A characteristic feature of episodic memories, as pointed out by Mahr & Csibra (M&C), is that they are epistemically generative, rather than epistemically preservative: An episodic memory may generate extra epistemic justification for beliefs without being a belief itself. In this respect, episodic memories are analogous to perceptions, which are not beliefs either, even though their contents often ground beliefs, but sometimes also conflict with what is believed (e.g., in known perceptual illusions). Semantic memories (including non-episodic memories of particular events), in contrast, consist of beliefs that merely preserve their justificatory status within the belief system. When a 46-year-old asks himself what happened during the fall of the Berlin Wall almost three decades ago, a justification for him to form the belief that people were dancing on the wall may derive from a mnemonic simulation of that scenario. This simulation may draw on sensory as well as agentive and emotional information from episodic memory traces that were grounded in some of his experiences at the wall on November 9, 1989 (Cheng et al. Reference Cheng, Werning and Suddendorf2016). A 20-year-old may also be justified to believe that people were dancing, but on the basis of recalling what she learned about the fall of the wall from other people or media. In her case, no new belief is formed, no extra justification gained. Another, closely related feature in need of explanation is why episodic memories have an epistemic status privileged over other sources of knowledge (although not infallible).
To account for the epistemic generativity of episodic memory, M&C postulate that subjects maintain a certain metarepresentational attitude toward the contents of their episodic memories – self-directed epistemic vigilance involving source monitoring. Through such a metarepresentational attitude, they hold, subjects are justified that the event information was obtained firsthand, that is, through their own experience. This “autonoetic” aspect privileges episodic memories over other sources of knowledge. Building on our notion of scenario construction as mnemonic simulation (Cheng & Werning Reference Cheng and Werning2016; Cheng et al. Reference Cheng, Werning and Suddendorf2016), we propose an alternative view: It is the way an event of the past is presented in episodic memory recall rather than a metarepresentational attitude of the subject that explains the epistemic generativity and privileged status of episodic memory and distinguishes it from semantic memory.
To M&C one may object that, even though semantic memory (e.g., “it takes me 40 minutes to drive to work”) fails to be epistemically generative, one may still assume a metarepresentational attitude toward its content. A subject may, for example, have the (justified) belief that the memory content is based on repeated experiences (“my commute has always taken me 40 minutes”). The subject may well represent a memory's source even though the memory is non-episodic. As a matter of intersubjective epistemic vigilance, the subject may even be able to recur to this source information when asked for a justification in communication. The capacity to assume a metarepresentational attitude toward a memory's content as a means of epistemic vigilance is thus not distinctive for episodic memory and, hence, not sufficient to explain its epistemic generativity and privileged status.
Moreover, if epistemic vigilance was the crucial function of autonoetic consciousness in episodic memory, as M&C postulate, then we would expect that there would have been an evolutionary pressure for humans to develop a measure for gauging the accuracy of their episodic memories. That is, the rememberer should be more confident of the content of her belief for those memories that are likely accurate than for those that are not, because epistemic vigilance is essential in social communication, and the confidence of the rememberer is an important source of information for the recipient to gauge the accuracy of socially communicated beliefs derived from episodic memories. We would, therefore, expect the self-assessed confidence in our own episodic memories to be strongly predictive of the accuracy of our memories. However, there is ample empirical evidence that contradicts this prediction. A number of studies suggest that even for highly memorable events, so-called flashbulb memories, the self-assessed confidence in a memory either has no, or only a very weak, correlation with its accuracy (Neisser & Harsch Reference Neisser, Harsch, Winograd and Neisser1992; Talarico & Rubin Reference Talarico and Rubin2003).
Let us now briefly sketch an alternative explanation. According to our view, in episodic memory recall a scenario of the event in question is mentally constructed such that the following holds:
-
(1) Information contained in episodic memory traces is combined with semantic information – prototypical features are filled in and need not be stored.
-
(2) Memory traces contain only physical “gist” information and are not conceptual mental representations themselves. They are, in particular, not compositional (Werning Reference Werning, Werning, Machery and Schurz2005) or systematic (McLaughlin Reference McLaughlin2009). A compositional representation is generated only during recall (Cheng & Werning Reference Cheng and Werning2013).
-
(3) Memory traces constitute a reliable causal link between experience and recall (Cheng & Werning Reference Cheng and Werning2016; Werning & Cheng Reference Werning, Cheng, Bernecker and Michaelian2017).
-
(4) The resulting scenarios are typically perspectival – which means more than just exhibiting a sensory viewpoint: they also involve an action direction and emotional stance (Russell & Hanna Reference Russell and Hanna2012).
-
(5) Scenarios are often vivid and in various respects resemble (but also differ from) the contents of perceptions.
First, from (2) it immediately follows that episodic memory traces are not beliefs because beliefs are conceptual mental representations. Only after recall is a hitherto non-existing belief formed. Second, the constructed scenario may serve as the source of epistemic justification. This is because, presupposing (3), scenario construction is – in a statistical sense – an epistemically reliable process (Goldman Reference Goldman1986). Episodic memory is hence epistemically generative. The fact that the constructed scenario is (typically) perspectival – following (4) – and in its vividness (often) resembles the contents of perceptions – following (5) – makes it finally likely that the presented event was experienced firsthand. This is a probabilistic relation, and the typical phenomenology of scenarios just serves as an indicator for its authenticity. This does not preclude that semantic memories may also in certain cases possess a perspective and some vividness. The epistemic privilegedness of episodic memory – as well as its epistemic generativity – is thus more a matter of statistics than of principle.
A characteristic feature of episodic memories, as pointed out by Mahr & Csibra (M&C), is that they are epistemically generative, rather than epistemically preservative: An episodic memory may generate extra epistemic justification for beliefs without being a belief itself. In this respect, episodic memories are analogous to perceptions, which are not beliefs either, even though their contents often ground beliefs, but sometimes also conflict with what is believed (e.g., in known perceptual illusions). Semantic memories (including non-episodic memories of particular events), in contrast, consist of beliefs that merely preserve their justificatory status within the belief system. When a 46-year-old asks himself what happened during the fall of the Berlin Wall almost three decades ago, a justification for him to form the belief that people were dancing on the wall may derive from a mnemonic simulation of that scenario. This simulation may draw on sensory as well as agentive and emotional information from episodic memory traces that were grounded in some of his experiences at the wall on November 9, 1989 (Cheng et al. Reference Cheng, Werning and Suddendorf2016). A 20-year-old may also be justified to believe that people were dancing, but on the basis of recalling what she learned about the fall of the wall from other people or media. In her case, no new belief is formed, no extra justification gained. Another, closely related feature in need of explanation is why episodic memories have an epistemic status privileged over other sources of knowledge (although not infallible).
To account for the epistemic generativity of episodic memory, M&C postulate that subjects maintain a certain metarepresentational attitude toward the contents of their episodic memories – self-directed epistemic vigilance involving source monitoring. Through such a metarepresentational attitude, they hold, subjects are justified that the event information was obtained firsthand, that is, through their own experience. This “autonoetic” aspect privileges episodic memories over other sources of knowledge. Building on our notion of scenario construction as mnemonic simulation (Cheng & Werning Reference Cheng and Werning2016; Cheng et al. Reference Cheng, Werning and Suddendorf2016), we propose an alternative view: It is the way an event of the past is presented in episodic memory recall rather than a metarepresentational attitude of the subject that explains the epistemic generativity and privileged status of episodic memory and distinguishes it from semantic memory.
To M&C one may object that, even though semantic memory (e.g., “it takes me 40 minutes to drive to work”) fails to be epistemically generative, one may still assume a metarepresentational attitude toward its content. A subject may, for example, have the (justified) belief that the memory content is based on repeated experiences (“my commute has always taken me 40 minutes”). The subject may well represent a memory's source even though the memory is non-episodic. As a matter of intersubjective epistemic vigilance, the subject may even be able to recur to this source information when asked for a justification in communication. The capacity to assume a metarepresentational attitude toward a memory's content as a means of epistemic vigilance is thus not distinctive for episodic memory and, hence, not sufficient to explain its epistemic generativity and privileged status.
Moreover, if epistemic vigilance was the crucial function of autonoetic consciousness in episodic memory, as M&C postulate, then we would expect that there would have been an evolutionary pressure for humans to develop a measure for gauging the accuracy of their episodic memories. That is, the rememberer should be more confident of the content of her belief for those memories that are likely accurate than for those that are not, because epistemic vigilance is essential in social communication, and the confidence of the rememberer is an important source of information for the recipient to gauge the accuracy of socially communicated beliefs derived from episodic memories. We would, therefore, expect the self-assessed confidence in our own episodic memories to be strongly predictive of the accuracy of our memories. However, there is ample empirical evidence that contradicts this prediction. A number of studies suggest that even for highly memorable events, so-called flashbulb memories, the self-assessed confidence in a memory either has no, or only a very weak, correlation with its accuracy (Neisser & Harsch Reference Neisser, Harsch, Winograd and Neisser1992; Talarico & Rubin Reference Talarico and Rubin2003).
Let us now briefly sketch an alternative explanation. According to our view, in episodic memory recall a scenario of the event in question is mentally constructed such that the following holds:
(1) Information contained in episodic memory traces is combined with semantic information – prototypical features are filled in and need not be stored.
(2) Memory traces contain only physical “gist” information and are not conceptual mental representations themselves. They are, in particular, not compositional (Werning Reference Werning, Werning, Machery and Schurz2005) or systematic (McLaughlin Reference McLaughlin2009). A compositional representation is generated only during recall (Cheng & Werning Reference Cheng and Werning2013).
(3) Memory traces constitute a reliable causal link between experience and recall (Cheng & Werning Reference Cheng and Werning2016; Werning & Cheng Reference Werning, Cheng, Bernecker and Michaelian2017).
(4) The resulting scenarios are typically perspectival – which means more than just exhibiting a sensory viewpoint: they also involve an action direction and emotional stance (Russell & Hanna Reference Russell and Hanna2012).
(5) Scenarios are often vivid and in various respects resemble (but also differ from) the contents of perceptions.
First, from (2) it immediately follows that episodic memory traces are not beliefs because beliefs are conceptual mental representations. Only after recall is a hitherto non-existing belief formed. Second, the constructed scenario may serve as the source of epistemic justification. This is because, presupposing (3), scenario construction is – in a statistical sense – an epistemically reliable process (Goldman Reference Goldman1986). Episodic memory is hence epistemically generative. The fact that the constructed scenario is (typically) perspectival – following (4) – and in its vividness (often) resembles the contents of perceptions – following (5) – makes it finally likely that the presented event was experienced firsthand. This is a probabilistic relation, and the typical phenomenology of scenarios just serves as an indicator for its authenticity. This does not preclude that semantic memories may also in certain cases possess a perspective and some vividness. The epistemic privilegedness of episodic memory – as well as its epistemic generativity – is thus more a matter of statistics than of principle.