Why Don’t Cognitive Training Programs Transfer to Real Life?

Cognitive mechanisms are implicated in nearly every type of psychological disorder, from addiction to mood disorders to psychopathy. Since the early writings of Aaron Beck more than 50 years ago (Beck, 1967), clinical scientists have devoted considerable time and effort into elucidating the specific aspects of cognitive functioning that underlie psychological disorders and symptoms. Based on these studies, a natural extension of this field led to the question of whether directly modifying cognition might yield improvements in symptoms and functioning. This question has been approached both through cognitive bias modification (CBM; attempts to modify symptom-specific biases in attention, interpretation, or memory) and cognitive training of “basic” cognitive functions such as working memory and inhibitory control. Aided by rapid improvements in technology, these computerized cognitive training approaches have been extensively tested as potential treatments for psychological disorders. Unfortunately, many of these studies have failed to find evidence that cognitive training improves clinically meaningful outcomes such as functioning or symptoms (known as “far transfer”), despite more consistent evidence that training improves specific aspects of cognition related to the training itself (“near transfer”). This poses a vexing question for clinical scientists: why do cognitive training interventions so often fail to transfer to real-world outcomes, even when these same interventions seem to engage the cognitive mechanisms of interest?

Previous commentaries, reviews, and meta-analyses have extensively litigated this question (Sala & Gobet, 2017; Simons et al., 2016; Wiers, 2018). Although the degree of optimism for cognitive training varies substantially across researchers, the consensus from many is that cognitive training has not worked as well as initially hoped. Clearly, one possibility is that far transfer might not be possible for most behaviors (e.g., Sala & Gobet, 2017). However, this interpretation is difficult to reconcile with findings from recent well-controlled, high quality trials that show specific benefits associated with cognitive training for certain symptoms (e.g., Hsu et al., 2021; Siegle et al., 2014). To better understand the conditions under which far transfer is possible, several recent publications provided concrete methodological suggestions for improving future research in this area (Redick, 2019; Smid, Karbach, & Steinbeis, 2020) and for precisely defining transfer effects (Harvey, McGurk, Mahncke, & Wykes, 2018). The goal of this commentary is not to rehash these suggestions, but rather to highlight three broader ideas for how future studies could deliver cognitive training with the greatest potential to yield far transfer.

Better Personalization

Cognitive training will almost certainly fail to transfer if the training being offered does not match the specific needs of an individual. Just as a cognitive-behavioral therapist would likely refrain from restructuring a thought that is not biased, cognitive training methods are likely only helpful if matched to specific biases or other cognitive needs. Multiple researchers have recently proposed that successful outcomes in cognitive training studies could be better achieved by incorporating personalized methods (Redick, 2019; Smid et al., 2020). Personalization of training can take many forms. At a broad level, this might include selecting participants based on demographic characteristics linked to successful training outcomes: for example, Price and colleagues demonstrated that across studies, attentional bias training was most effective for reducing anxiety symptoms among participants younger than age 37 (Price et al., 2016). In the case of cognitive bias modification paradigms, another promising aspect of personalization involves only selecting those participants who demonstrate the bias of interest on a pre-screening measure. In a recent large, pre-registered randomized clinical trial, Hsu and colleagues demonstrated that attention bias modification was associated with decreased depression symptoms in comparison to a sham condition; importantly, this study included only those participants who showed a pre-existing bias to negative images (Hsu et al., 2021). Recent studies of cognitive bias modification for negative interpretations have also embraced personalization, such as by allowing participants to select themes and situations that are personally relevant prior to beginning training with these stimuli (e.g., Beard et al., 2021). Other ongoing studies highlight the promise of personalizing interventions on multiple metrics simultaneously, through the use of machine learning methods applied to predictors of cognitive training outcomes (Shani et al., 2021).

Personalization can also be achieved by testing how specific neural mechanisms predict training outcomes, and assigning patients to treatment on the basis of these markers. Many researchers have long advocated for the need to assess both clinical outcomes and underlying neural mechanisms to understand how cognitive training interventions work (e.g., Siegle et al., 2007). Unfortunately, few studies of cognitive training have included pre and post-training neuroimaging assessments, and even fewer have assigned patients to specific interventions on the basis of neural markers (Baykara et al., 2021). The time and cost required for neuroimaging is undoubtedly a barrier to this work. However, recent findings highlight some of the ways in which clinical and cognitive outcomes of attentional training for transdiagnostic anxiety can be predicted by multiple measures of neural functioning. For example, in addition to functional MRI (Price et al., 2018); response to attentional training can also be predicted by peripheral measures such as by pupillometry (Woody, Vaughn-Coaxum, Siegle, & Price, 2020). Other research demonstrates changes in specific event-related potentials captured in EEG paradigms during cognitive training (e.g., Hartmann et al., 2016). Thus, future studies seeking to personalize training procedures may be enhanced by incorporating one or more of these potential neurocognitive predictors of training outcome.

Interventions at the Right Time and Place

Another barrier to transfer may be that cognitive training interventions are too far removed – both temporally and conceptually – from the outcomes that are desired. Many cognitive training studies have implicitly or explicitly relied on the “exercise routine” metaphor, in which it is assumed that training exerts effects through repeated practice of tasks. For example, in studies targeting rumination via working memory training, it is hoped that a general improvement in working memory will transfer to more effective use of working memory at the moments when someone is ruminating. This type of transfer may be possible, but a major limitation of this approach is that cognitive tasks are not typically practiced at the same time that a person is experiencing a given symptom. This is problematic for several reasons. First, the temporal gap between a training session and a moment when a person might “use” the training in action, which could be hours or even days, makes it difficult to pinpoint mechanisms of how training works, and difficult to tell how much the training sessions are responsible for momentary changes in symptoms or other outcomes. Second, cognitive functioning is variable within an individual, and intraindividual variability is tied to a number of dynamic factors such as state affect (Weizenbaum, Torous, & Fulford, 2021). Thus, the working memory resources a person uses in a cognitive training session on Tuesday afternoon might be appreciably different than the working memory capacity they have available on Friday night. One promising method that can meet this challenge is that of smartphone-based Ecological Momentary Interventions (EMIs). As technology rapidly advances, it is increasingly possible to achieve reliable measurement of momentary cognitive functioning via smartphone (Germine et al., 2019; Weizenbaum et al., 2021), which opens the door for cognitive training procedures to be delivered via this same mechanism. Future studies could take advantage of this approach by using smartphone-based assessments to measure momentary changes in a process of interest (for example, rumination), and then tailor cognitive training interventions exactly when needed, thereby merging the cognitive training field with the advancing science on this “Just-in Time” Adaptive Interventions (JiTAI) approach (Nahum-Shani et al., 2018). “Momentary” cognitive interventions such as these do not preclude the possibility that the traditional method of delivering cognitive training (e.g., spaced-out sessions at regular intervals) are also efficacious. Future studies would do well to evaluate ways to maximize transfer effects that combine or compare these approaches.

A related limitation of transfer effects is that cognitive training alone may not give participants enough information to understand how to apply those same cognitive resources in real-life contexts. This is likely a less significant problem for cognitive bias modification studies that incorporate stimuli directly relevant to the outcome of interest, (such as pictures of alcohol in attentional training procedures for alcohol use disorder), in contrast to paradigms that use basic executive functioning tasks (Wiers, 2018). Explicitly incorporating assessments of what participants think about the intervention, incorporating both quantitative and qualitative methods (e.g., Yardley et al., 2015), can be a fruitful addition in order to better understand how participants think about the training. For example, a recent study by this author found that many participants did not understand how working memory training was relevant to their real-life impulse control (Peckham et al., under review). However, there is encouraging evidence that combining cognitive training methods can be enhanced and made significantly more practical with the addition of explicit psychosocial interventions that help to actively practice real-life scenarios in which cognitive skills can be used (e.g., Bowie et al., 2012; Bowie et al., 2017). Future studies would do well to determine best practices for combining cognitive training procedures with explicit strategy coaching in order to refine our understanding about how to maximize transfer effects from such interventions.

Enhancing Multiculturalism in Cognitive Training Research

The existing literature on cognitive training also suffers from the same problem as the field of psychology as a whole: namely, a significant lack of racial and ethnic diversity among study participants. As others have noted (e.g., Hsu et al., 2021), many cognitive training studies rely on WEIRD (White, Educated, Industrialized, Rich, and Democratic; Henrich et al., 2010) samples that do not reflect the demographic realities of the broader community. There is also concerning evidence that cognitive psychology lags behind other areas of the field in terms of racial diversity of study participants, and also racial diversity among journal editors and authors (Roberts et al., 2020). These problems represent a significant barrier for transfer effects: if training paradigms are overwhelmingly designed by and tested among White individuals, there is reason to be skeptical that any transfer effects will reproduce outside of these narrow confines.

Comprehensive recommendations for enhancing inclusivity and multiculturalism in clinical psychology have been documented elsewhere (Galán et al., 2021), and nearly all such recommendations apply to the field of cognitive training as well. In addition to these suggestions, two additional areas of concern might be particularly relevant to cognitive training research. First, many proponents of cognitive training argue that the potential reach of this approach – being able to deliver training remotely via the internet – can allow for the dissemination and rapid scaling-up of such treatments. However, this strength is inherently limited by disparities in high-speed internet access, with clear racial disparities manifesting in lower rates of internet access for Black and Latinx households across the United States (Nallen, 2020). Addressing this barrier will require those of us in the cognitive training field to think creatively about issues such as how to ensure our interventions are available on devices that do not require constant internet connectivity and to ensure that internet access is not a barrier to study recruitment. A second potential threat to transfer effects stems from the overwhelming preponderance of study designs that use White faces for stimuli. This issue is likely most relevant to cognitive bias modification paradigms, such as attention training, that involve images as an inherent part of study design. The extent to which racial differences in face stimuli relate to differences in image processing is an area of ongoing research (e.g., Gonzalez & Schnyer, 2019), but very few cognitive training studies have explicitly incorporated racial diversity into study stimuli. Fortunately, there are multiple options for researchers seeking to enhance racial diversity in paradigms that require affectively valenced faces, such as the RADIATE database (Conley et al., 2018). Incorporating a more racially diverse array of stimuli will be an important priority for future studies of attentional bias modification in particular.

Summary and Conclusions

A recent article described the field of working memory training as going through phases of the “hype cycle,” in which a “trough of disillusionment” follows the “peak of inflated expectations,” although some technologies may reach beyond this to the “slope of enlightenment” (Redick, 2019). Given well-documented null findings and questions about the validity of transfer effects, this metaphor indeed appears appropriate for the field of cognitive training in general. Yet, by considering the many ways in which cognitive training may fail to transfer to real-life outcomes, there are also reasons to be hopeful that potential innovations could be revealed. By continuing to conduct rigorous research on transfer effects using the recommendations outlined above, cognitive training interventions designed to address symptoms and functioning may yet hold promise.