Time to Forget

Research Paradigms

Research on intentional forgetting has revealed that individuals are able to deliberately forget memory contents at least for a certain period of time (Anderson & Green, 2001; Anderson & Hanslmayr, 2014; Anderson & Levy, 2009; Bjork, Bjork, & Anderson, 1998; Hotta & Kawaguchi, 2009; Nørby, Lange, & Larsen, 2010; Noreen & MacLeod, 2014). Intentional forgetting has been investigated using different experimental paradigms, demonstrating both its costs and benefits. The most commonly used paradigms are the directed forgetting paradigm (item method and list method; Bjork et al., 1998; Bjork, LaBerge, & Legrand, 1968) and the think / no think paradigm (TNT; Anderson & Green, 2001). In both paradigms, subjects are instructed to either forget learned items or remember them.

In a typical directed forgetting experiment using the list method, individuals are required to learn a list of words before being instructed to either forget (forgetting condition) or remember (remember condition) the words on this first list. Afterwards, individuals are instructed to learn new words from a second list. In a final recall, individuals are asked to remember as many words as they can from both lists, even if they were previously asked to forget the first list. Two results are relevant here: (1) Fewer words from the first list are remembered in the forgetting condition than in the remember condition; and (2) in the forgetting condition, words from the second list are better remembered than words from the first list (Bjork, 1989; but see Sheard & MacLeod, 2005).

The TNT paradigm (Anderson & Green, 2001) is an experimental paradigm to assess an individual’s ability to suppress the retrieval of an unwanted memory. In the first phase, individuals have to learn pairs of words (e. g., lint–curtain) by practicing recalling the target word (e. g., curtain) whenever the cue (e. g., lint) is presented. In the second phase, two thirds of the cues are presented, and the participants have to either remember the learned word (think, e. g., curtain) or not remember the learned word by preventing it from entering their awareness (no think, variant: replacing it with another word). The remaining third of cues is not presented again and serves as the baseline for the third phase. In the third phase, a memory test for all item pairs is conducted. The typical pattern of the results is that individuals remember the no-think items worse, and the think items better, than the baseline items. In contrast to the directed forgetting paradigm (which compares experimental conditions between participants), the TNT paradigm allows individual forgetting abilities to be measured (by comparing word lists within participants).

Both experimental paradigms, directed forgetting and TNT, reflect task requirements that can also be found in everyday situations. Directed forgetting reflects the situation in which entire events (in this case, lists) or actions have to be forgotten because they are irrelevant now and will continue to be in the future. In everyday working life, this situation is experienced when, for instance, some regulations are no longer relevant, because they have been replaced by new regulations. The TNT paradigm reflects situations where individuals who are distracted by unwanted thoughts actively inhibit and possibly replace these thoughts with other content. In day-to-day work, these cognitive processes are required in situations where, for example, thinking about a conflict with a colleague or the student thesis on one’s desk can distract from one’s actual work task, if one does not try to forget these unwanted thoughts. It seems that intentional forgetting is also of great importance in everyday work and does not seem to be just a laboratory phenomenon.

Theoretical Explanations

Three theoretical frameworks have been proposed to explain the costs and benefits of intentional forgetting. The selective rehearsal hypothesis (Bjork, 1972) postulates that being instructed to forget stops the rehearsal of the learned words (Sheard & MacLeod, 2005). The retrieval-inhibition hypothesis (Anderson & Green, 2001, Bjork, 1989; Conway, Harries, Noyes, Racsmány, & Frankish, 2000) assumes that being instructed to forget temporarily inhibits the retrieval of the corresponding memory contents (costs), which in turn reduces proactive interference with the first list and supports learning and retention of the second list (benefits). The third explanation, the mental context change hypothesis is a two-process hypothesis that attributes the costs and benefits of intentional forgetting to different processes (Sahakyan & Kelley, 2002; also Bäuml, Hanslmayr, Pastötter, & Klimesch, 2008; Lehman & Malmberg, 2011). The first process (context change) explains forgetting with the notion that being instructed to forget creates a different mental context when learning the new items. The new context thus does not serve as a retrieval cue for the first list, making recall more difficult. The second process is the strategy change process: In the forgetting condition, people use better strategies to learn the words in the second list.

Empirical evidence from at least one experimental paradigm exists for all three theoretical approaches. Anderson (2005) postulates that both the retrieval inhibition hypothesis and the mental context change hypothesis rely on a general inhibitory mechanism for the to-be-forgotten items, in that mental contexts can also be inhibited, which results in a lower recall rate (Anderson, 2003).

Related Concepts

Intentional forgetting can be distinguished from related constructs pertinent to human cognitive control including self-regulation, self-control, thought control, and implemental mindset. Self-regulation is a broad construct referring to a process ranging from goal setting and goal striving to monitoring goal progress and protecting goal striving by overcoming obstacles and temptations (Hofmann, Schmeichel, & Baddeley, 2012; Schmeichel & Baumeister, 2004). Self-regulation, and self-control to an even greater extent, are similar to intentional forgetting. Self-control as a narrower subset of self-regulatory processes refers to the ability to override or inhibit automatic, prepotent impulses, urges, or emotions to protect purposeful, goal-directed behavior (e. g., Baumeister, Vohs, & Tice, 2007; Metcalfe & Mischel, 1999). Thus, self-control comprises the inhibition of cognitions and emotions. Suppressing unwanted memory contents can be seen as a type of self-control strategy that takes the form of limiting access to unwanted memory contents rather than of suppressing impulses or negative emotions. Therefore, the main difference between intentional forgetting and self-control is that intentional forgetting focuses primarily on memory storage and memory processes rather than on goal setting and striving, emotion regulation, and resisting and overcoming temptations, which include also motivational processes. Self-regulation, self-control, and intentional forgetting have in common that they are the subjects of several executive functions (Hofmann et al., 2012; Miyake & Friedman, 2012), and inhibitory control processes are discussed as an important executive function concerning all three constructs (Anderson, 2005; Hofmann et al. 2012; Schmeichel, 2007). It is important to note that intentional control, on the one hand, and self-regulation and self-control, on the other, stem from different research areas – namely, cognitive psychology and social psychology, respectively (Hofmann et al., 2012), with each investigating the respective phenomenon without referring to the other research area. However, recently Engen and Anderson (2018) have proposed in their analysis of this that memory control (i. e., intentional forgetting) might be an important component in emotion regulation – that is, one specific kind of self-control.

With respect to the question of how workers manage the abundance of information in their workplaces, examining memory processes such as intentional forgetting of irrelevant information rather than impulse control seems to be especially helpful.

Thought control is a cognitive control concept similar to intentional forgetting, and has mainly been investigated in clinical psychology. Thought control refers to the suppression of unwanted, intrusive thoughts (Wells & Davies, 1994; Wegner, Schneider, Carter, & White, 1987) which accompany anxiety disorders and posttraumatic stress. Unwanted intrusions are controlled with techniques such as thought stopping, controlled worry periods, cognitive reappraisal, and distraction (Wenzlaff & Wegner, 2000). In contrast to thought control, intentional forgetting is also directed at nonemotional memory contents, such as task-irrelevant information, rather than negative thoughts alone. In sum, we argue that the concept of intentional forgetting which refers to a process that controls irrelevant memory contents, can be especially helpful in dealing with the large volumes of digital information.

Complementing intentional forgetting is the concept of implemental mindset, which helps to focus on the task at hand by tuning an individual’s cognitions to action-related information (information on where, when, and how to act; Gollwitzer & Bayer, 1999). Both concepts reflect motivated information processing. Intentional forgetting, however, aims to achieve the opposite result, by suppressing unwanted memory contents.

Some Key Findings Regarding Intentional Forgetting

Intentional forgetting is a conscious process: Forgetting increases when participants have time to prepare suppression (Hanslmayr et al., 2010) and even remains stable when they are paid for correct responses (Anderson & Green, 2001) – showing that the forgetting effect does not rely exclusively on the deliberate withholding of responses that participants were actually conscious of. Intentional forgetting limits immediate retrieval but does not necessarily occur after a single attempt to permanently forget (Hotta & Kawaguchi, 2009; Nørby et al., 2010; Noreen & MacLeod, 2014). The more often people try to stop the retrieval, the worse the final recall turns out to be (Anderson & Green, 2001; Anderson & Huddleston, 2011; Hanslmayr, Leipold, & Bäuml, 2010; Lambert, Good, & Kirk, 2010). Thus, it seems that suppressing unwanted thoughts helps people to temporarily forget (e. g., a competing task), whereby sustained forgetting requires many attempts to suppress memory. However, van Schie and Anderson (2017) found that suppressing unwanted thoughts for a longer period of time was more effortful than to do so for a shorter period of time. Over time, sustained engagement in suppression of unwanted thoughts even impaired inhibitory control. Consequently, support of workers’ attempts to suppress task-irrelevant information temporarily or permanently in an information-rich context might be especially useful, as it would support the success of forgetting and reduce mental effort.

The line of research presented here seems to be in contradiction with the literature on the white bear phenomenon (see Wenzlaff & Wegner, 2000): To test this paradigm, participants were advised to not think of a certain object (e. g., a white bear) for several minutes. Results typically showed an ironic effect, as most participants reported frequent intrusions of the to-be-avoided object. This effect is often cited to show the ineffectiveness of thought suppression. However, one major difference from the TNT and directed forgetting paradigms is that in the white bear paradigm, the instruction to suppress already includes the to-be-suppressed memory, making it nearly impossible to ignore it. Rehearsing the instructions already involves a violation of the task goal. The TNT and directed forgetting paradigms work with reminders (specific cues or list affiliations) which do not refer to the target word. Further, van Schie and Anderson (2017) argue that suppression for a longer period of time reduces inhibitory control, which is the case with the white bear experiments.

Intentional forgetting has been shown to work not only with neutral memory contents, but also with emotional contents. This is important because to-be-avoided thoughts in everyday or working life are often complex, emotional, and negatively valenced. The forgetting effect has been widely replicated with both positive (Joormann, Hertel, Brozovich, & Gotlib, 2005; Nørby et al., 2010; for an exception, see Payne & Corrigan, 2007) and negative memory contents, albeit in an inconsistent manner. There is some evidence that the suppression of negative memories leads to inhibition comparable to that for neutral or positive stimuli (Depue, Banich, & Curran, 2006; Joormann et al., 2005; van Schie, Geraerts, & Anderson, 2013). However, some studies have failed to find an effect for negatively valenced words (e. g., Chen et al., 2012; Nørby et al., 2010; Marx, Marshall, & Castro, 2008; Yang, Lei, & Anderson, 2016). Thus, forgetting negative content seems possible but more difficult than forgetting neutral or positive memories (Yang et al., 2016). Recently, based on neurobiological and behavioral studies, Engen and Anderson (2018) have argued that inhibitory control of memory contents plays an important role in emotion regulation.

Intentional forgetting was originally investigated with word lists and word pairs (e. g., Anderson & Green, 2001; Bjork, 1972) but also with pictures (e. g., face-picture pairs; Chen et al., 2012; see also Depue et al., 2006; Gagnepain, Henson, & Anderson, 2014), action phrases (e. g., verbally learning or symbolically performing short actions such as inflating a balloon; Sahakyan & Foster, 2009), autobiographical events (e. g., individual events occurring in everyday life; Joslyn & Oakes, 2005; see also Noreen & MacLeod, 2013, 2014;Stephens, Braid, & Hertel, 2013), and habits (Dreisbach & Bäuml, 2014). Using the list method of the directed forgetting paradigm, Dreisbach and Bäuml (2014) asked their participants to learn and practice stimulus–response rules which changed after a while. Participants were instructed to either remember or forget the stimulus–response rules they had practiced. The accessibility of these rules was reduced when participants were cued to forget them rather than to remember them. These findings indicate that not only words but also habits can be intentionally forgotten. This is especially important in the work context when working routines are changed as a result of a technological or organizational change: The workers have to forget their old routines in order to learn new ones.

Only a few studies have examined intentional forgetting outside of the laboratory. Joslyn and Oakes (2005) demonstrated that people are capable of intentionally forgetting autobiographical memories (e. g., narrative descriptions of specific events that occurred in the participant’s everyday life) regardless of their valence and emotional intensity, which might explain the loss of at least some autobiographical memories over the course of one’s lifetime. Streb and colleagues (2016) demonstrated that a high ability to forget was associated with less stressful intrusions in daily life. In a longitudinal field study, Niessen, Binnewies, and Rank (2010) have shown that suppression of thoughts related to previous work roles positively predicted adaptation to a new work role (in this case, becoming self-employed) by reducing negative consequences of psychological attachment to those previous roles.

Individual Differences

It is important to note that people differ significantly in their ability to intentionally forget. Who finds it easier or harder to suppress distracting thoughts? There is evidence that age is negatively related to intentional forgetting (Anderson, Reinholz, Kuhl, & Mayr, 2011; Hasher & Zacks, 1988; Rizio & Dennis, 2014). Recently, Hulbert and Anderson (2018) conducted experiments to test whether inhibitory control depends on practice. They found that individuals reporting a greater history of trauma – which should indicate that they have had more practice in suppressing unwanted thoughts – also had enhanced memory inhibition abilities. This might also be true for individuals with a high need to suppress intrusive memories in everyday working life – for example, workers in professions dealing with emotionally demanding situations such as paramedics, workers in emergency call centers, or psychotherapists. Levy and Anderson (2008) considered executive control ability to be responsible for these differences (executive deficit hypothesis).

Affective states also seem to have an influence on intentional forgetting: Bäuml and Kuhbandner (2009) showed that people in a neutral or negative mood were able to intentionally forget (study material was neutral). In contrast, Noreen and Ridout (2016) found forgetting effects only for positive, but not for negative mood inductions, using valenced study materials. Further analyses revealed that participants in the negative mood condition had more difficulty with forgetting negative materials, whereas those in the positive mood condition did not. In line with this finding, several studies have shown that depression can lead to a modification of typical forgetting patterns: Dysphoric and depressive people generally have more difficulty in forgetting intentionally (e. g., Hertel & Gerstle, 2003; Joormann & Tran, 2009; Noreen & Ridout, 2016; Power, Dalgleish, Claudio, Tata, & Kentish, 2000). This effect is probably due to the downsized activity in the dorsolateral prefrontal cortex – a brain area that is usually activated during memory inhibition processes (Anderson et al., 2004; Benoit & Anderson, 2012).

A few studies have also investigated whether personality and cognitive abilities matter for intentional forgetting. There is some evidence that empathy and sensitization correlate with forgetting difficulties, especially for forgetting of negative material (Geiselman & Panting, 1985). Aslan, Zellner, and Bäuml (2010) further showed that working memory capacity positively predicts intentional forgetting (see also Delaney & Sahakyan, 2007). Fawcett and Taylor (2012) suggest that intentional forgetting is associated with establishing control over the contents of working memory. As working memory, executive attention, and fluid intelligence are located together in the prefrontal cortex (Kane & Engle, 2002), it is likely that high fluid intelligence may also be positively associated with intentional forgetting abilities. Thus, it seems likely that people with a lower working memory capacity, with increasing age, with less practice in forgetting, and with a negative, depressed mood are likely to have more difficulty in suppressing irrelevant information, which could be compensated for by an assistive system.

Situational Factors

Situational factors which might foster or hinder intentional forgetting have been neglected in cognitive psychology research so far. Studies have shown that additional or competing tasks requiring executive control capacity (Conway et al., 2000; Levy & Anderson, 2008; Miyake & Friedman, 2012) impair the inhibition of unwanted thoughts and behaviors (e. g., Lavie, Hirst, de Fockert, & Viding, 2004; Schmeichel, 2007; Ward & Mann, 2000). More recently, Noreen and de Fockert (2017) investigated inhibitory control of unwanted thoughts under a low and high working memory load task. They found that under higher working memory loads, individuals switched their strategies to prevent unwanted memory contents from coming to mind: Individuals directed their attention toward the high load task and engaged less in suppression (interference-based strategy).

Thus, it seems reasonable that typical situational factors at work such as a high workload and high task complexity may impair the intention to inhibit memory content. Still, there is a lack of studies on intentional forgetting at work that also consider situational factors related to the work environment. This lack is particularly important for the development of any assistive system, which is supposed to support users under varying situational factors such as time pressure, task complexity, etc.

Functions of Intentional Forgetting

Why should workers forget at work? Although the functions of intentional forgetting have not been empirically investigated as yet, a broad range of potential functions has been proposed (Anderson & Hanslmayr, 2014; Nørby, 2015): Intentional forgetting should serve to regulate negative affect, justify inappropriate behavior (e. g., by forgetting moral rules), maintain beliefs and attitudes (e. g., by forgetting contradictory thoughts), deceive others and oneself, preserve one’s self-image, forgive others, and maintain social relationships. Nørby (2015) further mentioned that intentional forgetting facilitates knowledge acquisition and learning (through the forgetting of redundant and false information), and context attunement (through the forgetting of outdated information). Thereby, forgetting redundant and false information helps to organize knowledge and to process information more efficiently.

The Context-Aware Cognitive Companion System Dare2Del

Dare2Del is an assistive system which supports users in managing the complex task of dealing with ever-growing amounts of digital information (see Figure 1). It helps users to decide which files can be archived or deleted and thereby get rid of irrelevant digital detritus. A second area of support is to hide information which is irrelevant in a given task context, thereby helping the user to focus. Dare2Del takes the form of a cognitive companion (Forbus & Hinrichs, 2006; Siebers, Schmid, Göbel, & Niessen, 2017), in which human–computer interaction is understood as mutual support. This is in contrast to assistive systems where the system provides information but intelligent decision making is completely left up to the user; it is also in contrast to fully autonomous systems where the system decides for the user.

In general, a companion system should fulfill the following requirements (see Siebers et al., 2017): (a) the system’s decisions should be transparent to the user – that is, the system must be able to explain its decisions by making the criteria that led to its decision explicit. (b) The system must be able to exploit predefined rules, such as specific company regulations or user preferences. (c) The system must be context aware such that it can infer what digital objects are relevant in a given working context, and it can infer when it would be appropriate to support the user. (d) The system must be able to adapt to the needs and preferences of specific users.

We illustrate how Dare2Del fulfills these requirements with two examples. In the first application scenario, Dare2Del recommends files to be deleted or archived, independently of whether the user still has the content of these files in his/her memory. When the contents of these files are represented in the user’s memory, deleting external cues (e. g., the file) will hinder retrieval of task-irrelevant memory contents, which are unwanted and distracting in the current situation. Also, Dare2Del allows organizations to include regulations and legal obligations for data retention. These regulations are included in the recommendations to delete data or files, as made by Dare2Del. Further, when a person starts using Dare2Del, they can also configure the times when the system will point to irrelevant files. And the number of files that will be presented each time can also be configured by the user. For example, one user might prefer to have short sessions at the end of every workday where only five files are suggested for deletion or archiving, while another user might prefer a longer session every Friday. The companion system will also detect when a new project or topic has been started, and will accordingly ask the user to delete or archive redundant or irrelevant files. Users should always be free to override the system dialogue (e. g., “not right now”). Furthermore, criteria about what files are irrelevant can be specified in accordance with company guidelines or personal preferences.

The user should be able to ask for an explanation whenever the system proposes a file to be irrelevant, regardless of whether they agree with the proposal or not (Siebers & Schmid, 2019). This explanation can be a verbalization of the underlying rule (e. g., “file X is considered to be a draft version of file Y”), thus informing the user why the file is proposed to be irrelevant (see Figure 2). Explanations can be further elaborated. A user might want to know why Dare2Del considers file X to be a draft, and Dare2Del can provide detailed information about the files that led to the classification as draft (e. g., “file X has a similar name to file Y and is older than file Y”). The files presented to the user are drawn from the sample of all files, using a probability distribution, and are tested against a set of rules. These rules classify a file as irrelevant or not irrelevant. Some of the rules are predefined as described above. However, it is crucial for the system that rules are learned from interactions with the user. After a selection of files has been presented, the user’s decisions to accept or decline the proposal that a file is irrelevant are used as training examples for an incremental machine learning approach. In Dare2Del, we use an inductive logic programming (ILP) approach (see below).

A second application scenario for Dare2Del is that digital objects are temporarily hidden to support the user in focusing on their current task. For example, one might not want to be distracted by incoming emails while checking monthly balances. One way of looking at this scenario is in the example of spreadsheets. Spreadsheets often contain complex information over many rows and columns. A large number of these rows and columns might be irrelevant when checking certain information. For example, when checking annual holidays and overtime for employees from a specific department in a financial accounting spreadsheet, the rows for employees from other departments as well as many of the columns (e. g., private address, salary, etc.) are irrelevant. Although spreadsheet programs offer numerous capabilities to hide and filter information, many users do not have sufficient knowledge and skills to apply these tools in an efficient way. Therefore, Dare2Del makes it possible to hide information in spreadsheets on the basis of current user intentions, to support the inhibition of irrelevant information and thus to prevent disturbances which direct attention away from the task at hand. Users’ intentions thus have to be inferred from their interaction with the program. In this, we can build on recently developed approaches to end-user programming and intention recognition (see below). However, it remains an open question as to how to identify the users’ actual aims and needs.

Dare2Del is conceptualized as a so-called white-box learning system, which combines classic knowledge-based approaches (Beierle & Kern-Isberner, 2014) and machine learning; most notably, the LIME (Local Interpretable Model-Agnostic Explanations) system (Ribeiro, Singh, & Guestrin, 2016) shows how the classification decisions of a (deep) neural network trained on images can be made transparent using visualization. However, information in the context of forgetting digital objects is presented as a symbolic representation, such as file names or attributes. Thus, current approaches in the context of image classification are not suitable.

An even more powerful, symbolic approach to machine learning is inductive logic programming (ILP; Muggleton & de Raedt, 1994; Muggleton, Schmid, Zeller, Tamaddoni-Nezhad, & Besold, 2018). ILP forms the core of the Dare2Del companion. It naturally combines the exploitation of prespecified knowledge and machine learning and allows for learning not only based on simple features, such as the size of a file, but also on relations between objects, such as the fact that two files are stored in the same folder. However, ILP approaches – like most machine learning approaches – typically require the system to first learn from a sufficiently large and representative sample of labeled training examples; only afterwards is the trained classifier applied to new data – that is, there is a strict separation between learning and application. However, in the present context, an incremental and interactive approach to machine learning (e. g., Fails & Olsen, 2003) is necessary to make Dare2Del adaptive and take user feedback into account. Therefore, the end user programming approach (trying to identify user intentions from observable behavior; Demolombe & Fernandez, 2005; Gulwani, 2011) is adapted within Dare2Del for hiding and filtering operations.

As mentioned above, the cognitive companion Dare2Del should be transparent to the user by providing explanations for decisions, should include the opportunity to exploit predefined rules, and should be able to adapt to users’ needs and preferences (Siebers et al., 2017). The system must also be context-aware in inferring when it would be appropriate to support the user. However, there is still an open question of when individuals need support to intentionally forget at work. The literature review has shown that there is sparse research on the question of why and when individuals are motivated to forget, in particular in the work context. Moreover, the widely used experimental paradigms on intentional forgetting require that participants are instructed to intentionally forget by the experimenter. In contrast, in daily working life, participants have to take the initiative to form the intention to forget rather than being asked to do so by another person.

Implications for Cognitive Companion Systems

In addition to the obvious function of facilitating concentration on the task at hand by deleting, archiving, or hiding irrelevant digital information, a cognitive companion such as Dare2Del can also support emotion regulation, maintaining concentration, goal attainment, and a positive self-image. In cases where digital objects such as emails and photos are related to negative emotions or negatively affect one’s relationships with others, archiving, deleting, and ignoring can also serve to regulate one’s emotions (see also Streb et al., 2016) and protect social relationships (Anderson & Hanslmayr, 2014). The present critical incident study also showed that workers reported intentional forgetting when they had time to do so (e. g., during a break), when there was time pressure, or when a new or complex task requiring increased attention started. Thus, the companion system should support forgetting especially in situations with higher cognitive load, when cognitive demands exceed individuals’ cognitive capacities to deliberately suppress irrelevant information. Moreover, as users have the possibility of configuring when the companion system shall propose irrelevant files, they can indicate times when there is little to do.