A puzzle about first-person imagination

It is easy to imagine being someone else from the first-person point of view. Such imaginings give rise to a puzzle. In this paper I explain what the puzzle is and then consider several existing attempts of solving the puzzle. I argue that these attempts are unsuccessful. I propose a Lewisian account of first-person imagination and make the case that this account solves the puzzle


Introduction
First-person imagination gives rise to a puzzle. The puzzle emerges when we imagine from the inside being someone else. On the one hand, our imagination is directed at the person we imagine being. On the other hand, the imagining is firstpersonal and seems to concern ourselves. To illustrate, consider a situation where I imagine being Napoleon looking out on a battlefield (Williams, 1973). Examine the content of my imagining. I imagine a certain scene, perhaps soldiers carrying muskets, as experienced from Napoleon's perspective. I, the imaginer, do not seem to be an additional element of the picture. At the same time, I imagine from the firstperson perspective that it is me who is having the relevant experiences and thoughts, and that I am performing the imagined actions. The problem is that when we take seriously the first point, i.e. the fact that I am not a separate part of the imagined situation, we seem to lose the second point, i.e. we seem unable to explain in which sense the imagining concerns me, rather than merely Napoleon. Could we say that I 1. (FIRST-PERSON PERSPECTIVE) The imagined scene is pictured from the firstperson perspective and the imagining concerns the imaginer herself. 2. (SELFLESSNESS) The imagined scene contains the imagined person but does not contain the imaginer as an additional element. 3. (POSSIBILITY) The imagining is coherent and the imagined scene is possible.
Admittedly, these claims are somewhat vague. But this is unavoidable, as they are meant to capture our pre-theoretic intuitions about the relevant cases of first-person imagination. Let us have closer look at why each of them seems plausible. First, consider FIRST-PERSON PERSPECTIVE. When I imagine being someone else, such as Napoleon, I am picturing a certain scene from the first-person perspective. I am imagining that I am experiencing this scene through the eyes of the imaginary person. I cannot adequately characterise the content of the imagining using only third-person terms. Rather, I have to describe it as a situation in which I am having such-and-such experiences and in which I am performing certain actions, etc. There is then an intuition that the imagining concerns me, the imaginer, myself. The

3
A puzzle about first-person imagination fact that inside imagining is first-personal and concerns the imaginer distinguishes it markedly from outside imagining. In outside imagination, what I imagine often has nothing to do with me (unless I explicitly imagine myself from the outside). For instance, I may imagine from a third-person perspective a situation in which Napoleon is riding a horse, commanding his troops, etc. Here, I am not part of the picture at all.
Reflect next on SELFLESSNESS. When I reflect on the qualitative content of my imagining, I do not find myself, CW, as a separate element within that content. 2 For instance, its visual content only represents (parts of) a Napoleonish figure and the surrounding environment. To put it crudely, when I picture being Napoleon alone in a room, the number of actors on my mental stage is one. Further, I do not imagine myself with added on Napoleonish features. Rather, I imagine a scene that contains the real Napoleon. Neither do I picture Napoleon with added on me-ish features. The Napoleon of my imagination may perfectly resemble the historical figure, and may have nothing in common with me. Finally, my imagining does not appear to have a relational structure. For instance, I do not seem to picture myself and Napoleon standing in the identity relation, by e.g. superimposing a picture of myself onto a picture of Napoleon. (Perhaps, these are things I might do when asked to imagine being Napoleon, but they are not the most natural way of complying with the request.) Assess finally the plausibility of POSSIBILITY. On the face of it, my imagining is coherent and what I imagine seems possible. I imagine a relatively ordinary situation, as viewed from a certain perspective. There does not seem to be an obvious or a subtle contradiction in the imagined content. As just noted, I do not seem to imagine an explicit identity between me and Napoleon. And I am able to imagine being Napoleon while explicitly assuming that I am not in fact him; I do not have to assume that "my biography is intertwined with Napoleon's" (Ninan, 2016), that I am e.g. a reincarnation of Napoleon (Velleman, 1996). Nothing in my imagining suggests that what I imagine is impossible or otherwise unusual.
All three of the above claims seem to be accurate characterisations of the imaginative episode. An adequate theory of first-person imagination should explain why they appear true, ideally by showing that they are true. The Lewisian account of first-person imagination manages to validate all three. Before considering this account, we should first survey alternative attempts of solving the puzzle. As it turns out, none of them is completely satisfactory.

The Naïve view
One can describe the above imagining by saying: "I imagine being Napoleon". Alternatively, one can report it as follows: "I imagine that I am Napoleon". A straightforward interpretation of the latter report suggests that I am imagining, against what was said above, that Napoleon and I are standing in the identity relation. Ninan (2016) labels this position the 'naïve view' (without intending the label to be disparaging). 3 A positive feature of the naïve view is that it delivers a neat match between a natural interpretation of certain imagination reports and the postulated content of our imaginings. 4 Further, the view may also accommodates FIRST-PERSON PER-SPECTIVE. My imagining represents my own perspective and seems to concern me, since I am part of the imagined content, featuring as one relatum of the identity relation.
A downside of the view is that it fails to maintain POSSIBILITY. If I am indeed imagining that I am identical to Napoleon, then my imagination does not present a possible scenario, since I could not be him. Furthermore, this fact should be transparent to me. I should be in position to recognize that my imagining is impossible. However, that is not the case-my imagining does not strike me as impossible. Williams expresses this point in the following passage: [According to the naïve view] what I imagine seems to be straightforwardly self-contradictory, which stops me in my tracks; and this will not do, for I know that, in imagining being Napoleon, I am not stopped in my tracks. (Williams, 1973, p. 44) Another problem is that the naïve view does not maintain SELFLESSNESS. Intuitively, my imagination is non-relational and centered solely around Napoleon. The naïve view denies this: both Napoleon and I are equal parts of the imagined scene and portrayed as standing in certain relation to each other.
The naïve view can only maintain one of the three pre-theoretic intuitions concerning imagining being someone else. Can other proposals do better? In his influential discussion of the issue, Williams (1973) considers a form of Cartesianism about first-person imagination. We can assess this view next. 5

3
A puzzle about first-person imagination

Cartesianism
According to Williams's version of Cartesianism, when I imagine being Napoleon I view myself as a featureless Cartesian center of consciousness. I imagine a situation that contains both Napoleon and my Cartesian ego, and picture my Cartesian ego as located where Napoleon is, capturing the idea that I am seeing the world through his eyes.
Williams's version of Cartesianism is best understood not as a proposal regarding our actual metaphysical nature, but rather as a view about the content of first-person imaginings (and modal statements of the form I could have been thus-and-so). The Cartesian self is introduced as my representative in the world of imagination, and not as an interpretation of what my 'real self' is like (Williams, 1973). Thus understood, Cartesianism does not assume that we really are Cartesian egos, it merely presupposes a Cartesian metaphysics for the world of imagination. The view therefore has a tri-partite structure, which Williams finds objectionable, comprising: (i) the real self, (ii) the Cartesian self, and (iii) the imagined person, i.e. Napoleon.
Prima facie, Cartesianism appears to be an accurate description of the phenomenology of imagining being someone else. It seems true that I bring little more to the world of imagination than my conscious perspective, and that I imagine this perspective as being located where Napoleon is. The view also appears to validate all three of the above claims. First, SELFLESSNESS is maintained since the qualitative content of my imagination consists solely of a Napoleonish figure (or the parts of the figure that are perceptible from the imagined perspective), and the figure's environment. There is no need to make room for someone resembling my real self. Second, Cartesianism preserves POSSIBILITY. A scenario where an immaterial entity with conscious mental states is connected to Napoleon's body may be far-fetched, but it is not impossible-most philosophers believe that even though our universe is not Cartesian, it could have been.
Third, at first glance Cartesianism appears to also uphold FIRST-PERSON PERSPECTIVE. The imagined possibility concerns me, one may think, because it contains my conscious perspective. On second thought, things are less clear. First, unless I really am identical with my Cartesian self, the situation does not literally involve me after all. What about the idea that it features me, since it contains my conscious perspective? What exactly could be meant by this? The situation does not contain any of my actual token-experiences. Further, if it did contain my conscious object picked out by S's tokening of the first-person concept (García-Carpintero, 2016). This presupposition contributes to the cognitive significance of the thought and plays a role akin to a belief state in the framework of (Perry, 1979). When the token-reflexive account is applied in a straightforward way to the case of first-person imaginings it results in the the naïve view, since it predicts that the central content of S's imagining that she is Napoleon is the singular proposition that S is identical to Napoleon. Should the Lewisian solution to the puzzle indeed turn out to be superior to the naïve view, this provides a pro tanto reason to prefer the Lewisian account of de se thought over its token-reflexive rival. Thanks to an anonymous reviewer for this journal for encouraging me to include a discussion of the token-reflexive approach to de se thought in the paper. perspective, then I should be the bearer of the imagined conscious states. But it seems that the relevant mental states belong in the first instance to the imagined Napoleon. This also suggests that the Cartesian picture does ultimately not adequately represent what I am imagining. According to Cartesianism, the subject of experience and the locus of behavioral control in the imagination world is my Cartesian self, not Napoleon. That is not how I picture things. The Napoleon of my imagination is not a mindless, physical shell, controlled by an external force. No, I imagine an ordinary person, with his own thoughts and experiences, who is in control of his body and actions. I simply also imagine from the inside being this ordinary, non-Cartesian Napoleon.
There is an additional problem. Suppose that I came to adopt a Cartesian worldview and started to believe that each of us is in fact a composite object, consisting of a physical body and a certain Cartesian ego. Suppose further that I then imagined being Napoleon in the Cartesian sense-I imagined being the combination of body NB and ego NB and viewing the world from the perspective of Napoleon's ego. 6 The content of this imagining, however, cannot be modelled as a situation in which I am body NB plus ego CW , since I explicitly imagine being body NB and ego NB . So Cartesianism does not provide a general recipe for modeling first-person imaginings. Ironically, it fails for imaginings that are explicitly Cartesian. Let us then discuss Williams's proposal next.

Williams's view
Williams rejects both the naïve view, according to which our 'empirical selves' figure in the imagined situation, as as well Cartesianism, according to which that role is played by our Cartesian selves. Both are, according to him, based on an illusion. The illusion consists in thinking that we need a third element, empirical or Cartesian, which represents us in the world of imagination. In contrast, Williams likens imagining to pretending, which, he suggests, has a binary structure. 7 He writes: In the description of this activity [i.e. pretending to be Napoleon], only two people need figure: the real me and Napoleon. There is no place for a third item [...] regarding which I imagine that it might have belonged to Napoleon. (Williams, 1973, p. 43).
Beyond postulating a binary structure for first-person imagination, Williams does not give us much by way of an explicit theory. Even though the account is sketchy, we can assess its general structure. First, it secures POSSIBILITY. The content of my imagining is simply a situation containing Napoleon doing this or that, e.g. looking out on a battlefield. This content does not differ from that of an ordinary outside imagining involving Napoleon. Since it is possible for Napoleon to e.g. look out on a battlefield, the account gets a tick here. Second, the view also validates OTHER DIRECTED CONTENT. As we have just seen, the content of the imagining is directed exclusively at Napoleon. I, the imaginer, do not come into the picture at all. So the account also ticks this box.
The problem, however, is that the view fails to secure FIRST-PERSON PER-SPECTIVE. Since the content of my imagining is entirely about Napoleon, it is unclear in which sense the imagining concerns me, the imaginer. As it stands, Williams's account does not have the resources to explain the crucial difference between imagining being Napoleon from a first-person point of view and imagining the same content from a third-person point of view. 8 This point can also be brought out by reflecting on Williams's analogy with pretence. As mentioned above, Williams sees a close parallel between pretending to be Napoleon and imagining being Napoleon. Importantly, both involve only two entities, the imaginer/pretender and Napoleon.
[...] there are only two persons involved in this, as I said, the real me and Napoleon. It is as unproblematic that I can imagine being Napoleon as that Charles Boyer could act the rôle of Napoleon. (Williams, 1973 p. 45) But the analogy between pretence and imagination breaks down at the crucial juncture. When Charles Boyer pretends to be Napoleon, the pretence does not really concern him. Charles Boyer himself is not involved in the pretence, or only in a causal sense. True, he produces the world of pretence, in that the content of the pretence depends on his actions, physical features, etc. But Charles Boyer is replaceable. The pretence would have been essentially the same had Napoleon's role been played by Alan Marshall instead (assuming that both had the same acting talent, physical appearance, etc). Not so in the case of first-person imagination. It makes a difference to the imagining that is is me, rather than you, who imagines being Napoleon. My imagining being Napoleon is importantly different from your imagining being Napoleon, in that mine concerns me and yours concerns you. In contrast, Charles Boyer's and Alan Marshall's pretending to be Napoleon do not differ in this way. The pretence created by Boyer does not concern him rather than Marshall. So Williams's attempt of explaining first-person imagination by analogy with pretence fails to account for the fact that de se imaginings are first-personal and concern the imaginer.
Let us sum up. Our discussion of attempted solutions of the puzzle started off from the naïve view. This view could secure only one of the three claims, i.e. the idea that first-person imaginings concerns ourselves. We then considered a Cartesian view of first-person imagination. This view was the mirror-image of the naïve view. It was able to account for the idea that we are imagining a genuine possibility, and that our imagination is focused solely on Napoleon. But it could not properly explain that the imagining is first-personal and concerns ourselves. Further, it did not yield an accurate picture of the imagined content, and was inapplicable to certain firstperson imaginings. Finally, we assessed Williams's account. It had the same profile as Cartesianism and was equally unable to explain the first-personal character of the imagining. None of the proposals considered so far could offer an adequate solution of the puzzle.
In the remainder, I will discuss two further proposals. In §6 I discuss the account of Ninan (2008Ninan ( , 2009Ninan ( , 2016, and in §7 the account of Recanati (2007Recanati ( , 2012. Both are developed within a Lewisian framework. It will therefore be easier to understand their views and my criticisms of them after having introduced the Lewisian framework first.

Lewis's theory of De Se belief and a puzzle about first-person belief
The mechanics of the Lewisian account of first-person imagination can best be elucidated by comparing it to Lewis's theory of de se belief. Compare for instance my first-person imagining that I am Napoleon to my first-person belief that I am him. The de se belief that I am Napoleon gives rise to a puzzle which closely parallels the imagination puzzle.
First, my belief represents things from the first-person perspective and directly concerns me. Whether the belief is true or false depends on how things are with me, on whether I, NN, am Napoleon. Call this claim FIRST-PERSON PERSPEC-TIVE dox . Second, the belief's content is centered on Napoleon. When I reflect on how things are according to my belief, the corresponding belief world revolves around Napoleon. To repeat a point from above, when I believe that I am Napoleon sitting alone in a room, the associated doxastic situation simply contains Napoleon sitting in a room. Depending on how deluded I am, the Napoleon of my belief world may not resemble me at all. Call this claim SELFLESSNESS dox . Third, the belief is coherent and seems to present a possible situation. What I take to be true of myself is in fact true of some believers-when tokened by Napoleon, the belief I am Napoleon is true. As I do not know a priori that I am not Napoleon, I cannot decide from the armchair whether the belief is true or not. For all I know a priori, it could be true. Call this claim POSSIBILITY dox . David Lewis (1979) has proposed a highly influential theory of first-person belief, such as I am happy or I am Napoleon. I will first show that this theory solves the belief puzzle. I will then make the case that a Lewisian account of first-person imagination can solve the puzzle about imagination.
According to Lewis (1979), all beliefs, including first-personal ones, can be modelled as a binary relations between a subject S and a content C. Lewis argued that contents should not be understood as classical propositions, i.e. sets of possible worlds or Russellian/Fregean structures that determine such sets. Rather, they should be understood as sets of centered possible worlds, i.e. triples of individuals, times, and worlds. A centered world < , , > belongs to the content of my belief iff the way the centerindividual i is at t in w is the way I take myself to be. For instance, I have the firstperson belief I am happy iff I stand in the belief relation to the set of centered possible worlds with happy center individuals: {< , , >∶ } . Lewis calls the belief relation that links a subject and centered content self-ascription (of a centered content). On this view, beliefs are not true or false at a possible world, their truth is relative to a world plus a time and an individual. Belief tokens get evaluated for truth at the believer's actual centered world, i.e. the triple <believer, believer's present time, believer's world>. A belief token B is true iff the believer's actual centered world is an element of B's content.
While this may sound technical, the account has intuitive appeal. 9 It captures the idea that we do not represent things from an eternal, god's eye point of view, but rather from our own individual perspective within space and time. Since we are not always certain about the objective location of this perspective, the cognitive significance of our beliefs cannot be captured by absolute and impersonal informationsome of our beliefs are genuinely perspectival. Here, we are primarily interested in whether the theory can solve our puzzle. If it can, we can regard this as additional support for the approach. So can it?
First, the theory maintains SELFLESSNESS dox . The content of my belief that I am Napoleon is the following set of centered possible worlds: it is a set of situations centered on Napoleon. I, NN, am not explicity represented in this content. Your de se belief that you are Napoleon has exactly the same content. In this respect, Lewis's account resembles Williams' proposal. There are just two individuals involved in the belief: the believer and Napoleon.
Importantly, Lewis's theory manages to also secure FIRST-PERSON PER-SPECTIVE dox . First of all, the content of the belief is perspectival and the believer ascribes this perspectival content to herself. Further, while the centered content of my de se belief that I am Napoleon does not explicitly represent me, its truth value depends on the way I am. As pointed out above, belief tokens get evaluated for truth at the believer's actual centered world. So whether my de se belief that I am Napoleon is true or not depends on me. Similarly, when you have a de se belief that you are Napoleon, your belief is evaluated at your centered world, and its truth value therefore depends on the way you are. Even though our beliefs have the same centered content, the truth of my belief token is relative to me and the truth of your belief token is relative to you. This relativity of truth provides the required explanation as to why my belief is about me or concerns me and why your de se belief is about you or concerns you. As a result, Lewis's theory maintains FIRST-PERSON PERSPECTIVE dox .
What of POSSIBILITY dox ? We can assess de se beliefs for possibility in terms of their content. I will here assume an orthodox conception of metaphysical possibility according to which possibilities correspond to ordinary uncentered possible worlds (for an alternative approach see the discussion of Ninan's account in §6). On this view, there is a close, yet somewhat indirect connection between centered belief content and possibility. A centered content is a set of centered possible worlds, a set of triples of the form < , , > . Such a set determines a set of ordinary uncentered possible worlds-after all, a centered world is simply an ordinary possible world with a marked individual and time. The corresponding set of uncentered worlds can be attained by simply removing the marked individual and time from each triple. If the resulting set of uncentered worlds is non-empty, the belief is possible in terms of content. We can assess the contents of belief types for possibility, since they are invariable across different contexts. Let us apply this to the belief in question. The centered content of my belief I am Napoleon is a set of centered worlds whose center individual is Napoleon. This determines a set of uncentered worlds containing Napoleon, corresponding to the uncentered content of the existential statement Napoleon exists. Hence my belief that I am Napoleon is possible when assessed in terms of content, since the associated set of uncentered worlds is non-empty.
While the content of my belief represents an ordinary metaphysical possibility containing Napoleon, we might say that my belief is impossible in another sense. We can explain this sense of impossibility as follows. The centered content of my de se belief's specifies a condition which I, CW, have to fulfil for my belief token to be true. In the above case, we can express this condition as being identical to Napoleon. Assuming that distinctness is necessary, this a condition which I cannot fulfil since I could not be Napoleon.
The Lewisian theory can explain that the de se belief that I am Napoleon presents us with a possible scenario. The uncentered modal content of the belief type is nonempty and corresponds to an ordinary set of possibilities containing Napoleon. It is plausible that the intuition underlying POSSIBILITY dox is based on the cognitive significance of the belief type, on how the belief represents things as being independent of the context in which it is held. And it is this notion of possibility that is captured by assessing the belief type for possibility in the way described above. At the same time, the Lewisian view can account for the fact that there is a sense in which my belief is impossible, in that a specific tokening of the belief places a conditions on the believer which she cannot fulfil. Assessing whether a belief token is impossible in this second sense depends on a posteriori background knowledge regarding the context in which the belief is tokened, and does therefore not affect the rational coherence of the belief. (See also the discussion of Recanati's criticism of the Lewisian account in §7.) Summing up. Lewis's theory of de se belief maintains all three desiderata and solves the puzzle about first-person belief. It is then reasonable to expect that a parallel account of first-person imagination can solve the puzzle about first-person imagination. And indeed, this expectation is borne out, as I will show in the next section.

Solving the puzzle: a Lewisian account of first-person imagination
Accounts of first-person imagination along Lewisian lines have been proposed by Recanati (2007Recanati ( , 2012 and Ninan (2008Ninan ( , 2009. 10 Following Recanati's and Ninan's lead, we can analyse first-person imagination in very close analogy to Lewis's theory of de se belief. First-person imaginings can be modelled as binary relations between a subject S and a centered content C. A centered world <i, t, w> belongs to the content of S's first-person imagining iff the way the center-individual i is at t in w is the way S imagines herself being. Consider my first-person imagining of being happy. A centered world belongs to its content iff the individual at the center is happy at the time and world of the center, {< , , >∶ } . The content of this imagining is the same as that of the de se belief that I am happy. And as in the case of de se belief, de se imaginings are evaluated for satisfaction at the imaginer. A de se imagining token I is satisfied iff the triple <imaginer, imaginer's present time, imaginer's world> is part of I's content. Is it really appropriate to evaluate imaginings for truth/satisfaction? Imagination does not seem to be an attitude aimed at representing how things are. 11 In response, we can first observe that it is common practice to assign truth conditional content to attitudes which do not have a mind-to-world direction of fit, such as desires. We can model the content of propositional attitudes in general as a set of situations in which the attitude is satisfied. The content of a de se attitude can be modelled as a (interesting (Egan, 2007)) set of centered possible worlds. As for de se belief, the notion of satisfaction is in general relative to the attitude holder. For example, S's de se desire D is satisfied iff S's actual centered world is part of D's content. When S de se desires to be happy her desire token D is satisfied iff S is happy (at the time and in the world of D). Both de se belief and de se desire share the same type of content, and both involve a subject-relative notion of satisfaction. The difference between the two attitudes lies in their diverging functional roles, i.e. in how they come about, how they interact with other attitudes, and what their role in the production of behavior is.
We can then apply this model to the case of de se imaginings. A de se imagining I is associated with a set of centered worlds that models I's representational content. As in the case of de se belief and de se desire, the satisfaction of a de se imagining is relative to the imaginer. S's de se imagining I is satisfied iff S's actual centered world is part of I's content. For instance, if she imagines being happy, her imagining I is satisfied iff S is happy (at the time and world of I). This makes pre-theoretic sense. There is an intuitive sense in which we can compare the imaginary situation to how things actually are, and in which we can determine whether the imaginer really is the way she imagines herself being. Again, the crucial difference to de se belief consists not in content or in the notion of satisfaction, but rather in the nature of the respective attitudinal relation, which is once more manifested in the contrasting functional roles of both attitudes. For instance, we do not adjust our imaginings to our perceptual environment, or retract an imagining upon learning that it is not satisfied, and imaginings typically do not directly guide our actions. Call the relation that holds between S and the centered content of S's first-person imagining imaginary self-ascription.
Back to our main question: can the Lewisian account of first-person imagination solve the puzzle about first-person imagination? First, the account maintains SELF-LESSNESS as the content of my imagining is exclusively centered on Napoleon. Its content is identical to that of my de se belief that I am Napoleon: the set of situations centered on Napoleon: {< , , >∶ = } . As in the case of de se belief, I, CW, do not explicitly figure in this content.
Second, the account also maintains FIRST-PERSON PERSPECTIVE. Like de se beliefs, de se imaginings have a perspectival content which the imaginer ascribes to herself in a non-factive mood. The centered content of imagining being Napoleon represents the imaginary scene from Napoleon's perspective and by self-ascribing this perspective, the imaginer simulates that perspective for herself. Further, the fact that the content of the imagining is evaluated at the imaginer explains the sense in which the imagining concerns her. Compare de se belief with de se imagination. Why is my de se belief that I am happy about me, whereas your de se belief that you yourself are happy is about you? Because the truth of my belief depends on whether I am happy, whereas the truth of your belief depends on whether you are happy. Similarly, why does my de se imagining being happy concern me, whereas your de se imagining being happy concerns you? Because the satisfaction of my imagining depends on whether I am happy, whereas the satisfaction of your imagining depends on whether you are happy. The same holds for a de se imagining of being Napoleon.
Finally, the account also validates POSSIBILITY. Again, just like the corresponding de se belief, the imagining's content determines a non-empty set of ordinary uncentered possibilities containing Napoleon. That the imagining is possible in terms of content accounts for the fact that the attitude presents us with a possible situation and suffices to preserve POSSIBILITY. Once more, we can also assess my token imagining for possibility in terms of whether it could be satisfied. So assessed, my imagining is impossible, since I could not be Napoleon.
The Lewisian account of first-person imagination retains the good parts of Cartesianism and Williams's view, while avoiding their pitfalls. Like Williams's view, it construes the content of first-person imaginings as merely containing the imagined person (i.e. Napoleon), and does without a third element, empirical or Cartesian, which represents us in the world of imagination. It preserves the idea that all that we bring to the world of imagination is our conscious perspective. At the same time, it avoids the mistake of reifying this perspective and populating the world of imagination with Cartesian egos. The conscious perspective is simply the perspective of our ordinary real selves. Further, we have pointed out that Cartesianism misconstrues the content of the imagining by portraying Napoleon as a physical puppet controlled by an immaterial puppeteer. On the Lewisian picture, the content of the imagination involves a perfectly ordinary Napoleon, who is the primary subject of the imagined mental states and in control of his own actions. Finally, the Lewisian theory can explain, unlike the two alternatives, that de se imaginings are first-personal and concern the imaginer. Like Recanati, Dilip Ninan (2008, 2009) has put forward a Lewisian analysis of firstperson imagination. His ultimate account is more complex than the simple Lewisian theory proposed here. But the added complexity is motivated mainly by cases of the "the impersonal-de se" which do not affect our present discussion. In fact, Ninan contends that the simple Lewisian theory is satisfactory for ordinary cases of inside imagination: "in cases where [de se ] ignorance is not at issue, this simple account [...] seems entirely adequate" (Ninan, 2008, p. 90). 12 While Ninan and I agree on how to model the content of the relevant first-person imaginings, we crucially differ in our assessment of the connection between firstperson imagination and possibility. Ninan thinks that first-person imaginability is a direct guide to possibility and to the truth of first-person modal judgments. I, on the other hand, do not. I believe that the link between first-person imaginability and possibility is broken, and that first-person imagination should not be used as a guide to de se modal truths. (I do however think that one can maintain a general connection between conceivability and possibility along the lines of (Chalmers, 2002).)

Ninan's view
Ninan distinguishes between two types of metaphysical possibility. First, he acknowledges uncentered metaphysical possibilities, corresponding to regular possible worlds. But, second, he also countenances centered metaphysical possibilities, corresponding to centered possible worlds (this notion of possibility is inspired by a proposal from Lewis (1983Lewis ( , 1986). Ninan thinks that first-person imagination offers us reliable access to centered metaphysical possibility and endorses the following principle: Centered Guide Imagining from the inside is a guide to centered possibility. If I can imagine a centered content p, that is evidence that there is a centered world <w, x> accessible from <actual world, me> such that p is true at <w, x>. (Ninan, 2008, p. 90).
Ninan further maintains that the existence of an accessible centered metaphysical possibility renders true de se modal statements, such as I could have been F. Such a statement is true when uttered by S iff there is a centered possible world accessible from S's actual centered world whose center individual is F (Ninan, 2008(Ninan, , 2009). The fact that I can imagine being Napoleon is evidence that there is a centered world with Napoleon at the center which is accessible from my actual centered world. On Ninan's picture, POSSIBILITY is then validated because being Napoleon is a genuine centered possibility for me, and the statement I could have been Napoleon is true when uttered by me.
I do not think that Ninan's picture is attractive, since it does not yield an adequate methodology for thinking about the self. There are three main reasons for this. First, there will often be conflicts between modal judgments from a third-person point of view and from the first-person perspective. Second, first-person imagination is in one sense too unconstrained, leading to de se illusions of possibility. Third, first-person imagination is in another sense too narrow, leading to de se illusions of impossibility. We should therefore not use de se imagination as a guide to de se modality.
First, on Ninan's view the statement I could have been Napoleon is true when uttered by me. At the same time, the statement CW could have been Napoleon is false, since there is no possibility where CW is Napoleon (assuming that distinctness is necessary). 13 But I am CW. So, could I, CW, have been Napoleon? More generally, when there is a centered possibility that renders true the statement I could have been F, but no corresponding uncentered possibility that renders true CW could have been F, am I, CW, the kind of entity that could have been F? There seems to be no way of answering this question. When two rival metaphysical conceptions of ourselves differ on whether or not we could have been F, we seem at an impasse.
Second, Kripke (1972) has made a convincing case that there is often a gap between what a concept tells us about its referent and the metaphysical nature of the referent itself. For instance, we cannot uncover the nature of water, i.e. that it is H 2 O, from a conceptual analysis of the term water. This point seems especially pressing for the first-person concept. The first-person concept puts hardly any constraint on its referent, i.e. on what type of thing we are. As Nagel puts it: "The concept of the self seems suspiciously pure-too pure-when we look at it from inside.
[...] When I consider my own individual life from inside, it seems that my existence in the future or the past [...] depends on nothing but itself." (Nagel, 1986, pp. 32-33). Correspondingly, there is a suspiciously wide range of scenarios which are imaginable from the inside. I can imagine being an immaterial soul. I can imagine being a purely material being. I can imagine being a different person, such as Napoleon. I can imagine being a rational parrot, or waking up as a beetle. I can imagine from the inside being a conscious artefact, such as animated teapot in a fairy-tale scenario. I can imagine being a conscious avatar in a computer simulation. And the list goes on.
Our persistence, as imagined from the inside, seems equally unconstrained. I can imagine surviving fission as Lefty or as Righty or not at all. I can imagine surviving a complete transformation of my body, switching bodies, and even outliving my body. I can imagine undergoing a comprehensive and radical change in my psychology. I can imagine surviving the replacement of immaterial souls. I can imagining from the inside successively living the lives of different people, such a first being Julius Cesar, then being Napoleon, and later being Greta Garbo, only to be reawakened as a conscious upload on a computer in the future. Interestingly, for each of these scenarios I can equally imagine failing to survive at any intermediate point throughout the scenario. 14 I do not think that it is plausible that our actual nature is this unconstrained. It is important to keep in mind that we are here considering what is metaphysically possible for us. I am not denying that the above scenarios are epistemically or conceptually possible, just as there is an epistemic/conceptual possibility in which water is 14 See (Weber, 2023a, b) for an explanation of why these appear possible from the inside. 13 (Ninan (2008, Appendix) provides an alternative centered semantics for third-person modal statements. But this semantics seems appropriate only for certain epistemic uses of such statements). XYZ. 15 What I am questioning is whether, assuming that I am in fact, say, a human organism, I could also have been a Cartesian soul or a teapot. It is furthermore doubtful whether we can extract any coherent conception of the self from countenancing all the scenarios which are imaginable from the inside as genuine possibilities, since some seem to place incompatible demands on our nature. Returning to the Kripkean picture from above (see also Nagel 1986, §3)), it seems more plausible to assume that the pure first-person perspective reveals very little about our underlying nature. Uncovering this nature requires a posteriori knowledge about the reference of the first-person concept. From this perspective, many of the modal seemings generated by first-person imagination should be regarded as a particular type of modal illusion-they are de se illusions of possibility. 16 Third, while first-person imagination is in many ways too liberal, it is in other respects too narrow. This aspect of first-person imagination creates de se illusions of impossibility. As Ninan points out (Ninan, 2008, p. 94), first-person imagination requires a conscious perspective-we cannot imagine from the first-person point of view being completely unconscious. Further, as Nagel (1974) has argued, the relevant conscious perspective cannot be too far removed from our actual one, e.g. by containing experiences connected to sensory organs which we are lacking. We cannot imagine from the inside what it is like to be a bat. As a result, there is no centered world accessible in first-person imagination where the center individual is completely unconscious or has bat experiences. Hence the following statements come out as true on Ninan's analysis I could not have been unconscious or I could not have had bat experiences. While these scenarios appear impossible from the inside, they are in fact possible: we are often unconscious, e.g. in a state of dreamless sleep, and it also seems plausible that we could have had different sense-organs. The experiential character of first-person imagination rules out certain scenarios which are in fact real possibilities for us. 17 I reject Ninan's analysis of de se modal statements. Instead, I take statements of the form I could have been F as uttered by S to have the same truth conditions as a corresponding third-person statement S could have been F. Both are true iff there is an accessible uncentered possible world according to which S is F. Further, as described in §4 and §5, the existence of an accessible centered world whose center individual is F only renders true an existential statement of the form Something could be F. We have seen above that this existential statement corresponds to the uncentered modal content of the de se imagining being F. On this picture, there is no direct link from de se imagination to de se modal truths. 18 Ninan is skeptical about severing the link between first-person imagination and possibility. He worries that this might endanger our access to possibility in general (Ninan, 2016, p. 276). In (Weber, 2023a), I argue that skepticism about the reliability of first-person imagination can be contained and is compatible with maintaining a general link between third-person conceivability and possibility. To conclude, Ninan's view, according to which first-person imagination is a guide to what is metaphysically possible for us, seems unattractive, because it fails to acknowledge the oftentimes deceptive nature of first-person imagination. Recanati (2007Recanati ( , 2012 considers the Lewisian theory in its basic outline and rejects it in favour of his own proposal. His criticism of the Lewisian theory is as follows:

Recanati's view
Even if the content does not include the self, the self comes into the picture at the next step, in the act of entertaining that content in the relevant mode. If this act is represented as the act of self-ascribing the property [i.e. the centered content] that is the content of the thought, just as in perception but with an additional element of pretence (in order to distinguish imagination from genuine experience), then don't we have the same contradiction one step later? How can I (pretend to) self-ascribe the property of being Napoleon [...if this is a property] that it is impossible for me to instantiate? (Recanati, 2007, p. 204) I take it that Recanati's rhetorical question is incomplete, and should be in full: 'How can I rationally (pretend to) self-ascribe the property of being Napoleon [... if this is a property] that it is impossible for me to instantiate?' Recanati's charge against the Lewisian account seems to be that the account ultimately fails to uphold POSSIBILITY, since it predicts that the subject should be in a position to notice the impossibility of her attitude at the step of assessing the imagined content for satisfaction. Therefore, the account erroneously counts imagining being Napoleon as incoherent or irrational.
This objection can be met. Recanati (2007, p. 204) himself mentions two potential responses. The first is to sever the link between imaginability and possibility; the second consists in his own account. But there is another option.The Lewisian account is not committed to the claim that it is irrational (in the relevant minimal sense) to hold an attitude that the attitude holder could not satisfy. Consider, the case of belief. Whether a belief is rational or not depends in the first instance on its content, and not on whether a tokening of the belief could be true at the believer. We can distinguish transparent from opaque impossibilities. Transparent impossibilities are attitudes with empty centered content, such as the belief I am happy and nonhappy. Opaque impossibilities are attitudes with non-empty centered content that cannot be satisfied by the attitude holder, such as my belief that I am Napoleon. It is irrational to hold a transparent impossibility. 19 In principle, the attitude holder should be in a position to recognize that the content of her attitude cannot be satisfied at any centered world. She can then conclude that it is a fortiori not satisfied at her own centered world, since the centered content of attitudes types does not vary across contexts. Importantly, she does not need to know what the features of her actual centered world are. Not so in the case of opaque impossibilities. These are not impossibilities tout court, since they are satisfied at some centered worlds. However, they cannot be satisfied at the attitude holder's context (i.e. her actual centered world). Even when the attitude could not be satisfied at the subject's context, it may not be irrational for her to hold that attitude, because that fact may not be obvious to her. She may be ignorant or mistaken about relevant features of her context. For instance, she may not know what time it is, or who she is. Assume that I am unaware that I, CW, am the individual at center of my actual centered world, and that I am under the mistaken impression that the center-individual is Napoleon. As a consequence, I take the content of my belief I am Napoleon to be fulfilled at my actual centered world. It is therefore not irrational for me to have this belief. Equally, when I am mistaken about the time of my centered world, I may falsely believe on Monday that it is Tuesday, and rationally impose the condition being on a Tuesday on a time that cannot possibly fulfil it.
Being ignorant or mistaken about a feature of one's context is one way of rationally holding an attitude that one cannot satisfy. Another way is to temporarily suspend judgment on this question. I may be confident about the objective properties of my context, but decide to waive this confidence. Consider a suppositional context, where I am supposing that I was swapped as a baby in hospital and that my real parents are Chuck Norris and Sarah Silverman. The corresponding suppositional content that I am Chuck Norris's and Sarah Silverman's child could not be satisfied by me (at least, if we agree with (Kripke, 1972) about the necessity of origin). Nonetheless, it is not irrational to make this supposition. Imagination arguably resembles supposition in this regard. I may temporarily lift assumptions about my actual context for the purposes of imagination. That notwithstanding, the satisfaction of my imagining remain relative to my actual centered world. Imagining being Napoleon determines a content that could not be satisfied by me, and yet, this fact alone does not render it irrational. The Lewisian theory does not entail that it is always irrational to hold an attitude that one cannot satisfy.
Let us now look at Recanati's own proposal. He contends that the satisfaction of first-person imagination is not always relative to the imaginer. Recanati writes: Imagination is different: what it represents need not concern the imaginer.
[...] I will, therefore, coin the term 'quasi-de se' to refer to the type of thought one entertains when one imagines, say, being Napoleon. The type of imagining at stake is clearly first personal, yet the imaginer's self is not involved-not even at the 'evaluation' stage. The properties [i.e. centered content] that are imaginatively represented are not ascribed to the subject who imagines them [...] (Recanati, 2007, pp. 206-207) If we do not ascribe the imagined centered content (or 'properties' in Recanati's terminology) to ourselves, to whom is it ascribed? According to Recanati, we ascribe it to the person we imagine being.
[...] when I imagine being Napoleon [...] I imagine certain properties being instantiated, but-in central cases at least-it is not to myself that I (self-) ascribe the properties in question. Rather, I ascribe them to Napoleon. (Recanati, 2007, p. 204, my emphasis) Recanati's proposal deviates in a crucial respect from the simple Lewisian theory. According to that theory, the centered content of first-person imaginings is always (imaginarily) self-ascribed, and their satisfaction is always relative to the imaginer. This explains why such imaginings are first-personal and concern the imaginer. In contrast, Recanati suggests that when we imagine being someone else, we ascribe the relevant content not to ourselves but to the person we imagine being. This proposal faces various difficulties. Here, I want to focus exclusively on the one that relates to our puzzle. 20 The central problem is that, like Cartesianism and Williams's view, Recanati's proposal fails to secure FIRST-PERSON PERSPECTIVE. Since it denies that the content of the relevant first-person imaginings is self-ascribed, it forgoes the Lewisian explanation for why these imaginings are first-personal and concern the imaginer. Recanati is aware of this challenge and tries to provide an alternative explanation within the confines of his own proposal. His explanation has two elements. First, he points out that the target of our imagination, Napoleon, is imagined as having first-personal mental states. Second, in the act of imagining, we simulate these first-personal mental states. He writes: [...] that is what the quasi-de se is: the simulation of a reflexive state. The first personal character of quasi-de se states is inherited from the primary state which quasi-de se imagination simulates. (Recanati, MS, p . 7) Recanati's explanation seems insufficient. It seems possible to engage in an act of imagining that satisfies Recanati's two conditions without thereby engaging in a genuinely first-personal imagining. Assume that I imagine Napoleon looking out on a battlefield from a third-person perspective. In addition, I imagine that Napoleon has certain first-personal mental states, such as perspectival visual experiences and the de se belief I will win the battle. I may then even simulate these mental states, perhaps to make the imagining particularly vivid. That notwithstanding, I may still conceive of the relevant first-personal experiences as belonging to Napoleon rather than me. The whole episode then corresponds to a particularly vivid outside imagining of Napoleon having certain first-personal experiences and beliefs. The imagining misses the crucial ingredient that would turn it into a genuinely first-person imagining: I am not genuinely imagining that I myself am the person with these experiences, and I am not treating the relevant first-personal mental states and actions as my own mental states and actions for the purposes of imagination. It is this final step that is achieved by imaginarily ascribing the actions and mental states to myself, and by treating the satisfaction of the imagining as being relative to me.
In addition, Recanati's two conditions also do not seem necessary for first-person imagination, since we can imagine being a subject that lacks first-person mental states. Consider a subject that (temporarily) does not receive any perspectival perceptual input and that also does not have any other occurrent first-personal attitudes. The subject consciously contemplates only general mathematical theorems, say. We can, it seems, imagine from the inside being such a subject. Against Recanti, it is therefore not a prerequisite of first-person imagination that the imagined person has first-personal mental states. Recanati's analysis fails since the proposed conditions seem neither sufficient nor necessary for first-person imagination.
The discussion of Recanati's proposal has revealed two things. First, his reason for rejecting the Lewisian account of first-person imagination is unconvincing. Second, his own proposal for explaining the first-personal character of inside imagination is inadequate and fails to solve the puzzle. We should be purists and stick to the Lewisian original.
8 First-person imagination and the self I have argued that only the Lewisian account offers an adequate solution to the puzzle about first-person imagination. The account explains the first-personal character of de se imaginings: they have a perspectival content which the imaginer (imaginarily) self-ascribes, and the imagining's satisfaction depends on the imaginer. The account also accommodates the intuition that the content of imagination is focused on the person one imagines being, and that it presents a possible scenario: the assigned content is simply a set of situations involving the imagined person and does not feature the imaginer.
In the discussion with Ninan it became apparent that the conception of first-person imagination may have important consequences for our views of the self. 21 Many of the debates about the self hinge on certain first-person modal claims, e.g. claims regarding what kind of changes we could survive or what kind of entity we could be. For instance, Descartes influential argument for Dualism is based on the modal premise that we could exists as immaterial entities. Locke's central argument for a psychological theory of personal persistence is based on the modal premise that body switching is possible for us. In practice, support for these premises is often sought in first-person imagination. The modal premises seem compelling, because we can imagine corresponding scenarios from the inside; we can imagine being an immaterial entity or switching from one body to another. Should the picture of first-person imagination and its relation to possibility which I have proposed here survive scrutiny, this line of support for Dualism or Lockeanism is shaky (even granting a general link between conceivability and possibility). If we are indeed not part of the imagined content, then such imaginings do not directly reveal possibilities for us. When I imagine being an immaterial soul, the uncentered modal content of my imagining is simply a set of possibilities containing an immaterial soul. Importantly, this soul does not have to be identical with me. We should therefore not take first-person imaginings as direct evidence for singular modal statements like: I, CW, could have been an immaterial soul or I, CW, could have switched from this body to another. From this perspective, Dualism and Lockeanism turn out to be based on a peculiar type of first-person modal illusion. A proper understanding of imagination from the first person point of view may therefore have far-reaching consequences for our views about what kind of thing selves ultimately are.