Personal memory: Is it personal,is it memory?

We are to a large extent the product of our memory. With the exception of the fleeting instant captured in the present, we continuously re-enact the past and imagine the future by relying on our recollections. As such, we would like to think that our memory is a faithful representation of our individual experiences, weaved into our private mental literature (borrowing a metaphor attributed to Aldous Huxley). However, a large body of evidence from both cognitive psychology and brain research challenges this assumption; memory, so it appears, does not necessarily convey an accurate or even solely personal representation of reality (Hirst and Echterhoff, 2012; Loftus, 2005; Schacter et al., 2011; Schacter and Loftus, 2013; Stone et al., 2012).

To understand this circumstance, a brief excursion into the brain’s mechanisms of memory formation, storage, and retrieval is warranted. The encoding of experience does not necessarily culminate in long-term memory. Some experiences may leave only a fleeting trace and fade into oblivion. For lasting memories to become extant, encoded information is believed to undergo a process of consolidation, which is a post-acquisition stabilization of memory (Dudai, 2004, 2012; McGaugh, 2000; McKenzie and Eichenbaum, 2011; Müller and Pilzecker, 1900). During the consolidation process, the memory trace is protean and sensitive to external influences. This can be demonstrated experimentally by exposing humans or animals to interfering agents during the time window of consolidation (Dudai, 2004; McGaugh, 2000). In humans, these interferences are commonly salient sensory distractors and new information, while in animal models, pharmacological agents that inhibit macromolecular synthesis are particularly effective. Once consolidation is over, these interferences lose their potency—with one critical exception. When the memory is retrieved, it seems to re-enter a transient phase in which it again becomes susceptible to the same amnesic agents that were effective in the original consolidation window (Dudai, 2012; Nader et al., 2000; Sara, 2000). This process is dubbed “reconsolidation.” One interpretation is that this allows the individual to flexibly adapt to changes in the environment by not maintaining an overly rigid memory representation (Schacter et al., 2011), and promotes social coherence by mnemonic convergence (Hirst and Brown, 2011).

A ubiquitous source of new information is the social milieu. No other species create social structures with the complexity demonstrated in humans (Byrne and Whiten, 1989; Dunbar and Shultz, 2007; Haun et al., 2013) and humans are, for the most part, extremely social by nature (Allen, 1965; Asch, 1951; Byrne and Whiten, 1989; Dunbar and Shultz, 2007; Kelman, 1961; Sherif, 1936). We usually live in large and organized societies, follow norms, and are influenced by our fellows in the selection of food, consumer goods, friends, and a myriad of other everyday choices and decisions (Boyd and Richerson, 1988; Byrne and Whiten, 1989; Dunbar and Shultz, 2007; Haun et al., 2013; Hirst and Echterhoff, 2012; Izuma, 2013; Kelman, 1961; Roediger et al., 2001; Ruff and Fehr, 2014; Simonsen et al., 2014; Zhu et al., 2012).

In such environs, inter-personal information is potentially a highly informative source of evidence that goes beyond the sensory limitation of the individual. For example, mathematical models demonstrate that in stable environments it can be more efficient to learn from others than from your own limited personal experience (Boyd and Richerson, 1988, 2004; Haun et al., 2013). Given these factors, it is not surprising that humans are sensitive to social cues and information. In fact, the development of complex social structures may be one of the main driving forces shaping the evolution of the human brain (Byrne and Whiten, 1989; Dunbar and Shultz, 2007).

Ample experimental evidence has established that information from an inter-personal source can be effective in influencing memories (Gabbert et al., 2006; Meade and Roediger, 2002; Roediger et al., 2001). For example, personal recollections can be shaped by conversation, creating a collective memory shared by several individuals (Hirst and Echterhoff, 2012; Stone et al., 2012). It is of note that such influence can impact not only what we remember but also what we forget via the omission of information from the conversation (Stone et al., 2012). Practical consequences of these effects have been studied in regard to how they influence eye-witness testimonies (Chambers and Zaragoza, 2001; Echterhoff et al., 2005; Loftus, 2005; Schacter et al., 2011; Wright et al., 2009). Moreover, cultural and societal factors may participate in this process of reworking memory, adding an additional layer of complexity (Wang, 2016). These influences can span generations by affecting the storytelling narrative passed on in families and larger social groups (Fivush and Merrill, 2016).

In a series of experiments, we attempted to further elucidate the processes by which social influence affects personal memory, and in particular, identify the brain systems that support mnemonic flexibility (Edelson et al., 2011, 2014a, 2014b). We based our behavioral design on established experimental protocols taken from conformity and social-influence research (Allen, 1965; Asch, 1951; Deutsch and Gerard, 1955; Kelman, 1961; Sherif, 1936), and adapted these to the context of episodic memory encoding and retrieval. The protocols were further adapted for a functional brain-imaging environment, referencing previous work on the neural correlates of social influence (Berns et al., 2005; Klucharev et al., 2009) and memory investigations using naturalistic films to elicit mnemonic processes (Ben-Yakov and Dudai, 2011; Furman et al., 2007; Hasson et al., 2008).

We used a four phase protocol spanning a 2-week period (Figure 1). Initially, participants viewed a film in small social groups ranging between five and eight participants. The film was an audiovisual documentary following the complex life of illegal immigrants, and was chosen because of its naturalistic content. After 3 days, participants were asked to individually answer questions regarding the content of the film (Test 1, Figure 1). On the seventh day, participants performed the same test again in a functional magnetic resonance imaging (fMRI) brain scanner. This time, before answering each question, they were exposed to answers that their fellow group members supposedly provided (Test 2, Manipulation phase, Figure 1). In a subset of the questions, for which the target participant had initially a confident correct answer (as identified by Test 1), the answers provided by all fellow group members were made to be incorrect. In the rest of the questions, the answers provided by the fellow group members were manipulated so as to support the participant’s initial answer and confidence. These latter questions served to enhance the credibility of the protocol.

OPEN IN VIEWER

Figure 1.

(a) Experimental outline. Participants viewed a movie in small social groups and subsequently performed three memory tests individually. Test 1 assessed the participants’ initial memory and confidence before the manipulation phase (Test 2). The correction phase (Test 3) served to identify memory errors that persisted after the social manipulation was removed. The manipulation phase had three different experimental conditions: the Manipulation condition, in which all provided answers were incorrect; the No-Information condition, in which the letter X was displayed instead of answers, and the Credibility condition, in which variable patterns of answers were displayed. The source of the external information (co-observers/computers) was manipulated in separate experimental groups. (b) “Change of mind” flowchart model, spanning from an incorrect answer with high confidence to a correct answer with high confidence. As illustrated, shifting from a low confidence erroneous judgment (Xi) to a low confidence correct judgment (Xj; case I) represents less of a change than shifting to a high confidence correct judgment (Xk; case II). Furthermore, a subject may maintain the same answer but alter his/her confidence in that answer (i.e. case III; move from Xj to Xk), revealing a change in the strength of the judgment.

In a final phase of the experiment, performed in the fMRI scanner 2 weeks after the initial viewing, the veracity of the information provided by the fellow group members was discredited. Participants were informed that the answers given by their co-observers were in fact determined randomly and are thus irrelevant (rather than incorrect). Participants were then asked to answer the test questions again (Test 3, Correction Phase, Figure 1). This last phase was performed in order to insure that no normative motivation remained to conform (Asch, 1951; Kelman, 1961; Smith and Mackie, 2007). If participants really modified their long-term memories to include the erroneous information provided by their fellows, this effect would persist at this stage (persistent errors) (Allen, 1965; Kelman, 1961; Reysen, 2005; Smith and Mackie, 2007). If on the other hand the erroneous information will no longer have an effect, it is hard to claim that a long-lasting memory change had indeed occurred (transient errors).

In line with previous behavioral studies, we found that in our episodic memory protocol, participants’ initial responses were in fact very susceptible to social information (Edelson et al., 2011). Indeed accuracy levels dropped in the manipulation questions from 100% accuracy to around 30% when participants were presented with a group consensus conflicting with their original correct answer. Importantly, in 40% of these errors (i.e. ~25% of all manipulated questions), long-lasting errors (that persisted 7 days later in Test 3) were created. Using functional brain imaging it was possible to examine the processes and brain structures being activated during the manipulation stage and to test whether these predicted effects were later evident in behavior (Wagner et al., 1999).

We found that the creation of persistent errors was correlated with activation and connectivity between two regions in the mediotemporal lobe of the brain. These are the hippocampus and the amygdala. A large body of evidence links hippocampal function to mnemonic processing and this region is considered to play an important role in memory encoding and some forms of retrieval (Morris, 2007; Scoville and Milner, 1957; Squire and Wixted, 2011). For example, lesions in the hippocampus and the surrounding cortex lead to dense “global” amnesia that is loss of the ability to store newly acquired declarative memories for a period longer than a few minutes. As for amygdala, it has also been associated with the acquisition and storage of some forms of memory, as well as salience detection and the processing of emotional and social information. (Adolphs, 2003; Dolan and Vuilleumier, 2003; Klüver and Bucy, 1937; Nader et al., 2000; Phelps, 2004). Furthermore, amygdala activity has also been linked to the modification of memories via influence on hippocampal circuitries (Dolan, 2002; Phelps, 2004). In our experiment, enhanced activity in the hippocampus itself was also found when the external influence was presented as provided by an inanimate source (i.e. computer algorithms, Figure 1(a)), but, the enhanced activity and the connectivity of the amygdala was unique to the social information. Moreover, the amygdala activity in this paradigm was not predicted by the participant’s emotional state assessed after each trial. These findings support the “extended common currency” view (Ruff and Fehr, 2014), by which social and non-social information may initially engage different brain areas (such as the amygdala) but that these signals ultimately converge and are processed by identical circuitries (in the case of episodic memory in the hippocampus).

It is of note that in 60% of the events in our studies, participants were able to subsequently correct socially induced memory errors when the source of the misinformation was later discredited. Our data indicate that this was not due to an additional change in mnemonic representations in mediotemporal circuits. What then determines if the subjects will correctly select their original memory over the misleading information?

Influential behavioral and neurocomputational models contend that when individuals are faced with several options, evidence is accumulated for each option until reaching a threshold that leads to a decision (De-Martino et al., 2013; Vickers, 1970). The option that first reaches this criterion is selected and the confidence in this option correlates with the difference in evidence between the selected and non-selected alternatives (De-Martino et al., 2013). Thus, confidence and response modifications may be different representations of a shift in one postulated parametric space (for three different examples see Figure 1(b)). In the current case, it can be assumed that evidence for the original internal representation of the experience competes with evidence for the representation acquired following the social influence. At one extreme of the spectrum, one can position the original answer with maximal confidence and at the other extreme the socially endorsed answer with maximal confidence (Figure 1(b)). When the external environment changes and evidence from an option is considered more credible or informative, the evidence accumulation can be biased towards that option (e.g. by a larger computational weight on the accumulated evidence for this option). The result is a shift in position along a postulated “change of mind” axis (Figure 1(b)). Such changes will always result in modifications to confidence but only when the watershed point between options is crossed will this change result in a modification of the response (see case I and case II vs. case III in Figure 1(b); for a review of the relationship between confidence and accuracy, see Roediger et al., 2012).

Using these formal principles, we next studied the brain activity during the Correction Phase (Test 3, Figure 1(a)) of our protocol (Edelson et al., 2014a). We found that activity in a frontal brain area, the anterior lateral prefrontal cortex, correlates with the size of the position shift, per event, on the “change of mind” axis. This brain area has been previously reported to participate in processes related to the selection between competing memory traces and reversal learning, and shows enhanced engagement in response to social cues (Christoff and Gabrieli, 2000; Koechlin and Hyafil, 2007; Mendelsohn et al., 2008; Mitchell et al., 2009; Rushworth et al., 2011). Further, in our studies, the activity in this brain region was inversely correlated with the activity in the amygdala during the initial social influence (Test 2, Figure 1(a)).

Taken together, the results of our research program reviewed above, suggest that the exposure to social influence can provide an opportunity for memory change. The process probably involves reconsolidation of the reactivated old information immediately following its confrontation with the new, conflicting information provided by the social influence, or the activation of other mechanisms that can alter existing memories. Further, the process may be specifically embodied in the amygdala–hippocampal circuitry. Evidence of the original representation does remain (whether in the form of details in a revised trace or as a weakened original trace). The relevance of this evidence can later increase or decrease depending on the dynamic nature of the social environment. This fact can be utilized by frontal high-order (sometimes termed as metacognitive) circuits to provide an additional degree of flexibility. These frontal and mediotemporal systems may operate adaptively in tandem since the stronger the initial modification (as reflected by amygdala and hippocampal co-activation), the lesser is the frontal engagement subsequently observed.

If amygdala activity plays a role in the modification of memory due to the input of social information, reducing amygdala activation should result in the reduction of these long-lasting memory modifications. Toward that end, we administered the neuropeptide oxytocin, which has been found to decrease amygdala activation in male participants (Domes et al., 2007; Striepens et al., 2011). This was done by administering 24 International Unit (IU) of the neuropeptide intranasally in the social manipulation stage (Figure 1(a)). Oxytocin administration was indeed found to reduce the robustness of long-lasting socially induced memory errors (Edelson et al., 2014b). This effect was accompanied by an increase in transient errors during the manipulation phase (Test 2), possibly due to the neuropeptide’s reported properties relating to enhanced affiliation (Feldman, 2012).

Our studies are hence in line with the conclusion that our personal memory is in fact only partially personal. The influence of social information on the mutable nature of memory may afford an evolutionary advantage, allowing rapid adjustment of past representations to accommodate for changes in the present environment. Flexible brain systems seem to provide humans with multiple levels of adjustment depending on whether new social information is available or the general contextual setting has changed. This sensitivity of memory to inter-personal information may also serve to ensure that the accumulation of memory over time far exceeds the capacity and life span of the individual brain. Such mechanisms may operate conjointly with other cultural, historical, and socialization processes and interact with the physical availability of alternative historical information (Assmann, 2008; Schwartz, 1982).

Our concept of memory has shifted from a metaphor of a private library capturing our individual pasts, or private literature, to a multi-node network serving different functions. These functions may not necessarily recognize veracity as a prime objective. For example, the important role of the memory system in the assimilation of culture or norms (Wang, 2016) and in future planning and imagination (Dudai and Carruthers, 2005; Schacter et al., 2007; Schacter and Madore, 2016), actually could be considered to favor a degree of “error,” “noise,” or deviance from an accurate and objective personal history. As noted, these deviations may provide our species with evolutionary advantages. Rapid technological development in recent years raises intriguing questions in this context. It is already evident that external devices (such as smartphones and computers) can store a tremendous amount of objective personal information. How the presence of objective personal libraries in our hands will affect both the common and the scientific concept of memory is yet to be seen.