ArticlePDF Available

I did not expect to be dreaming: Explaining realization in lucid dreams with a Bayesian framework



The commonsense view is that a lucid dream starts when the dreamer realizes that they are currently dreaming. The notion of realization, however, has been accepted at face value, with little consideration of whether the dreamer realizes that they are dreaming in the sense of actual reasoning, or if it is a mere epiphenomenon of lucid dream initiation. This article offers a solution to this problem by, first, arguing that the transition to lucidity can occur as a result of successful reasoning, and second, building a model of this reasoning in terms of probabilistic reasoning. The established Bayesian model explains realization in lucid dreams taking under consideration two factors: the beliefs that the dreamer holds on what is generally probable and improbable, and the dreamer’s admissibility of being in a dream. Defended against important objections, the model offers an explanation of lucid dream initiation, relevant for future research on dreaming.
I did not expect to be dreaming: Explaining realization in lucid dreams with a Bayesian
Piotr Szymanek
Faculty of Philosophy, Jagiellonian University
Gołębia 24, 31-007 Kraków, Poland
Author note
Piotr Szymanek:
This research received no specific grant from any funding agency in the public, commercial,
or not-for-profit sectors.
I have no known conflict of interest to disclose.
The commonsense view is that a lucid dream starts when the dreamer realizes that they
are currently dreaming. The notion of realization, however, has been accepted at face value,
with little consideration of whether the dreamer realizes that they are dreaming in the sense of
actual reasoning, or if it is a mere epiphenomenon of lucid dream initiation. This article offers
a solution to this problem by, first, arguing that the transition to lucidity can occur as a result
of successful reasoning, and second, building a model of this reasoning in terms of
probabilistic reasoning. The established Bayesian model explains realization in lucid dreams
taking under consideration two factors: the beliefs that the dreamer holds on what is generally
probable and improbable, and the dreamer’s admissibility of being in a dream. Defended
against important objections, the model offers an explanation of lucid dream initiation,
relevant for future research on dreaming.
Key words: lucid dream, dream cognition, reasoning, Bayes, probability.
Lucid dreams (henceforth: LDs) have been troubling philosophers for hundreds of
years, dating back even to Aristotle (350 BCE/1918-2000), who noticed that sometimes while
asleep, we can tell that our current experience is but a dream. The topic of lucid dreaming
provokes many questions, as it is the exception which proves that there is a rule. To dream
lucidly is to dream in a strange manner, without the aspect that seems so immanent to
dreaming: the uncritical acceptance of what is occurring in one’s surroundings. In a dream, a
close friend can suddenly turn into a hive of bugs, which start to chase the dreamer through
the different ages of historical Europe and while sometimes such occurrences can be
surprising, in most cases the dreamer just shrugs their dream shoulders, accepting that all of
that is happening for real.
There are many myths surrounding LDs, some of which tend to get mystical, assigning
extraordinary qualities to lucidity. While the scientific inquiry has tried to deal with LD since
the second half of the 20th century, hoping that explaining lucidity will help to explain other
phenomena, such as consciousness, there are still many notions related to LD that have been
taken at face value. Those notions demand an explanation if they are to be used in the
discourse of lucidity. One such notion relates to the question of how and why LDs are even
The commonsense view is that LDs start when the dreamer realizes that they are
currently asleep and dreaming. However, theorists of LD do not usually clarify what the
notion of realization means. If lucid dreamers realize that they are dreaming in the sense of
drawing a correct conclusion, this realization could be described as a form of successful
reasoning. Then, a question arises, how is it possible that the dreamer can sometimes (and
sometimes not) realize that they are dreaming? And in what way this process could be
This article offers an argument supporting the claim that a dreamer’s realization can in
fact be a mechanism of the LD initiation, followed by a plausible account of this mechanism.
In other words, I intend to show that lucid dreams, even though not all of them, can be
initiated with a successful reasoning that the dreamer conducts and that leads him to the
realization that their current experience is but a dream. I also intend to present a model of said
reasoning, developed within a probabilistic framework, and defend it against important
The notion of realization in lucid dreams
There are two definitions of LD: a stronger and a weaker (Windt, 2015). The stronger,
founded on the ideas of Frederik van Eeden (1913), describes an LD as a “dream
characterized by full intellectual clarity and the ability to control the dream” (Windt, 2015, p.
106). The weaker and more common definition characterizes LD in terms of “insight into the
fact that one is currently dreaming” (ibid.).
The notion of full intellectual clarity present in the stronger definition is not clear.
According to Windt (2015), Stephen LaBerge agrees with the stronger definition, yet in his
work (1985) he described lucid dreams in many different ways: as dreams in which a dreamer
knows that they are dreaming or as dreams in which the dreamer is conscious of the fact that
they are dreaming. It seems that the notion of clarity can be replaced with concepts such as
insight, knowledge, or awareness, and the whole definition will continue to sound intuitively
It is possible that the state of the dreamer discussed above is not limited only to insight
or only to conscious awareness, but to many cognitive properties composing the LD.
However, it is safe to say that what binds all characteristics of LD is the dreamer’s capability
to (somehow) tell that they are currently asleep and dreaming. This can be a working, minimal
definition for the current problem: LDs are dreams in which the dreamer can correctly mark
the current situation as a dream. But how is such a state possible?
Researchers interested in LD tend to state that the dreamer realizes or recognizes that
they are currently dreaming (Green, 1968; LaBerge, 1985, 1990; LaBerge & Rheingold, 1991;
Revonsuo, 1995; Metzinger, 2003; Windt 2015; Adams & Bourke, 2020). Stumbrys and
colleagues (Stumbrys, Erlacher, Schädlich, Schredl, 2012) state that “in lucid dreams the
dreamer has to observe and evaluate his or her present experience to recognise the dream state
and become lucid” (p. 1457). This idea is probably a result of the fact that most of cases of
LD initiation involve a feeling of coming to know that bizarre occurrences experienced are in
fact a part of a dream (LaBerge, 1990). It also seems very intuitive that a state of being able to
mark the current situation as a dream should involve a moment of recognition, similar to a
situation in which one realizes that they are on the wrong street while going somewhere
without paying attention to surroundings.
But how should the notion of realization in LD be understood? A situation in which
the dreamer realizes that their current experience is part of a dreaming narrative seems to be
related to some kind of successful reasoning. In an exemplary situation, the dreamer notices a
bizarre occurrence (e.g. spots Santa Claus and the host of his elves) and confronts it with their
knowledge, searching for an explanatory hypothesis. Due to a lack of better hypotheses, the
dreamer concludes that they must be asleep and dreaming, as there is no Santa Claus and no
elves. This marks the point of realization, thus the initiation of LD. Thus, LD realization is
actually the outcome of a successful reasoning based on the current experience and the
dreamer’s prior knowledge. But does this reasoning happen before every single case of LD? Is
every LD triggered by something unusual, such as the appearance of Santa Claus and his
As it turns out, LDs can also occur without perceiving any anomalies in the content of
a dream and without the moment of experienced realization. Subjects sometimes report that
their dream was instantly lucid just after a short break in sleep (Green, 1968; LaBerge, 1985)
or that they “just knew” that they are dreaming (Adams & Bourke, 2020). Moreover, studies
have shown that there are many other so-called triggers of LD activation. According to
LaBerge (1985), the major triggers are cases of perceptual inconsistencies, emotional arousal,
and a sort of instant recognition of the dream-like nature of experience (LaBerge, 1985;
Adams & Bourke, 2020). The minor triggers, more rare and specific, are cases like those in
which the dream is lucid from the very beginning; in which the dreamer recollects having a
similar dream before (déjà rêvé, French for “already dreamt”), which starts LD; or in which
the dreamer associates the initiation of LD with a specific event of touching an object
(LaBerge, 1985).
Additionally, even if we consider only these cases of LDs in which the dreamer
experienced a vivid sensation of recognition, we still cannot be sure if they all involve
reasoning of any kind. As Green (1968) noticed, even though typically the dreamer can realize
that they are currently asleep and dreaming after being filled with anxiety or fear (emotional
arousal) or after detecting that there are some incongruent elements in their surroundings,
scientists and philosophers should not simply assume that these events are more than an
epiphenomenon of LD initiation. The problem here is that the dreamer’s mind could in fact
produce the feeling of intellectual achievement that lucid dreamers typically experience
(Windt, 2015) without any relevant relationship between this feeling and finding oneself in
the state of marking the surrounding narrative as a dream. In other words, LD could occur out
of any reason whatsoever, but with the dreamer convinced that they realized that what is
happening right now is a dream.
In sum, it seems that we can neither generalize all LDs as involving realization nor can
we treat the presence of realization experience as definitive. What is left for the current
discussion, then, is a very specific case of a “cognitive realization”: the dreamer’s realization
that they are asleep and dreaming that which is an endproduct of successful reasoning, as
opposed to a “sudden realization” which can involve the same experience but does not
involve reasoning. But are there any reasons to believe that the cognitive realization is
relevant for LDs at all?
Cognitive realization as a route to lucid dreams
Here I want to prove that the cognitive realization can be the mechanism of LD
initiation, even though it is not always the case. As we have already seen, there are cases of
LD that simply do not fit the notion of cognitive realization. There are still good reasons,
however, to believe that cognitive realization is a possible route for LD initiation.
Studies on neural correlates of LD involving electroencephalograms (EEGs) have
shown activations in frontal areas of the brain parallel to lucidity that do not typically occur
during normal REM sleep (Voss, Holzmann, Tuin, & Hobson, 2009; Dresler et al., 2012).
According to Dang Vu, Schabus, Desseilles, Schwartz and Maquet (2007), frontal
deactivation during non-lucid dreams relates to aspects of dreaming such as “the uncritical
acceptance of bizarre dream content” and “the delusional belief of being awake during
dreams” (p. 104). To the best of current knowledge, regions involved in LDs — anterior
prefrontal, parietal and temporal cortex — are related to metacognitive processes in the
waking state (Baird, Mota-Rolim, Dresler, 2019).
Moreover, there are activations in a specific portion of the frontal cortex, namely the
right dorsolateral prefrontal cortex (dlPFC), which occur during LD (Dresler et al., 2012), and
do not usually occur during normal REM sleep (Voss et al., 2009). Moreover, studies have
shown that stimulating dlPFC increases lucidity during REM sleep (Stumbrys, Erlacher, &
Schredl, 2013). Importantly, activations in dlPFC are related to executive functions that can
be identified with volition, self-reflective awareness, and insight (Voss et al., 2009), and with
a wide range of higher cognitive capacities (Dresler et al., 2012).
It seems that LDs involve activations typically connected with higher cognitive
functions, which is compatible with considering the dreamer’s realization a successful
reasoning of some kind. In support of this claim, it is worth noting that LDs tend to happen
during dreams of “high cognitive and/or affective intensity” (Cohen & Price, 1988, p. 112).
These neuropsychological findings look promising, but they do not dispel the doubts
we had with the experiential data. Besides the general problem of “reverse inference” in
neuropsychology (Poldrack, 2006), it might be that specific parts of the brain activate for
some external reasons not related to reasoning of any kind, nevertheless giving the dreamer
the experience of intellectual achievement. Additionally, the presented neuroscientific studies
focus on LD in general, not on LD initiation, so we do not know if relevant brain areas are
active during the transition to lucidity, or if they just activate later on.
There is a need of finding a proper argument that would justify the claim that dreamers
can make successful reasoning about their current experience and draw the conclusion of “I
am in a dream,” and with that reasoning, the LD will start. Proving that is proving that
cognitive realization is in fact a possible route of LD initiation.
A convincing argument can be found among techniques of inducing LD. There are
three classes of methods to induce lucidity: cognitive techniques, external stimulation and
miscellaneous techniques (Stumbrys, Erlacher, Schädlich, Schredl, 2012). Among the
cognitive techniques, there is lucid-awareness training involving the dreamer’s critical attitude
(Cohen & Price, 1988), now dubbed the “reality testing” or “reflection” technique (Stumbrys
et al., 2012). It was first proposed by Paul Tholey (1983), who based his claim on over a
decade of research (LaBerge, 1985). According to Tholey’s idea, developing a reflective
attitude toward the current experience of waking life by asking oneself repeatedly — at least
five to ten times a day — if one is dreaming or not, results in the appearance of this attitude
during the actual dream state. Although some subjects have reported being so susceptible to it
that they had LD on the first night, on average, the technique starts to work after a month
(LaBerge, 1985). Contemporary research supports its general effectiveness (Stumbrys et al.,
2012). What is important for this technique is that the critical question of “Am I dreaming or
not?” should be asked especially in situations that seem dreamlike (LaBerge, 1985).
If we understand the cognitive realization as an endproduct of successful reasoning,
the effectiveness of the described technique supports the relevance of cognitive realization in
LDs. Asking oneself whether one is dreaming or not is in fact triggering the process of
reasoning — to answer this question correctly one must draw a correct conclusion based on a
set of premises, including the content of the experience and one’s prior knowledge. Actively
doubting if the current experience is real or not (or, to be philosophically safe: of the waking
or dreaming kind) is, in fact, a training of reasoning, which will be made clear in the last
pages of this article.
Additionally, in favor of the claim about the relevance of the cognitive realization, the
full version of Tholey’s approach also involves improving dream recall (Cohen & Price,
1988), which corresponds to the fact that one of the major triggers of LD initiation, mentioned
earlier, is recognizing the dreamlike quality of on-going experience (LaBerge, 1985). If
training in dream recall helps in inducing LDs, then this kind of “dreamlikeness” recognition
seems to be not merely an epiphenomenon of LD initiation, but a relevant part of the
mechanism of this initiation. If a dreamer can easily recollect their previous dreams, they
obtain a new premise about the current experience (namely, that it does seem similar to other
dreams) that helps to draw the correct conclusion of “I am in a dream”.
Another trigger, from the minor ones, that seems to be related to dream recall is the
déjà reve, the dreamer’s subjective impression that he or she is “relieving an experience he [or
she] has had previously while asleep” (Funkhouser & Schredl, 2010, p. 60). However, the
déjà reve is not just a simple recollection of a dream, but a more nuanced experience,
similarly to déjà vu not being simply a recollection of previous events. According to Schredl
et al. (2017), experiencing déjà reve is experiencing an uncanny similarity to a dream dreamt
before, even though this dream could not be remembered before the experience took place.
Considering that, we should be cautious in linking dream recall to this phenomenon easily.
The following section develops this discussion.
To sum up what has been said so far, the cognitive realization turns out to be relevant
for the problem of LD initiation. The onset of lucidity in the moment of realization can be an
outcome of a successful reasoning with a conclusion that one is dreaming. It should be
emphasized, though, that 1) there are other mechanisms that can also be responsible — in
other, more outstanding cases — for the initiation of LD; 2) not every experience of
realization is necessarily a case of cognitive realization.
Triggers involving cognitive realization
As we have already seen, some lucid dreams are triggered by perceptual
inconsistencies, which seem central to the discussion of cognitive realization. Cases of
perceiving inconsistencies — or incongruities (Green, 1968) —involve the dreamer finding an
element from the content of the dream “incongruent with the laws of waking reality” (Cohen
& Price, 1988, p. 111). The earlier example of Santa Claus and his elves is representative of
this kind of trigger. Another example, given by Cohen and Price, is the perception of “flying
without mechanical aids” (ibid.). Let us look at the Table 1, where this example is depicted.
Table 1
An example of dreamer’s reasoning
1: This table is floating in the air.
2: Tables do not typically float in the air. It is
3: Impossible things happen mostly in dreams.
Conclusion: I am in a dream.
The example has the form of a regular schema of reasoning, with a set of premises,
and the conclusion that is drawn basing on those premises. The cognitive realization that “I
am in a dream” marks the point of LD initiation, as the dreamer finds themselves capable of
marking the current experience as a dream. But could this schema cover more cases then the
given one? Answering this question requires examining other triggers of LD initiation.
The second major trigger is the case of dreamer’s recognition of the dreamlike quality
of their current experience. As Green (1968) describes it, this recognition is spontaneous and
involves noticing that the situation possesses the quality of “dreamlikeness.” Adams and
Bourke (2020) point out to two important cases of this dreamlikeness realization: the feeling
of déjà rêvé discussed above and a realization of control: the dreamer’s realization that they
can manipulate the dream content, accompanied with a “heightened sense of agency” (p. 14),
as compared to wakefulness. This heightened sense of agency probably results from the
extraordinary experience of control over the dream content (Adams & Bourke, 2020), which
seems to fit any other case of anomaly. Of course, there is a problem of where does the said
control come from, but given our minimal definition of LD, the control itself is not enough for
the dream to be lucid (we will get back to that problem later). If we were to establish a more
universal schema of reasoning based on the perception of inconsistencies, including the
perception of control over the content of the dream, the schema would present itself as in the
Table 2.
Table 2
Schema of anomaly-based reasoning
1: There is something odd going on.
2: Current events seem inconsistent or bizarre.
3: Inconsistent and bizarre events happen mostly in dreams.
Conclusion: I am in a dream.
However, when it comes to the feeling of déjà rêvé, it cannot be simply equated with
perception of inconsistencies, and the same goes for the spontaneous recognition of
“dreamlikeness” of the experience, since the trigger of dreamlike quality does not necessarily
involve any evident anomalies (Green, 1968). Importantly, there might be certain cases in
which everything about the experience seems completely normal, in the sense that there are
no anomalies in the content of the dream and there is no dreamlike quality that could be —
spontaneously or not — recognized, and yet the dreamer realizes that they are dreaming.
According to LaBerge (1985) déjà rêvé occurs when the dreamer experiences an actual or
apparent recollection that they have had a similar dream before, which suggests that there are
situations in which the dreamer experiences déjà rêvé unfounded in previous dream: the
dreamer realizes that they already had this dream before, when in fact they did not. Such cases
of misrecolletion or unfounded recognition cannot be included within the notion of cognitive
realization (even if the dreamer experiences persuasive feeling of realization). Moreover,
cognitive realization cannot be possibly relevant for the case of falling asleep directly to LD
(mentioned in Section 1, “The notion of realization in lucid dreams”), too. It must be admitted
here that these two types of LD initiation cannot to be explained by the mechanism discussed
But now let us consider those cases, that do involve a dreamlike quality of experience
and those cases in which the content of the dream does include some elements known from
previous dreams. Table 3 presents a schema of reasoning based on such premises.
Table 3
Schema of dreamlikeness-based reasoning
1: There is something odd going on.
2: Current events seem dreamlike or familiar because they were already seen in a dream.
3: Dreamlike events and events that repeatedly occurred in prior dreams are probably a
part of another dream.
Conclusion: I am in a dream.
This schema depicts the situation in which the dreamer correctly recognizes the
experienced content to be a dream because, to put it simply, it does feel “dreamy”. Here the
premises for the conclusion of “I am dreaming” are included in the dream’s content not as
inconsistent, unsettling events or things, but as events or things that the dreamer associates
with prior dreams or dreaming in general. An immediate question is whether there are cases of
a dreamlikeness-based reasoning accompanied by the feeling of déjà rêvé or not. Or, to put it
differently: are some cases of déjà rêvé cases of actual recollection? This problem is
intriguing, but not for the sake of this discussion. For now, it is only important that the
mechanism of cognitive realization includes an actual recollection of a previous dream,
independently of whether the dreamer had the déjà experience of having this dream before, or
We can now join Tables 2 and 3 to establish an universal schema of the dreamer’s
reasoning. However, before we get to that, let us first consider the last class of major triggers,
and one minor trigger from our original list (as seen in the Section 1, “The notion of
realization in lucid dreams”).
This third class of major triggers, which — as we will see — can be subsumed under
the discussed mechanism of cognitive realization, includes all kinds of emotional arousal and
is the most common class among dreamers that have not professed an interest in LD (Green,
1968). Although LaBerge speaks about emotional arousal in general (LaBerge, 1985), the
relevant experiences are mostly anxiety and stress (Cohen & Price, 1988; Green, 1968) as the
arousal is typically a result of a nightmare (Green, 1968).
Green (1968) notes that the fact that LD can be triggered by the unpleasant nature of a
dream’s content corresponds with the experience that people sometimes have in waking life
while under heavy stress. It is a feeling of hope that whatever is going on right now will turn
out to be a narrative of a dream; that one will wake up soon and realize that it was just a
dream. It seems that merely experiencing something stressful or scary can have a similar
impact on the mind as experiencing incongruent with laws of waking reality, bizarre events.
This connection is especially vivid in cases of heavy derealization related to occurrences of
anxiety and panic attacks (Hoyer, Braeuer, Crawcour, Klumbies, Kirschbaum, 2013). A study
on different triggers of LD (Adams & Bourke, 2020) shows that the trigger of nightmare does
in fact correspond with the dreamer’s attitude of denial towards the ongoing experience: “I
was telling myself (...) that it was only a dream and I needed to wake up” (p. 13).
In the light of this remark, one might expect that intense emotional arousal makes the
dreamer more critical about what is currently happening, both in the dream and in the
situation of heavy stress in waking life. In other words, it may be that lucidity in nightmares
and other dreams involving intense emotions is more likely to occur because intense emotions
make one more likely to make a critical judgment about the current experience. Therefore, the
schema of anomaly-based reasoning from Table 2 remains the same.
However, apart from the unusual case of falling asleep directly into LD and déjà rêvé,
which has been mentioned and detached from the mechanism discussed earlier, LaBerge
(1985) describes one more (minor) trigger that can result in lucidity: a situation in which the
dreamer associates an event of touching something with the initiation of LD.
It seems that if touching any kind of thing could induce lucidity, then such a case
cannot be subsumed under neither of the two schemas, as there is no relevant reasoning
involved. There are two examples of such a case evoked by LaBerge (1985) that are,
according to him, hard to classify. But are they really a stand-off?
From the two situations that LaBerge (1985) recalls, the first one — as he even admits
— involves elements typical for anomalies (bizarre elements). In the recollection, the dreamer
states that they saw a white dove that flew down and landed on their forehead. The dreamer
immediately found themselves in a state of lucidity.
The second case of allegedly different LD initiation was LaBerge’s personal
experience. He describes a dream in which he was approached by a bird as he was taking a
train by the sea. The bird landed on his open hand, and from this moment on, the dream was
lucid (LaBerge, 1985).
Those cases would be insufficient for stating that there exists another, special kind of
LD initiation even if they were completely unique and unfitting to the other cases, merely
because they are not well documented. Moreover, they actually do not seem unique and
unable to fit within the established schema from Table 2. As a matter of fact, they both share
features that fit premises from the schema: birds do not typically sit on people’s forehead or
on the hands of train passengers, unless they are well trained. The last minor trigger from our
list appears to be nothing more than a particular case of anomaly.
We can now join the schemas from Table 2 and Table 3, capturing all discussed
relevant triggers that lead to LD initiation. It will result in the final schema (Table 4).
Table 4
Combination of Tables 2 and 3
1: There is something odd going on.
2: Current events:
a) seem inconsistent or bizarre OR;
b) seem dreamlike or familiar because they were already seen in a dream.
3: Inconsistent, bizarre and dreamlike events and events that repeatedly appeared in prior
dreams occur mostly in dreams.
Conclusion: I am in a dream.
There are, however, at least three issues that arise if we consider the reasoning
depicted in the schema above the cognitive realization route of LD initiation.
First, as mentioned before, this mechanism does not cover all possible cases of the
onset of LD. There are situations in which the second premise is false or absent, yet LD is
initiated, and among these cases, the most important are those of unfounded realization and
those of falling asleep directly into LD. These cases cannot be explained by reasoning the way
all others can. It is possible that they involve some reasons external to the dream content, and
as such they must be explained by a different mechanism.
Second, it is possible and quite likely that at the same time many typical cases of LD
initiation in which second premise is true are not necessarily caused by any kind of reasoning.
This kind of LD initiation might even involve the same experience (sudden realization). What
is set at this point is that the various triggers included in the schema can be cases of cognitive
Third, there are too many situations in which the second premise is true, yet LD does
not initiate. In other words, if the final schema is correct, an inconvenient question arises: why
are LDs so rare? According to studies by Jayne Gackenbach and Thomas Snyder (1988), 58
percent of the population reported having experienced LD at least once in their lives, and 21
percent reported experiencing LDs at least once a month. A recent meta-analysis shows very
similar estimates (Saunders, Roe, Smith & Clegg, 2016). Considering the frequency of
incongruent, bizarre elements occurring in dreams, these numbers are too low in the light of
the established schema.
The third issue is indeed a bit of a difficulty, but there are two possible answers to it —
we will return to that matter as the proper model of reasoning is built. Before that happens,
however, we must examine whether the existing theories of LD initiation can or cannot
account for the route of cognitive realization.
Theories of lucid dream initiation
The first attempt at an explanation of LD initiation is from Oliver Fox, a poet and
occultist who lived at the turn of the 20th century. Fox (1962) documented his experiences
with LDs, and believed that the onset of LD comes from critical thinking. According to him,
perceiving an inconsistency in the surrounding narrative can result in lucidity if the dreamer is
not satisfied by any explanation other than “I am currently dreaming.” Although it does
describe the process of realization, Fox’s conception does not explain how the mechanism of
critical attitude exactly works: this theory says nothing more than that we are sometimes able
to correctly conclude that we are dreaming after encountering an inconsistency. Fox’s
conception is thus pointed in the right direction, but insufficient. The Bayesian model, which
will be developed here, is in some sense an extension of Fox’s theory.
The second idea, on the other hand, does not involve any sort of critical assessment or
search for explanation on the part of the dreamer. According to Thomas Metzinger (2003),
even though non-lucid dreamers can act on desires, they cannot see themselves as agents, and
so do not experience themselves as deliberately directing their actions or attention. A lucid
dreamer, on the other hand, possesses a stable first-person perspective, which enables one to
conceptualizing oneself as the thinker of thoughts and the owner of point of view; this
conceptualization is possible if the availability of self-related information increases.
Metzinger claims that during the transition from typical dream to lucid dreaming, what
changes is “the content and functional profile of the phenomenal self model” (Metzinger,
2003, p. 537).
Metzinger’s conception gives an account of LD which focuses on the relation between
a dreamer and their experiences, leaving the problem of realization untouched. Thus, this
theory can explain cases that do not involve any realization, but it does not account for the
mechanism involved when a dreamer realizes that they are asleep and dreaming in the sense
of conducting a reasoning. It might also be that the mechanism proposed by Metzinger works
in every LD initiation but is not explanative for cognitive realization — that is, it only
describes one of many cognitive aspects of change between normal sleep and lucidity.
The third conception, like Fox’s, focuses more on the mechanism of realization in the
initiation of LD. Using the distinction between operant thinking — which is deliberate and
goal-directed — and respondent thinking — which is involuntary and independent of goals,
Cohen & Price (1988) explain the dreamer’s realization as an activation of operant thinking.
This activation enables the dreamer to observe the content of the dream, which is produced by
respondent thinking. In this conception, the dreaming narrative is generated by respondent
thinking but it can be changed (to some extent) voluntarily, thanks to operant thinking. Cohen
and Price believe that operant thinking emergence is connected with an increase in self-
awareness, which might result in the dreamer questioning the reality of their ongoing
experience (Cohen & Price, 1988).
Thus, speaking in terms of modern dual-process theory, as depicted by Daniel
Kahneman (2011), we could say that according to Cohen and Price’s idea, the System 2
processes (slow, analytic and effortful) can sometimes notice that the ongoing System 1 (fast,
intuitive and effortless) narrative is in fact a dream (Windt, 2015). This is an elegant and
consistent idea, but it does not state explicitly how the mechanism of realization works in the
LD initiation, because it says nothing about the relationship between the elements of the
dream content and realization. In what conditions does System 2 decide that the content of
System 1 narrative must be a dream? The reasoning model developed here will resolve this
problem within a more recent framework, namely, the Bayesian framework.
Importantly, the theory by Cohen and Price, assumes that our metacognitive functions,
cognition about cognition (Dunlosky, Metcalfe, 2009), are not particularly deficient in the
state of dreaming compared to their function in waking life. This assumption was endorsed in
some studies, leading to the idea that there is not “dream cognition” and “waking cognition,”
but rather a continuous spectrum of cognitive changes in dreams and in waking life (Kahan &
LaBerge, 1994; Kahan & LaBerge, 1996; Kahan, 2001; Kahan & LaBerge, 2011). This is to
say that dreaming and waking thoughts are different in their content, but not necessarily in the
cognitive processes that they involve. Windt (2015) differentiates between cognition and
cogitation, of which the first is the “associative and fluid processing characteristic of System
1” (p. 444), relating in dreaming to the generation of dreaming narrative, while the second
involves metacognition and — generally — processes typical for System 2. While it seems
that the latter is “less prominent and highly unstable in dreams” (ibid.), according to Windt,
cogitation is not absent in dreams, but rather is mixed with cognition in a continuous range.
As it would probably be false to claim independence of System 2 processes in dreaming,
cogitation in dreams does not fall entirely into the System 1-like stream of cognition (Windt,
We will later see that this claim — that dreams are cognitively and metacognitively
similar to wakefulness — is controversial. For now, however, let us assume that it is roughly
correct. Now, if metacognitive functions, such as thought monitoring (Filevich, Dresler,
Brick, Kuhn, 2015), are available in dreaming in a manner similar to in wakefulness, then we
need to face the paradoxical fact that the dreamer is sometimes (rarely) able to evaluate its
current experience and realize that they are asleep and dreaming, and sometimes (usually) not.
Thus, the struggle with the problem of rarity of LD continues: why are LDs rare given that the
dreamer is able to realize that they are dreaming every time they notice something bizarre in
the content of the dream? One answer to that problem will be presented in the Bayesian
model, developed throughout the next sections.
Bayesian model of realization in lucid dreams
In the schema that I have established before (Table 4) the second and third premises
involve an intuitive description of events that, if observed, can trigger LD initiation. It seems
reasonable that perceiving inconsistent or dreamlike occurrences can make one doubt about if
they are real, but what does exactly “inconsistent” or “dreamlike” mean? Do we all have the
same standards of regularity?
A rather new theoretical approach to the inquiry of cognition, namely, predictive
coding, might give a clue to answer these questions. Recently, cognitive science has been
adopting the general idea that the brain, “unbeknownst to consciousness, engages in a process
that approximates Bayesian inference” (Andersen, 2017, p. 7) to predict the incoming sensory
information based on its prior expectations. In that approach, perception work not only
bottom-up, but also top-down: when the sensory input is ambiguous, our expectations towards
the world can shape what we see, hear, touch, smell etc., to the point where they could
completely dominate the processes of perception making one see what they believe is
happening (Andersen, 2017). Crucially, the predictive coding entails that while perceiving, the
brain constantly compares different, competing representations with the input and the
representation that wins as the most probable, given the current input, is what is consciously
experienced as the model of the world (Hohwy, 2013).
The framework of predictive coding provides a model of perception, but also of other
cognitive processes, such as reasoning. If indeed the brain resembles as a Bayesian inference
machine that computes the probabilities of different representations, the processes of
reasoning could be modeled as a comparison and computation of data, expectations and
representations that act as potential hypotheses.
Hohwy (2013) describes that idea, the idea of probabilistic reasoning, using a short
scenario: a man is sitting in an empty house and hears a strange sound from behind the door.
The man’s task is to figure out what is causing that sound. Each of the possible causes is
called a hypothesis, and all hypotheses come from the man’s general expectations towards the
world: the “prior knowledge” (Manktelow, 1999) or the “prior beliefs” that they hold about
the world. Now, some of these hypotheses have a higher prior probability than others. Prior
probability is the “probability of the hypothesis prior to any consideration of its fit with the
evidence” (Hohwy, 2013, p. 16).
On the other hand, there is the likelihood of the hypothesis. While the prior probability
reflects the subjective estimation about how probable is it that something happens, the
likelihood of the hypothesis is the probability that if a given hypothesis is true, the effect
shows up. Let us mark the strange noise from the example as “B” and some hypothesis as “A”
— the likelihood of A is the probability of B|A, then.
Now, there are some hypotheses that fit an effect extremely well but are at the same
time extremely improbable independently of that effect. In Hohwy’s example, a hypothesis
with a very high likelihood would be that there is machinery outside the house designed to
make such noises to make the inhabitant wonder what is causing the sound. However, that
hypothesis seems very improbable when considered prior to any consideration of its fit with
the data. The high likelihood is not a sufficient condition for the brain to choose the best
hypothesis, for if it relied only upon how likely it is that effect will show up given a certain
hypothesis, it would choose hypotheses that are extremely improbable when considered alone.
A hypothesis that has both a high likelihood and a high prior probability is that there is a
person outside making the noise — of course, as long as we are not dealing with a “last man
on Earth” scenario.
I addressed the question of what “inconsistent” or “dreamlike” means and if all human
beings have the same standards of regularity. Based on the theory of probabilistic reasoning, I
can now replace the notions such as inconsistency or dreamlikeness with the terms of
subjective probability, which results in a new schema (Table 5).
Table 5
Schema of reasoning from Table 5 improved with Bayesian terms
1: There is something odd going on.
2: Current events are improbable under the hypothesis that I am awake.
Conclusion: I am in a dream.
What the dreamer therefore estimates is the posterior probability: the probability of A
(“I am dreaming”) under condition that B occurred (e.g. the dreamer met Santa Claus and his
elves). This posterior probability, as it was mentioned, is based on two main ingredients:
likelihood and prior probability (or just prior) (Hohwy, 2013). The likelihood, following the
example, would be the probability of (B) spotting Santa Claus and his elves under the
hypothesis (A) of “I am asleep and dreaming,” while the prior would be the probability of A
Now, according to the model, crossing a certain threshold of calculated posterior
probability would result in realization that one is asleep and dreaming, and, respectively, a
low value of posterior probability would lead the dreamer to conclude falsely that they are
awake. With use of Bayes’ rule (Griffiths, Kemp, & Tenenbaum, 2008) we can depict the
relevant calculations as shown in Figure 1.
Figure 1
Bayes’ rule
Note. A Hypothesis of being in a dream. B Event that seems inconsistent or
dreamlike or seems familiar because it was already seen in a dream. P(A) Prior
probability; probability of being in a dream. P(B) Probability of occurrence of B.
P(B|A) – Likelihood of A; probability of occurrence of B under the condition of being
in a dream. P(A|B) Posterior probability; probability of being in a dream under the
condition that B occurred.
A problem appears, though, because the dreamer does not really consider B
improbable — they consider P(B|~A) improbable, where ~A means the hypothesis of “I am
awake” or “I am not dreaming.” Hypothesis A, unlike most of our hypotheses about
explanations of events, says something about the whole epistemic situation of the system that
calculates probabilities. In this light, there is no P(B) alone, only P(B|A) and P(B|~A). This
problem can be solved if we use another Bayesian equation (Figure 2), derived from the
standard Bayes’ rule (Jeffrey, 2002). On the left side of the equation in Figure 2 there is a ratio
of the probability of hypothesis A under the condition that B occurred to the probability of
hypothesis ~A under the condition that B occurred. Let us name this ratio the Reality-check
ratio (RCR). The RCR depends on the value of two other ratios:
, which will be respectively dubbed the Cartesian ratio (CR) and the Bizarreness ratio (BR).
Figure 2
Bayes’ rule after transformation
Note. AHypothesis of being in a dream. ~A – Hypothesis of being awake (not being
in a dream). B – Event that seems inconsistent or dreamlike or seems familiar because
it was already seen in a dream. P(A) – Prior probability of A; probability of being in a
dream. P(~A) Prior probability of ~A; probability of being awake. P(B|A)
Likelihood of A; probability of occurrence of B under the condition of being in a
dream. P(B|~A) Likelihood of ~A; probability of occurrence of B under the
condition of being awake. P(A|B) Posterior probability; probability of being in a
dream under the condition that B occurred. P(~A|B) – Posterior probability;
probability of being awake under the condition that B occurred.
The model predicts that crossing a certain value of RCR should result in hypothesis A
winning, which would lead the dreamer to the conclusion “I am dreaming.” Let us present
how it would work in the example of Santa Claus. The event of spotting Santa Claus (with a
host of elves) takes place in a dream. According to the model, the dreamer conducts a
reasoning, which is modeled as a calculation of probabilities, as depicted above. The dreamer
puts together how probable it is that Santa Claus exists given that this is a dream or
wakefulness, and in contrast how probable it is that they are dreaming compared to the
probability that they are awake. The calculation of probabilities leads to establishing the value
of the RCR, which might or might not exceed a certain threshold. If it does, the dreamer will
draw the hypothesis of “I am dreaming” as the conclusion of their reasoning, while if it does
not, the dreamer remains in the state of a normal dream.
Let us assume that the dreamer is a child and believes in Santa Claus. Then, the
probability of B|A will not be as low as in the case of a typical adult who does not believe in
Santa Claus, and therefore will hold a high P(B|A) and as low P(B|~A) as they allow the
possibility that Santa actually exists.
The typical adult will then have a high BR (as they do not believe in Santa Claus) but
given that they are not a trained lucid dreamer, the CR will be extremely low, as the prior
probability of being in a dream is low, and the prior probability of being awake is high. Thus,
the adult will not realize that they are in a dream because the resulting RCR will be very low.
This is the point in which the model gives a possible explanation to the problem of
LDs’ rarity. It seems that what this model says — similarly to Cohen and Price’s theory — is
that the dreamer should infallibly draw the conclusion “I am in a dream” every time
something unusual happens in the dream’s content. The problem with such an explanation, as
mentioned before, is that LDs are rare. If dreamers could always infallibly draw the
conclusion that they are in a dream (given that metacognitive functions are not absent in
dreaming) after spotting some improbable events, LDs would be as frequent as the bizarre
events happening in dreams are; yet they are rare.
How can this model deal with this problem? According to the model, if the RCR is
high enough, the dreamer realizes that they are asleep and dreaming. This means that
cognitive realization can happen in LD only if the BR and/or CR reach certain threshold(s), so
only if the dreamer notices something extremely bizarre and/or they generally hold a
relatively high belief that they are currently asleep and dreaming.
It would seem that people generally do not hold any beliefs on if they are awake or in
a dream; that no one (with the possible exception of Rene Descartes, the namesake of the CR)
holds prior beliefs about their current epistemic situation. Asking someone if they are
currently dreaming results in most cases with them stating that they are positive that they are
awake. But the same holds with people that are in fact dreaming – they “tacitly assume that
they are awake” (LaBerge, 1985, p.152). It might be that our prior beliefs about being in a
dream or awake are in some sense invisible, because we consider the probability of being in a
dream so small that only very incongruent, unsettling events can make us doubt if events
around us are really happening.
People hold prior beliefs about their epistemic situation — they believe that it is highly
improbable that they are currently dreaming. This belief is consistent with the fact that LDs
are rare events; they are rare because the dreamer is in some sense tied to their belief that they
are awake and that their experience comes from the real world. An argument for the statement
above comes — again — from the techniques of inducing LD. If it is possible to make LDs
happen more frequently by asking oneself repeatedly “am I dreaming or not,” it seems that the
admissibility of the hypothesis of being in a dream can increase. Reality-testing LD inducing
techniques make people more open to the possibility that they are currently dreaming; their
P(A) increases (and so does their CR).
To sum up what has been said so far: I hold that mechanisms of LD initiation include
the route of cognitive realization, which is, in fact, an endproduct of a successful reasoning
made by the dreamer after noticing bizarre, dreamlike events or events that have been
experienced in previous dreams. This route does not account for all kinds of LD initiation but
must be considered given the effectiveness of reality-testing techniques. This mechanism of
cognitive realization can be modeled in a Bayesian framework — the model predicts that the
dreamer realizes that they are asleep and dreaming if the reasoning, modeled with calculation
of subjective probability, leads them to the hypothesis “I am asleep and dreaming.” Two
factors are important for this calculation of probability: the prior probability of being in a
dream/awake (CR) and the calculated probability of noticing a bizarre event given that the
dreamer is in a dream/awake (BR).
Noteworthy, the model should not be understood as a logical structure that human
beings should follow in order to dream lucidly, though it can provide some clues in that
matter. The Bayesian model of realization implies that dreamers perform successful reasoning
with a conclusion of “I am dreaming” and that underlying processes of this reasoning can be
modeled as computations on probabilities. By no means does it imply that dreamers engage in
conscious calculations of probabilities, though the reasoning on the conscious level might
have a language-like, representational structure.
This model solves the problem of LDs’ rarity by pointing out that not only is the
bizarreness of dream events important, but the priors that the dreamer holds about their
epistemic situation — which are by default “I am awake”, hence the low CR — are also
substantial. However, it will be now discussed that this problem might have a second solution
whatsoever, related or unrelated to the Bayesian model.
Discussion of the model
The presented Bayesian account of realization in LD, which claims that successful
reasoning based on the calculation of probabilities is the mechanism of said realization, forges
the common view into a proper model. It is not, however, fault-free.
The main problem lies in the question whether this model is indeed applicable.
Dreams in general seem to be so variable that a unified model of LD might be always found
wanting, similar to models of such complex and erratic phenomena as culture or religion,
which on many occasions have been criticized as oversimplified. The current model does not
cover all cases of LDs, but it never promised to do so; its only obligation was to depict and
explain the case of LD initiation which involves the dreamer’s cognitive realization. It might
be argued, however, that even this typical case is too complex of a phenomenon, considering
that realization in LDs is of various kinds. According to Windt (2015), lucidity can be of a
purely experiential quality (A-lucidity) or explicit (C-lucidity), and sometimes the reasoning
involved seems to be on the right track but does not reach the correct conclusion (pre-lucid
dream) which could be depicted by the faulty reasoning of “I am now a dog, so it cannot be a
dream.” Additionally, perhaps dreamers could engage in faulty reasoning (e.g. “I am a human,
so it must be a dream”) and nevertheless end up in lucid dream through this.
This problem hinges on another. It was assumed before (see Section 4, “Theories of
lucid dream initiation”) that metacognitive functions can work in dreams similarly to
wakefulness and it has been suggested that generally, high order cognition can precede
lucidity (Adams & Bourke, 2020). However, as opposed to the studies mentioned earlier,
there are studies indicating that metacognition is unstable in the majority of dreams (Windt,
2015), less pronounced during sleep (Perogamvros et al., 2017), impaired during the REM
sleep phase (Desseilles et al., 2011) or that it does not work in normal dreams at all (Windt &
Metzinger, 2007) while it does in LDs (Filevich, Dresler, Brick, Kuhn, 2015). But if
metacognition is completely absent in normal dreams, then the effectiveness of the reality-
testing technique is paradoxical: if it is possible to train oneself in recognizing that the current
experience is a dream, which must be related to the metacognitive function of thought
monitoring, at least some metacognitive functions must work in dreams. Hence, a possible
explanation of this discrepancy would be that that the metacognition in dreams can work the
same as in wakefulness, but it usually does not. In other words, metacognitive functions are
not necessarily absent in normal dreams, but they can be impaired and they usually take on a
different, distorted shape. This would be coherent with the suggestion that System 2 processes
are in dreams mixed with System 1 processes (Windt, 2015) and with the suggestion that
metacognition in dreams can be divided to System 1 low-level metacognition and System 2
high-level metacognition (ibid.).
Now, if that is correct, then we can solve the problem of different kinds of realization.
If metacognitive functions are not absent in dreams and they are might be distorted, than the
Bayesian model can still be applied, as monitoring one’s current experience (leading to
relevant reasoning) is still possible in dreams, while the qualitative difference of reasoning
within different dreams can account for different types of lucidity and pre-lucid dreams. A
possible empirical study that could address this hypothesis will be discussed in the last
However, the discussion on metacognition draws us back to the problem of LDs’
rarity. If metacognitive functions do not work the same during sleep and in waking life, then
LDs can be said to be rare because the dreamer rarely has the chance to realize that they are
dreaming. Given that answer, the solution presented within the Bayesian model might turn out
to be useless. However, the deficits of metacognition do not present any explanation for the
question of how it is possible to train oneself in dreaming lucidly via obtaining a critical
attitude towards reality. I propose here that the answer to the problem of LDs’ rarity might be
dual: not only are our metacognitive functions less pronounced and distorted during sleep in
general, but additionally, it is not easy to realize that one is dreaming because of the default
low level of CR. An interesting hypothesis is that the reality-checking technique changes the
value of CR and at the same time affects the in-dream activity of brain functions related to
metacognition. More studies covering the relationship between metacognition and dreaming
are needed to allow for more careful consideration of that matter.
It seems that dreams, generally speaking, are multifactor phenomena, which are hard
to grasp in a unified model. However, that probably should not discourage scientists and
philosophers from trying to do so. Lucid dreaming is a phenomenon that relies on many
different variables. The Bayesian model cannot possibly explain the place of all these
variables, as it is only an idealization, aspiring not to a holistic explanation of LDs, but to a
more modest explanation of a particular phenomenon: the cognitive realization in dreams. The
model tries to account for this phenomenon appealing to only two variables: a dreamer’s prior
beliefs about the world and about their current epistemic situation. In the research on
phenomena as complex as dreams, a unified model can be easily denied on the basis of the
claim that it simplifies the nature of explanandum and does not work in atypical, minor cases.
This however can lead to what has been called by anthropologist Mary Douglas “bongo-
bongoism,” the practice of countering every generalization with a minor, outstanding
exception (Douglas, 1997), which seems to be sterile if even a working model is ever to be
Thus, the model put forth in this work by no means tries to explain the phenomenon of
LDs in general or to consider every variable relevant to the processes of dreaming. Moreover,
it has several constraints.
First, this model does not consider the relationship between the capability of the
dreamer to tell that they are asleep and dreaming and the aspect of deliberate action or control
of the dreaming narrative. Windt’s stronger definition (see Section 1, “The notion of
realization in lucid dreams”) encloses the ability to control the dream and the ability to act
voluntarily is mentioned on different occasions in the context of LD (Dresler et al., 2012;
LaBerge, 1990; LaBerge, 1985; Green, 1968). There might be a specific mechanism that binds
the modeled reasoning with the availability of control, but the current Bayesian model does
not account for this connection.
The second constraint is that is seems that the Bayesian model has little or no
empirical application due to the apparent impossibility of controlling the variables of prior
beliefs, but even more due to the elusive, multifactor, and uncontrollable nature of dreaming
itself. Testing the model directly would require conducting a study that measures the prior
probabilities held by the dreamer or evokes them experimentally and then relates them to the
frequency of LDs triggered by relevant events within a dream. Given contemporary
technology, this endeavor seems futile.
It seems, however, that there are practical ways to test the model, though they are
mostly indirect. In the following final subsection, we will see possible empirical inquiry based
on the assumptions and predictions of the established model.
Possible empirical applications
The Bayesian model is based on the assumption that metacognitive functions in
dreams work similarly enough to those of waking life to allow a dreamer to evaluate their
current experience and conclude that it is a dream. I have established, however, that given the
studies opposing this view and the apparent paradox of the reality-testing technique, it must
be true that metacognition works at least differently (1) in dreams than in waking life, and (2)
within different dreams. Proving this could account for various kinds of lucidity, giving the
Bayesian model a chance to cover them despite the differences.
The first claim seems plausible in light of previous remarks. The latter, more specific
to the model, would require testing to determine if there indeed are different kinds of
metacognition in dreams and if they relate to different kinds of lucidity. This could be
achieved by evaluating subjective reports for various aspects of mental states (Kahan,
LaBerge, 1996). Controlling for the conscious expressions of reasoning (explicit inference)
would probably be fruitful, too, as differences in metacognition are expected to relate to broad
differences in reasoning about the current experience. Proving that there are differences in
metacognition related to different types of lucidity would ground the model deeper in
empirical premises, while proving the opposite would create a demand to explain why there
are various types of lucidity (and so-called pre-lucid dreams) that seem to exploit the same
mechanism of realization with different outcomes.
This could be also tested with the methods of cognitive neuroscience. There are
studies on metacognition in lucid dreams and normal dreams (Filevich, Dresler, Brick, Kuhn,
2015) but to the best of my knowledge, none have compared neurocorrelates of different types
of lucidity, especially concerning metacognition.
Neuroscientific methods could also be applied for testing the sole mechanism of the
Bayesian account. We have seen some studies on neural correlates of LDs in general, but what
the Bayesian model depicts is the moment of transition from normal dreaming to LD.
Examining this moment for neural correlates would allow the observer to track if there is
specific activation in regions responsible for, e.g., metacognition, just before the dreamer
gives the first signs of dreaming lucidly with the method first used by Hearne (1978).
Possible experimental studies could also be made with respect to the prior beliefs,
which are the core variable in the mechanism of realization, and their effect on the frequency
of LDs. The Bayesian model predicts that approaching something that the dreamer considers
very improbable can result in their realization that they are currently asleep and dreaming.
Now, most people would claim that spotting a dinosaur is a very improbable event. But it
seems plausible that spotting a dinosaur repeatedly — even on a screen or perhaps in virtual
reality — could somehow change one’s prior belief about dinosaurs, making one less
suspicious about approaching a dinosaur, which in effect will result in fewer LDs initiated by
meeting an extinct saurian. Though such a fan of dinosaurs might not really believe that
dinosaurs exist in the present, their implicit estimation of this probability could become less
sensitive with every dinosaur-related movie they watch. Proving that experimental
manipulation of this kind (not necessarily related to Jurassic Park) has an effect on the
frequency of LDs would not only validate the prediction of the Bayesian model, but also
suggest the strength of the effect of allegedly irrelevant everyday experience on what we
perceive as probable and improbable.
At the same time, there probably exist some events that — in opposition to a modern-
day dinosaur — we consider probable in waking life, but very improbable in a dream. A
common example is pain: one who does not believe what they have just seen often asks
another to pinch them. It may be that most people possess a prior belief that one cannot feel
pain in a dream, so the experience of pain temporarily lowers the prior probability of the
hypothesis “I am in a dream,” even though dreamlike events are currently taking place. This
issue is not so easy to transfer to empirical grounds in the field of LDs, but it does seem to be
a matter for further investigation, not only related to LD initiation but to the general topic of
identifying situations as real or unreal.
In light of these examples, the Bayesian model seems to have potential empirical
Concluding remarks
This article sought to forge the commonsense notion of realization in LD, which has
been widely used with little consideration, into a proper model of LD initiation. The view that
LDs start when the dreamer realizes that they are currently asleep and dreaming can be
understood as a view that the dreamer can make a kind of reasoning about their current
experience and successfully conclude that what they are experiencing is a dream. Until now,
there has not been a model which could account for that phenomenon.
I first argued, based on studies on the reality-checking technique of LD initiation, that
cognitive realization — the “I am dreaming” conclusion of successful reasoning — is a route
of LD initiation. Thus, the phenomenon of LD can occur due to a dreamer’s critical evaluation
of what is currently happening. It is important to emphasize, however, that because there are
cases of lucid dreams that do not fit the notion of cognitive realization, this mechanism cannot
account for every kind of lucid dream initiation.
Second, I established the model which predicts that the dreamer conducts a reasoning
that can be modeled within a probabilistic, Bayesian framework. After noticing bizarre or
dreamlike events (or events that occurred in previous dreams), the Reality-check ratio is
calculated depending on two factors: the values of the Bizarreness ratio (the calculated
probability of noticing a bizarre event given that the dreamer is in a dream/awake) and
Cartesian ratio (the prior probability of being in a dream/awake). If the Reality-Checking
Ratio exceeds a certain threshold, the dreamer concludes the hypothesis “I am dreaming”
In the last part of the article, the constraints and problems of the model have been
discussed, as well as its possible empirical application. Testing predictions and assumptions of
the Bayesian model may require innovative methods but is by no means impossible.
It is notable that the matter discussed in this manuscript is interesting not only in the
context of LDs, but in the more general context of scientific inquiry on the mind. Phenomena
such as derealization, as well as new technologies of virtual reality are probably related to the
topic of how one perceives the surrounding world as reality or mere illusion. The Bayesian
framework, which has been adopted here as probabilistic reasoning, can probably be fruitful
in the broad field of explaining the marks of dreamlike and waking; virtual and actual; real
and unreal.
I would like to thank Michał Klincewicz and Mateusz Hohol for insights related to the
framework proposed in the manuscript.
Adams, L., & Bourke, P. (2020). Examining the triggers of lucid insight. Dreaming, 30(2),
Andersen, M. (2017). Predictive coding in agency detection. Religion, Brain & Behavior,
9(1), 65-84.
Aristotle. (350 BCE/1918-2000). On dreams. (J. I. Beare, Trans.). Retrieved from [31.01.2020]
Baird, B., Mota-Rolim, S. A., Dresler, M. (2019). The cognitive neuroscience of lucid
dreaming. Neuroscience & Biobehavioral Reviews, 100, 305-323.
Cohen, D.B., & Price, R.F. (1988). Lucid dream induction. An empirical evaluation. In: J.
Gackenbach & S. LaBerge (Eds.), Conscious mind, sleeping brain. Plenum Press.
Dang-Vu, T.T., Schabus, M., Desseilles, M., Schwartz S., Maquet, P. (2007). Neuroimaging of
REM sleep and dreaming. In D. Barrett & P. McNamara (Eds.), Praeger perspectives.
The new science of dreaming: Vol. 1. Biological aspects (pp. 95–113). Praeger
Publishers/Greenwood Publishing Group.
Desseilles, M., Dang-Vu, T. T., Sterpenich, V. & Schwartz, S. (2011). Cognitive and
emotional processes during dreaming: a neuroimaging view. Consciousness
and Cognition, 20(4), 998-1008.
Douglas, M. (1997). Natural Symbols. Routledge.
Dresler, M., Wehrle, R., Spoormaker, V. I., Koch, S. P., Holsboer, F., Steiger, A., … Czisch,
M. (2012). Neural correlates of dream lucidity obtained from contrasting lucid versus
non-lucid REM sleep: A combined EEG/fMRI case study. Sleep, 35(7), 1017–1020.
Dunlosky, J., & Metcalfe, J. (2009). Metacognition. Sage Publications, Inc.
van Eeden, F. (1913). A study of dreams. Proceedings of the Society for Psychical Research,
26. Retrieved from
%20dreaming%20pdfs/vanEeden_PSPR_26_1-12_1913.pdf [02.07.2020]
Filevich, E., Dresler, M., Brick, T. R., & Kuhn, S. (2015). Metacognitive Mechanisms
Underlying Lucid Dreaming. Journal of Neuroscience, 35(3), 1082–1088.
Fox, O. (1962). Astral projection. University Books.
Green, C. E. (1968). Lucid dreams. Institute of Psychophysical Research.
Griffiths, T., Kemp, C., & Tenenbaum, J. (2008). Bayesian models of cognition. In: R. Sun
(Ed.), The Cambridge handbook of computational psychology (pp. 59–100).
Cambridge University Press.
Hearne, K. M. (1978). Lucid dreams: An exploratory study of consciousness during sleep.
Unpublished doctoral dissertation, University of Liverpool.
Hohwy, J. (2013). The predictive mind. Oxford University Press.
Hoyer, J., Braeuer, D., Crawcour, S., Klumbies, E., & Kirschbaum, C.
(2013). Depersonalization/derealization during acute social stress in social phobia.
Journal of Anxiety Disorders, 27(2), 178–187.
Jeffrey, R. (2004). Subjective probability: The real thing. Cambridge University Press.
Kahan, T. L. (2001). Consciousness in Dreaming: A metacognitive approach. In K. Bulkeley
(Ed.), Dreams: A reader on religious, cultural and psychological dimensions of
dreaming (pp.333–360).
Kahan, T. L., & LaBerge, S. (1994). Lucid Dreaming as Metacognition: Implications for
Cognitive Science. Consciousness and Cognition, 3(2), 246-264.
Kahan, T. L., & LaBerge, S. (1996). Cognition and metacognition in dreaming and waking:
Comparisons of first and third–person ratings. Dreaming, 6(4), 235–249.
Kahneman, D. (2011). Thinking, fast and slow. Farrar, Straus and Giroux.
LaBerge, S. (1985). Lucid dreaming. Ballantine Books.
LaBerge, S. (1990). Lucid dreaming: Psychophysiological studies of consciousness during
REM sleep. In R.R. Bootsen, J.F. Kihlstrom, & D.L. Schacter (Eds.), Sleep and
Cognition (pp.109–126). APA Press.
LaBerge, S., & Rheingold, H. (1991). Exploring the world of lucid dreaming. Ballantine
Manktelow, K. (1999). Reasoning and thinking. Psychology Press.
Metzinger, T. (2003). Being no one. The self-model theory of subjectivity. The MIT Press.
Saunders, D. T., Roe, C. A., Smith, G., & Clegg, H. (2016). Lucid dreaming incidence: A
quality effects meta-analysis of 50 years of research. Consciousness and Cognition,
43, 197–215.
Schredl, M. & Funkhouser, A. (2010). The frequency of déjà vu (déjà rêve) and the effects of
age, dream recall frequency and personality factors. International Journal of Dream
Research, 3(1), 60-64.
Schredl, M. T., Funkhouser, A., & Göritz, A. S. (2017). Frequency of déjà rêvé: Effects of age,
gender, dream recall, and personality. Journal of Consciousness Studies, 24(7-8), 155–
Schwartz, W., & Godwyn, M. (1988). Action and representation in ordinary and lucid dreams.
In: J. Gackenbach & S. LaBerge, (Eds.), Conscious mind, sleeping brain (pp. 419–
427). Plenum Press.
Snyder, T.J., & Gackenbach, J. (1988). Individual differences associated with lucid dreaming.
In: J. Gackenbach & S. LaBerge (Eds.), Conscious mind, sleeping brain (pp. 221–
259). Plenum Press.
Stumbrys, T.J., Erlacher, D., Schädlich, M., & Schredl, M. (2012). Induction of lucid dreams:
A systematic review of evidence. Consciousness and Cognition, 21(3), 1456–1475.
Stumbrys, T., Erlacher, D., & Schredl, M. (2013). Testing the involvement of the prefrontal
cortex in lucid dreaming: A tDCS study. Consciousness and Cognition, 22(4), 1214–
Perogramvros, L., Baird, B., Seibold, M., Riedner, B., Boly, M., Tononi, G. (2017). The
Phenomenal Contents and Neural Correlates of Spontaneous Thoughts across
Wakefulness, NREM Sleep, and REM Sleep. Journal of Cognitive Neuroscience,
29(10), 1766–1777.
Poldrack, R. A. (2006). Can cognitive processes be inferred from neuroimaging data? Trends
in Cognitive Sciences 10(2), 59-63.
Revonsuo, A. (1995). Consciousness, dreams and virtual realities. Philosophical Psychology,
8(1), 35–58. https://
Tholey, P. (1983). Techniques for inducing and manipulating lucid dreams. Perceptual and
Motor Skills, 57, 79–90.
Voss, U., Holzmann, R., Tuin, I., & Hobson, J. A. (2009). Lucid dreaming: A state of
consciousness with features of both waking and non-lucid dreaming. Sleep, 32(9),
Windt, J. M. & Metzinger, T. (2007). The Philosophy of Dreaming and Self-Consciousness:
What Happens to the Experiential Subject during the Dream State? In: D. Barrett & P..
McNamara (Eds.), The New Science of Dreaming. (pp. 193–247). Praeger
Imprint/Greenwood Publishers.
Windt, J. M. (2015). Dreaming: A conceptual framework for philosophy of mind and
empirical research. The MIT Press.
... Prediction Errors, Multisensory Integration, and Dynamic Attention Guide Lucid Dreams PC mechanisms were also proposed to account for LD experiences (26,37,38). In LD, prediction errors trigger updating of predictions that, at some point, reach a metacognitive self-model explaining the ongoing mental representations as dream images (37). ...
... Prediction Errors, Multisensory Integration, and Dynamic Attention Guide Lucid Dreams PC mechanisms were also proposed to account for LD experiences (26,37,38). In LD, prediction errors trigger updating of predictions that, at some point, reach a metacognitive self-model explaining the ongoing mental representations as dream images (37). We assume that in LD, prediction errors are streamed forward through the processing hierarchy, reaching supramodal, prefrontal, and parietal cortical regions [in contrast to non-LD when activity is largely attenuated in these areas (16,(31)(32)(33)]. ...
... Other cognitive techniques instruct participants to recall their intention of becoming lucid and to rehearse this objective, especially before falling asleep (99). These practices use prospective memory to increase the probability of the prediction (prior) of being in a dream against the alternative prediction of being awake (37). ...
Full-text available
Lucid dreaming (LD) is a mental state in which we realize not being awake but are dreaming while asleep. It often involves vivid, perceptually intense dream images as well as peculiar kinesthetic sensations, such as flying, levitating, or out-of-body experiences. LD is in the cross-spotlight of cognitive neuroscience and sleep research as a particular case to study consciousness, cognition, and the neural background of dream experiences. Here, we present a multicomponent framework for the study and understanding of neurocognitive mechanisms and phenomenological aspects of LD. We propose that LD is associated with prediction error signals arising during sleep and occurring at higher or lower levels of the processing hierarchy. Prediction errors are resolved by generating a superordinate self-model able to integrate ambiguous stimuli arriving from sensory periphery and higher-order cortical regions. While multisensory integration enables lucidity maintenance and contributes to peculiar kinesthetic experiences, attentional control facilitates multisensory integration by dynamically regulating the balance between the influence of top-down mental models and the precision weighting of bottom-up sensory inputs. Our novel framework aims to link neural correlates of LD with current concepts of sleep and arousal regulation and provide testable predictions on interindividual differences in LD as well as neurocognitive mechanisms inducing lucid dreams.
Full-text available
Approximately half the population have experienced a lucid dream. Yet, it is not well understood how an individual realises they are dreaming (i.e. lucid insight). A few studies suggest it can be triggered by a nightmare, or by the identification of inconsistencies/dream-like qualities/peculiarities. The present study aimed to produce a detailed typology of lucidity triggers to inform consideration of the nature of associated thought processes. 91 lucid dreamers were identified in an undergraduate sample of 148. Lucid dreamers were asked to describe what it was, if anything, that had made them realise they were dreaming. Thematic analysis of responses extracted evidence of four overarching triggers consistent with previous research: identification of an abnormality, identification of a dream-like quality, an emotionally-arousing dream event, and miscellaneous. It was uniquely identified that 'abnormalities' within the dream were those inconsistent with waking knowledge rather than with the accompanying dream content. Novel triggers were identified that were classifiable as subthemes under the previously-identified triggers, and triggers co-occurred in a third of cases. Novel triggers included 'déjà rêvé', the feeling that one has dreamt the experience before, as well as 'self-comfort/denial' involving an emotionally-driven denial of the reality of a distressing dream. Nightmare-induced lucidity appeared to arise via qualitatively diverse paths: via self-comfort/denial and/or via the identification of abnormalities/dream-like qualities. The nature of the triggers indicates that higher-order cognition can precede, and promote, lucid insight. This sheds light on a key theoretical issue of whether lucid insight is a prerequisite for the re-emergence of higher-order cognition.
Full-text available
Lucid dreaming refers to the phenomenon of becoming aware of the fact that one is dreaming during ongoing sleep. Despite having been physiologically validated for decades, the neurobiology of lucid dreaming is still incompletely characterized. Here we review the neuroscientific literature on lucid dreaming, including electroencephalographic, neuroimaging, brain lesion, pharmacological and brain stimulation studies. Electroencephalographic studies of lucid dreaming are mostly underpowered and show mixed results. Neuroimaging data is scant but preliminary results suggest that prefrontal and parietal regions are involved in lucid dreaming. A focus of research is also to develop methods to induce lucid dreams. Combining training in mental set with cholinergic stimulation has shown promising results, while it remains unclear whether electrical brain stimulation could be used to induce lucid dreams. Finally, we discuss strategies to measure lucid dreaming, including best-practice procedures for the sleep laboratory. Lucid dreaming has clinical and scientific applications, and shows emerging potential as a methodology in the cognitive neuroscience of consciousness. Further research with larger sample sizes and refined methodology is needed.
Full-text available
Thoughts occur during wake as well as during dreaming sleep. Using experience sampling combined with high-density EEG, we investigated the phenomenal qualities and neural correlates of spontaneously occurring thoughts across wakefulness, non-rapid eye movement (NREM) sleep, and REM sleep. Across all states, thoughts were associated with activation of a region of the midcingulate cortex. Thoughts during wakefulness additionally involved a medial prefrontal region, which was associated with metacognitive thoughts during wake. Phenomenologically, waking thoughts had more metacognitive content than thoughts during both NREM and REM sleep, whereas thoughts during REM sleep had a more social content. Together, these results point to a core neural substrate for thoughts, regardless of behavioral state, within the midcingulate cortex, and suggest that medial prefrontal regions may contribute to metacognitive content in waking thoughts.
Full-text available
Lucid dreaming is a state of awareness that one is dreaming, without leaving the sleep state. Dream reports show that self-reflection and volitional control are more pronounced in lucid compared with nonlucid dreams. Mostly on these grounds, lucid dreaming has been associated with metacognition. However, the link to lucid dreaming at the neural level has not yet been explored. We sought for relationships between the neural correlates of lucid dreaming and thought monitoring. Human participants completed a questionnaire assessing lucid dreaming ability, and underwent structural and functional MRI. We split participants based on their reported dream lucidity. Participants in the high-lucidity group showed greater gray matter volume in the frontopolar cortex (BA9/10) compared with those in the low-lucidity group. Further, differences in brain structure were mirrored by differences in brain function. The BA9/10 regions identified through structural analyses showed increases in blood oxygen level-dependent signal during thought monitoring in both groups, and more strongly in the high-lucidity group. Our results reveal shared neural systems between lucid dreaming and metacognitive function, in particular in the domain of thought monitoring. This finding contributes to our understanding of the mechanisms enabling higher-order consciousness in dreams. Copyright © 2015 the authors 0270-6474/15/351082-07$15.00/0.
Agency detection is a central concept in the cognitive science of religion (CSR). Experimental studies, however, have so far failed to lend support to some of the most common predictions that follow from current theories on agency detection. In this article, I argue that predictive coding, a highly promising new framework for understanding perception and action, may solve pending theoretical inconsistencies in agency detection research, account for the puzzling experimental findings mentioned above, and provide hypotheses for future experimental testing. Predictive coding explains how the brain, unbeknownst to consciousness, engages in sophisticated Bayesian statistics in an effort to constantly predict the hidden causes of sensory input. My fundamental argument is that most false positives in agency detection can be seen as the result of top-down interference in a Bayesian system generating high prior probabilities in the face of unreliable stimuli, and that such a system can better account for the experimental evidence than previous accounts of a dedicated agency detection system. Finally, I argue that adopting predictive coding as a theoretical framework has radical implications for the effects of culture on the detection of supernatural agency and a range of other religious and spiritual perceptual phenomena.
Over a century ago, William James wrote: “The universal conscious fact is not ‘feelings exist’ and ‘thoughts exist’ but ‘I think’ and ‘I feel’” (James 1890: 226). In other words, central to consciousness is the self-as-observer, what James called ‘the knower.’ As Baars explains, “the self is that which has access to consciousness” (1997: 153). (Also see Dennett, 1978.) The notion of self-as-observer implies dual levels of awareness: The contents of experience (participant perspective) are the object of awareness (observer perspective).
We report a quality effects meta-analysis on studies from the period 1966–2016 measuring either (a) lucid dreaming prevalence (one or more lucid dreams in a lifetime); (b) frequent lucid dreaming (one or more lucid dreams in a month) or both. A quality effects meta-analysis allows for the minimisation of the influence of study methodological quality on overall model estimates. Following sensitivity analysis, a heterogeneous lucid dreaming prevalence data set of 34 studies yielded a mean estimate of 55%, 95% C. I. [49%, 62%] for which moderator analysis showed no systematic bias for suspected sources of variability. A heterogeneous lucid dreaming frequency data set of 25 studies yielded a mean estimate of 23%, 95% C. I. [20%, 25%], moderator analysis revealed no suspected sources of variability. These findings are consistent with earlier estimates of lucid dreaming prevalence and frequent lucid dreaming in the population but are based on more robust evidence.
Lucid dreaming has been said to be within the capability of all individuals (LaBerge, 1985). Based on analyses of the incidence of this dream experience among university students and among persons with an expressed interest in dreaming, a majority have reported experiencing at least one lucid dream during their lifetime, and about 20% have reported experiencing lucid dreams with relative frequency. Our goal in this chapter is to describe and to integrate what has been learned through research about individuals who experience lucid dreams. To this end we will present data derived from the study of four separable but not unrelated functional domains for which subject differences associated with lucid dreaming, or lucidity, have been found. These functional domains are (1) oculomotor/equilibratory; (2) visual/imaginal; (3) intellectual/creative, and (4) personal/interpersonal. The extent of individual differences in lucid dreaming and the methods by which these differences have been investigated will also be discussed. Because methodology is an integral part of research into individual differences, methodological considerations will first be presented.