Although there were some books on philosophy of memory written in the past century, both the previous and the coming decades will be remembered for a revival of the production of books that helped to deepen the philosophical analysis on this often neglected topic. Chris McCarroll’s Remembering from the Outside (OUP, 2019) will certainly be one of those works remembered. His aim is in certain way modest: unlike those who have a Kantian spirit, he does not intend to give a complete account neither of memory nor of one of its classical sub-kinds, episodic memory. He exclusively focuses on the study of one very specific memory phenomenon: “observer memories”, that is, visual memories where one sees oneself in the remembered scene from an external perspective, from-the-outside. But it is precisely the specificity of his scope that allows him to present an in-depth and detailed analysis that would not be possible otherwise, by deploying a vast range of insights from philosophical reasoning, scientific references, and examples coming from visual arts and literature.

In this short piece, I will focus on two main lines of criticism to McCaroll’s ideas because this is the purpose of a critical notice, without implying that the quality of this marvelous work is dubious¹. On the contrary, it is clearly exceptional, and will be of interest not only to memory researchers but also to those engaged in the study of visual imagery and the perspectival mind.

The aim of McCarroll’s work is to understand the nature of observer memories by moving beyond the traditional and common sense conception that memories of this kind are less accurate, faithful, and genuine than their counterpart: visual field memories, that is, memories in which the remembered scene is seen as one originally experienced it, from one’s original point of view. As McCarroll explains, field memories seem more authentic because they preserve, in principle, the original visual perspective, whereas observer memories do not. Using Bernecker’s (²⁰¹⁰) distinction between memory truth and memory authenticity, we could believe that despite observer memories being true —in the sense that they depict an event that actually happened—, they are not authentic because they differ from the past visual representation that was not from-the-outside. The observer memory representation would not only present the event remembered from a perspective that was not part of the original experience, but would also present an additional content: a representation of oneself experiencing the event. That is why observer memories would be less accurate, faithful, and thus genuine than field memories.

In order to deal with this line of reasoning and dethrone the traditional and common sense conception of observer memories, McCarroll makes two major moves in his work. The first one consists in showing that observer and field perspectives are not really mutually exclusive, and can be both present during an experience, while encoding it and also remembering it. Multiple sources of information, modalities and perspectives are in fact available during an experience, and this multiplicity is reflected in the construction of memories of that experience. The defense of the richness of our experiences and memories takes up most of the chapters of his book (chapters 2, 3 and 5). In his second move, which is mostly developed in chapter 6 (but also to some extent in chapter 4), McCarroll argues that in observer memories there is no additional explicit representation of oneself experiencing the event: self-presence is transparent, identification-free and given by the mode of presentation. The content—in the sense of the intentional object—would be the same in field and observer memories, so the former would not be more genuine than the latter.

The first line of argument is quite convincing but not unproblematic, as I show next. Against the traditional and common sense conception of observer memories, it is possible to argue that memory retrieval is reconstructive, which means that multiple causes produce memory content at the time of recall and hence memory representations generally differ from past perceptual representations. Within this framework, both observer and field memories would have a reconstructive nature. Arguments in favor of reconstruction are not really needed because in the last couple of decades, science has provided massive evidence in favor of this thesis at all levels of explanation (see for example ^{Schacter & Addis, 2007}). But McCarroll warns readers about the efficacy of this strategy²: it could still be argued that observer memories are more reconstructive than field memories because along with the reconstructive content, they present an anomalous perspective. That is why McCarroll adopts a different tactic and develops the Constructive Encoding approach, which is, in my opinion, one of the most original parts of his work. According to this proposal, the content that is available during the encoding of an experience is rich: it is not limited to sensory data, but includes multimodal information (such as kinesthetic and affective information) and —most importantly for McCarroll’s purposes— multiperspectival information. During an experience, information does not necessarily have to be egocentric: all modalities can present information from an allocentric point of view, that is, from a frame of reference different from oneself or one’s body. This is also the case for spatial and visual imagery. The three main examples that McCarroll introduces to give credit to this claim are very convincing: (a) the allocentric cognitive maps that are available during perception and allow us to navigate the spatial environment (^{O’Keefe & Nadel, 1978}); (b) amodal perception, that is, the capacity we humans have of perceiving the whole object through mental imagery, although only parts of the object available from a particular visual perspective affect the sensory receptors; (c) and cross-modal informational translation, such as the possibility of generating (observer) visual imagery in the absence of visual perception through information coming from other modalities, like kinesthesia. Therefore, because visual allocentric information is in general present during our experiences and the process of encoding them, observer memories are not more reconstructed and less authentic than field memories.

The Constructive Encoding approach is quite a compelling and original thesis that is backed up with scientific data. It is not itself problematic, but raises questions about the need of such a thesis and its consequences for the general framework that McCarroll intends to defend. In fact, the original argument that criticizes the authenticity of observer memories only makes sense under certain conceptualization of memory. If genuine memories need to preserve all or some significant part of the content of the original representation, observer memories are clearly less genuine than their field counterparts. The preservation of content is a necessary condition in classical and neo-classical causal theories of memory (^{Martin & Deutscher, 1956}; ^{Bernecker, 2010}; for a categorization of theories of memory, see ^{Michaelian & Robins, 2018}). But the preservation of content is not a requirement in all memory accounts. In a functionalist framework that defines a memory state according to the role it plays in our cognitive architecture, observer and fields memories could be both equally genuine³. This is also the case in Michaelian’s (²⁰¹⁶) simulationist account of memory, which considers memories as simulations of past episodes not different in nature from future projections and other kinds of imaginative simulations. According to Michaelian’s proposal, observer memories are not less genuine than observer memories: both are the result of a simulation operated by the episodic construction system, which “makes its best guess” as to what the scene must have looked like from one perspective or the other (^{Michaelian, 2016}, p. 137). Even a (causal) theory of memory that adopts a distributed conception of memory traces (^{Sutton, 1998}) considers that field memories are not more fundamental and intrinsically tied to reality than observer memories: both are the product of a construction and both can be accurate and genuine (^{Sutton, 2010}). As Sutton explains, the thought that “construction entails distortion only makes sense against a background assumption that genuine personal memory must replay or archive the past in an exact copy of an original experience” (^{Sutton, 2010}, p. 33). Consequently, although the Constructive Encoding approach can be considered as a truthful account of our experience and our encoding processes, its introduction to prove that observer memories are authentic memories is only needed in a preservationist framework. Functionalist and radical constructivist theories of memory do not need this approach to accept the authenticity of observer memories. In fact, not the approach itself, but the idea that observer perspectives are genuine because they maintain the external perspective that one adapted at the time of the event⁴ is plainly incompatible with the theories mentioned above. In conclusion, the real enemy that McCarroll challenges with the Constructive Encoding approach is traditional preservationism; that is why the conception of memory that he implicitly assumes in order to give a full memory status to observer memories is also a kind of preservationism. It could be called “quasi-preservationism” because McCarroll allows memory to incorporate new content that was not included in the original experience, whereas a strict preservationism considers that a memory representation does not include content not included in the original representation. Nonetheless, in his account, preservation of some content coming from the original experience is essential but generation and construction of content at retrieval are not. Hence the label “quasi-preservationism”. This categorization certainly goes against McCarroll’s explicit endorsement of a constructive and active nature of memory, yet it inevitably follows from the prominent place given to the Constructive Encoding approach in his defense of the authenticity of observer memories.

Therefore, a quasi-preservationist conceptualization of memory is a consequence of the Constructive Encoding approach. But there is another undesirable consequence for the main picture of memory that McCarroll’s wants to depict. Because of the important role played by the Constructive Encoding approach to save the genuineness of observer memories, the notion of accuracy and faithfulness of memory representations becomes inevitably associated with the preservation of content. McCarroll could have done a different move to avoid this supposition. For example, he could have tried to dissociate construction and post-event generation of content from distortion and falsehood by arguing that field memories are also the product of constructive processes which may include post-event information or by showing that in many cases new content is needed in order to construct a faithful and accurate memory of the past. Although the latter strategy seems more evident in memories with a narrative form (see ^{Campbell, 2006}; ^{Habermas, 2012}), it can probably be extrapolated to visual memories with little difficulty. While remembering traumatic experiences such as physical violence or murder, a field memory that is focused on a salient feature of the traumatic event—such as the weapon—to the detriment of other important details, contains less information and thus is less faithful to the past event than an observer memory that presents a visualization of the entire scene from a bird’s eye view.

Furthermore, the Constructive Encoding approach and the idea that we may encode memories from-the-outside is in tension with the hypothesis of chapter 5, which proposes the existence of a plurality of perspectives at retrieval. This plurality is understood in different ways. First, McCarroll argues that there is no necessary correlation between different perspectival modalities. Visual, kinesthetic and emotional imageries are dissociated, so an observer visual memory can be coupled with an internal emotional perspective, and vice versa. Second, a memory of a single event may present multiple perspectives of the same modality: in the case of visual imagery, the same memory may shift between field perspective and different observer perspectives. Third, perspectival boundaries may even be blurred, and a memory may be recalled from both a field perspective and an observer perspective simultaneously.

Whereas this thesis, which may be called “Plurality of Perspectives at Retrieval”, presents a very interesting and novel approach to understand memory construction, it also raises some questions. First, the idea of internal and external emotional imagery is itself problematic, especially as it is treated in chapter 5. Kinds of emotional perspectives do not seem to be similar to kinds of visual perspectives. Whereas observer or external visual memories refer to memories presented from an unoccupied point of view rather than from the point of view of some other person (thesis defended by McCarroll in chapter 4), external emotional memories are never from an unoccupied point of view and rarely from the point of view of some other person. Most of the time, the “from-the-outside” of an observer emotional memory must be interpreted as from-the-outside of the past self who experienced the event, but is still internal to the subject who remembers, and thus from-the-inside. It can refer to the way I appraise now a past event, or to the way I appraised the past event in some time x after the event. So the terms “internal” and “external” do not mean the same when applied to emotional perspectives or to visual perspectives, and the common and indiscriminate use of these terms to talk about perspectives of different nature seems misleading. What is more, the emphasis on the plurality of perspectives in distinct modalities hides the fact that there are some interactions between these different perspectives and that their combination is not merely accidental. McCarroll mentions Valenti, Libby & Eibach’s work (²⁰¹¹) on observer perspective imagery and its relation to regret for actions and inactions, but he does not present a survey and analysis of the rich empirical literature on the interaction between visual and emotional perspectives, which is specially relevant to understand and treat clinical conditions such as traumatic and intrusive memories (^{Holmes & Matthew, 2010}). A reader interested in empirical research may feel that a section like the one proposed is missing in a chapter focused on the plurality of perspectival modalities.

Secondly, the thesis about the plurality of perspectives at retrieval is in tension with the Constructive Encoding approach, which states that memories may be encoded from-the-outside, and also with an idea that McCarroll develops in chapter 6: that the perspective is given to a sort of “perspectively neutral” memory content by the mode of presenting that content at retrieval. So there are three different theses that seem mutually incompatible between them: (a) the observer perspective of memories is determined at encoding; (b) the observer perspective of memories is determined at retrieval; (c) visual perspectives are multiple and even simultaneous in a memory so there is no pure observer memory or pure field memory. Unfortunately, McCarroll does not explain how these different theses may become compatible or not, but suggests in the last page of the concluding chapter (p. 198) that there are different kinds of observer memories, those that are the product of a switch of perspective operated at retrieval, and those that originated in observer perspective experiences.⁵ If the authenticity of observer memories is grounded in the fact that observer perspectives are available during the encoding of the experience, as the Constructive Encoding approach suggests, it is natural to wonder if memories that are “converted” to observer memories at retrieval are more reconstructed and so less authentic than those originated in observer perspective experiences. The Constructive Encoding approach does not account for the authenticity of “converted” observer memories, so McCarroll should have said something more about it. If McCarroll’s original idea was to propose the existence of different kinds of observer memories, it is a pity that he did not develop it in detail, maybe in a specific chapter devoted to different kinds of observer memories.

On the other hand, the thesis (c) poses other kinds of problems. As McCarroll explains, the variations in point of view during a single memory experience are well documented in empirical literature. In fact, they account for the dynamic and ever-changing stream of our consciousness. But the simultaneity of perspectives in a single “frozen snapshot” of a visual memory is a much stronger thesis and the different examples mentioned by McCarroll to support it (Cubism, dual viewpoint in gesture, multiperspectival imagery in comics) are not very convincing. On one hand, if multiperspectival visual memories are a kind of visual memories among others, this new type would add more complexity to the already complex picture suggested by McCarroll, and would have certainly deserved more explanation. Are all multiperspectival memories originated in multiperspectival experiences? Or do some of them come from pure field experiences and others from pure observer experiences? Does the original perspective of the experience count as a criterion to identify visual memory kinds along with its current perspective? How many kinds of visual memories are there? These are all legitimate questions for which an attentive reader might expect some answers somewhere in the text. On the other hand, if visual memories are in fact multiperspectival, the distinction between field and observer memories loses its sense, and McCarroll’s project of explaining the nature of observer memories becomes difficult to understand. Nevertheless, there seems to be a little confusion over the idea of multiperspectival visual memories. Information used to construct visual memories can be multiperspectival, such as in a Cubist painting, where the physical appearance of a person or an object is depicted from distorted multiple perspectives. But this does not mean that the visual memory itself needs to be multiperspectival. Although field and observer memories can be both constructed from multiperspectival information, they still seem to differ in one essential feature: the presence or not of the self who remembers. If the self-presence is the distinctive trait that determines the perspective of a visual representation, the perspective of a visual memory representation can be either field or observer, but not both.

This leads us to the other major theme that McCarroll develops in chapter 6 to prove the authenticity of observer memories: the presence of the self in remembering from-the-outside. The traditional and common sense conception of observer memories considers that they are not really genuine because they do not show immunity to error through misidentification (IEM). The reason: they are explicit de se thoughts, that is, thoughts about oneself in which the self is explicitly represented, as part of the explicit content of the thought. So they involve an identification component that is potentially misleading: one may misidentify oneself when one remembers someone doing an action. Someone other than oneself could have done it. So although the event remembered can be accurately depicted, the identification of the person who participated or performed the event might not be. This kind of thought contrasts with de se implicit thoughts, in which the representation of the self is not part of the explicit content but appears implicitly, through the mode of presenting that content. The first-personal aspect of such implicit de se thoughts (such as those presenting proprioceptive information) guarantees that one cannot be mistaken about the identity of the bearers of such states. I cannot misidentify myself when, for example, I feel pain or am hungry. Similarly, it seems that one cannot misidentify oneself when one remembers an action from the same perspective as it appeared to oneself in the past, when it was performed. So for the traditional conceptualization of memory, field memories are genuine memories because they are IEM, whereas observer memories are not because of their lack of immunity.

In consequence, McCarroll’s second main strategy in favor of observer memories is to try to prove that “if episodic memory is IEM, then remembering from-the-outside, as an instance of episodic memory, will manifest IEM” (McCarroll, 2019, p. 159). Once again, his enemy is preservationism. Only a radical preservationist such as Recanati (²⁰⁰⁷) can consider that the IEM is a condition of the genuineness of memories. A constructivist would probably find no link between IEM and genuineness. And if (s)he finds it, it would probably not be grounded on the presence or not of an explicit representation of the self in the visual memory image of an event, especially if (s)he approaches the subject from a naturalistic perspective. For example, the ownership of memories is in fact sometimes disputed (particularly between twins) and the visual perspective does not play any role in proving who is right: disputed memories are just likely to be recalled from a field perspective as non-disputed memories (^{Sheen, Kemp & Rubin, 2001}; ²⁰⁰⁶). That is why the argumentation developed in chapter 6 reads as a response to Recanati, given from an expanded version of Recanati’s general framework and in a certain way only valid within its limits. What is more, it confirms once more McCarroll’s adoption of quasi-preservationism instead of the constructivist approach that he professes at the beginning of the book. “Conceiving of observer perspectives as implicit de se has explanatory value” (McCarroll, 2019, p. 180) only in a preservationist framework. Other conceptualizations of memories allow us to understand observer perspectives as genuine episodic memories without conceiving them as implicit de se thoughts, that is, without denying that the representation of the self can be part of the memory content.

In any case, for the purpose of proving that observer memories are not different from field memories in what concerns IEM, McCarroll makes two moves. The first one consists in arguing that observer memories are identification-free, and the second one, that the presence of the self is not an added content to the visual representation of the past event but is given by the mode of presentation of that content and so, like field memories, observer memories are implicit de se thoughts.

Concerning the first point, McCarroll presents some examples in order to show that in observer memories one sees oneself without having to explicitly identify oneself. According to his explanation, this is possible because self-presence is not the object of one’s awareness: there is no ghostly self whose experience of seeing is part of the memory content. The self is thus more implicitly than explicitly represented, so it is transparent and does not involve a process of identification: in an observer memory, one’s identity is given immediately and non-inferentially. This contrasts with photographs of oneself whose identification, following Rowlands (²⁰¹⁷), can certainly be immediate, but only in a contingent way and not as a necessary feature of one’s encounter with the photograph. McCarroll does not delve into this interesting topic, but the difference between a contingent and a necessary immediate identification for photographs and visual memories respectively is not so obvious. Whereas probably most of our memories are identification-free, some of them—maybe only a few—may require an explicit identification process: I have a clear image of myself driving along the Sydney to Melbourne Coastal Drive, but my partner insists that it was him who mostly drove all along. This makes me doubt about the identity of the driver and, without still being convinced that he was the one who drove all along, this doubt makes the image of me driving blurry, and so the human silhouette of the driver is not anymore identified with me in an immediate and non-inferential fashion. Was it me who drove? Or my partner? This kind of doubt about the identification of myself in a visual memory image is not so different from the kind of doubt that may arise when my image in a photograph is not very distinct and I have reasons to believe that the silhouette may correspond to other person than me. Similarly to visual memories, this identification component in photographs is more an exception than the rule, so it is expected that most of the identifications of our photographic self-images are immediate and non-inferential. If this is the case, it is not evident per se why in visual memories this immediate identification would be a necessary feature whereas in photographs it would only be a contingent one. The way we identify ourselves in visual memories may finally not be so different from the way we identify ourselves in photographs.

Coming back to McCarroll’s argumentation, the next move is to argue that the self-presence of observer perspectives implicitly arises from the mode of presentation. For this purpose, McCarroll proposes to replace the dualistic mode/content distinction characteristic of Recanati’s framework by a tripartite structure inspired by the School of Brentano, which distinguishes between act, object, and content or mode of presentation (also used by Rowlands, 2017). The idea is that the same object with the same type act (remembering) can fall under different modes of presentation, among which, field and observer perspectives. The mode of presentation affects how one remembers the object, so the complete content of our memories changes according to the mode of presenting the object, but the object that one remembers remains the same. That is why, explains McCarroll, “in remembering from an observer perspective one is simply remembering the event, rather than remembering having seen oneself during the event” (McCarroll, 2019, p. 176). This kind of “expanded” framework to understand mental states allows McCarroll to rule out the traditional idea that observer memories present a content that is added to the original one: the representation of the self. Nonetheless, it is not very clear in the argumentation how the localization of the self in the mode of presentation helps to avoid an identification component in observer memories and thus to prove that they are IEM. The previous argument presented in support of the identification-free nature of observer memories is in principle valid even in a framework where the self is given in the content. In fact, McCarroll’s proposal that the interoceptive information and other forms of internal embodied subjective imagery that are used in the construction of observer memories are responsible for “the subjective, first-personal, and from-the-inside essence of remembering from-the-outside” (McCarroll, 2019, p. 172), suggests that the presence of the self in observer memories is given by the nature itself of the information and not by the mode of presenting it, as he pretends. If “the self-presence of observer perspective memory is internally constructed” (Ibid.) and “may result from this multimodal integration of information” (Idem, p. 66), how can it be produced by the mode of presentation?

What is more, in McCarroll’s framework, the idea that visual perspectives are given by the mode of presentation raises questions about the conceptualization and usefulness of the notion of “mode of presentation”. Both field and observer memories are cases of episodic memories, that is, memories of events personally experienced. As such, the intentional object of both memories is presented “as one that has been formerly witnessed, orchestrated or otherwise encountered” by the rememberer (^{Rowlands, 2017}, p. 49). So the intentional object of field and observer memories is already subsumed under a specific mode of presentation: the episodic one. McCarroll does not mention the episodic mode of presentation but indirectly assumes it in his framework: only the mode of presentation can distinguish an episodic from a semantic memory. But then the relationship between these more fundamental modes of presentation and the perspectival ones becomes confusing, as well as the explanatory role played by the notion of “mode of presentation”. To give an example: the representation—full of nostalgia— of me strolling along the beach with my ex-partner would be a memory whose intentional object is subsumed under the episodic mode of presentation (which allows me to remember that image as something I formerly experienced), but also under the visual observer perspective mode of presentation (which allows me to see myself in the image), and under the affective observer mode of presentation as well (which allows me to feel a new emotion towards that past event). This multiplication of modes of presentation to account for different traits of a memory is somehow tangled, as well as the multiplication of presences of the self that it implies: the presence of the past self who previously experienced the event, the presence of the past self who is “seen” in the visual memory, the presence of the current self who is feeling nostalgia. That is why the resignification given to the notion of “mode of presentation” to simultaneously refer to different kinds of mental states and standpoints of our mental life does not present the same robustness as the original Fregean concept applied to propositional attitude ascriptions.⁶ In conclusion, within the conceptual framework proposed by McCarroll to understand mental states, the explanatory power of the concept of “mode of presentation” becomes suspicious in a certain way, and more should have been said to clear up these doubts.

In any case, one thing is clear: the hard, thoughtful and original work done by McCarroll in this wonderful book. Although some readers may regret the quasi-preservationist framework implicitly assumed to explain the nature of observer memories, it is undeniable that McCarroll not only achieves his main goal by debunking the idea that observer memories are less genuine than field ones, but also presents a masterly explanation of visual observer perspectives and the perspectival mind.