 500 ms compared to 0 ms or 1,000 ms. These findings suggest that differences in timing on the scale of milliseconds may influence learning driven by the striatum and could potentially explain differences in findings across otherwise comparable studies. Of course, other methodological differences across the current study and Lighthall et al. (2018) may also drive differences across studies; thus, future research should evaluate the reproducibility of the current findings. One such difference is the task type. We chose a prototype learning task given its relevance across the lifespan and that the effect of feedback timing has yet to evaluated. However, in a prototype A/B learning task, feedback may not be informative on a trial-by-trial basis and thus, affect the electrophysiological response to feedback.

If a consistent advantage for learning from delayed feedback is identified in older adults, or if individual differences arise, manipulating feedback timing may serve to optimize learning in this population. Techniques aimed at enhancing learning may be particularly useful for older adults with acquired neurologic impairments. Adults with acquired neurologic injury often receive rehabilitation services in which they re-learn skills and learn new compensatory strategies. Rehabilitation specialists must administer treatments in ways that target the impaired system but also successfully engage functioning systems of learning to induce neuroplastic change. As rehabilitation fields continue to move toward theory-driven treatments that delineate which treatment ingredients engage proposed mechanisms of action and how identified ingredients should be administered, feedback timing should not be overlooked.

This work suggests that in a prototype A/B category learning task, the timing of feedback (immediate vs. delayed) may have distinct consequences for learning in younger and older adults. Differences in learning across groups and timing conditions were associated with differences in electrophysiological response to feedback. Notably, older adults learned better from delayed relative to immediate feedback, potentially due to age-related changes in the neural mechanisms responsible for processing feedback. Future research is needed to determine the probable locus of the learning advantage with delayed feedback in older adults and can work toward understanding how feedback timing may be manipulated to promote learning across the lifespan.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by Institutional Review Boards of Mass General Brigham. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

KN: Formal analysis, Investigation, Visualization, Writing – original draft, Writing – review & editing. RC: Conceptualization, Investigation, Methodology, Writing – review & editing. VT-B: Investigation, Writing – review & editing. YA: Conceptualization, Formal analysis, Funding acquisition, Methodology, Resources, Software, Supervision, Writing – review & editing. SV-R: Conceptualization, Formal analysis, Funding acquisition, Methodology, Resources, Supervision, Writing – review & editing.