- seCtIon FIVe: BoRDeRLAnDs
The same sensory areas are activated when something is seen or heard as when
it is imagined or remembered, and the same seems to be true of dreams. For
example, the relative increases in activity in sensory areas and decreases in
prefrontal areas are consistent with multisensory dreams lacking in executive
control of action or decision-making. The emotions in dreams are consistent
with increased activation in the amygdala, orbito-frontal cortex, and anterior
cingulate, and the involvement of memory is related to activation of the hip-
pocampus and connected areas (Maquet et al., 2005). Animal studies have
revealed more about the connections between learning, memory, and dream
content. For example, rats were trained to run on a circular track and activity in
the hippocampus was recorded during the activity and when asleep (Louie and
Wilson, 2001). Of more than forty REM episodes, about half repeated the unique
signature of brain activity that was created as the animal ran. The correlation was
so close that when the animal dreamed, researchers could reconstruct where it
would be in the maze if it were awake and whether it was dreaming of running
or standing still. More recently, studies of activity in the place cells of the hippo-
campus, which is precise enough to reconstruct a rat’s position, have suggested
that sleeping rats ‘preplay’ routes that they have seen will lead to food before
actually exploring them, forming mental maps of the projected journey to and
from the food (Ólafsdóttir et al., 2015).
Could we one day be able to deduce people’s dreams from their brain activity?
Scary as this prospect might seem, the first steps have already been taken. In the
Gallant Lab at the University of California at Berkeley, scientists recorded many
hours of fMRI data while people watched videos (Nishimoto et al., 2011) and cre-
ated a huge ‘dictionary’ to relate the shapes, edges, and movements in the videos
to activity at several thousand points in the viewer’s brain. When they then showed
a new video to the same person, they could use the dictionary to reconstruct a
recognisable, if fuzzy, version of the video being watched. A similar method has
since been applied to people sleeping inside a scanner and woken from REM sleep.
By using the recorded data and the detailed dictionary, images of what they were
dreaming about could be reconstructed (Horikawa et al., 2013). The computational
power required was vast, but the principle has been proven: it should be possible
to look at someone’s brain activity and know what they are dreaming about. This
is a huge step forward in our understanding, but perhaps only serves to make the
gulf between physiology and experience seem more obvious.
Hobson’s ‘protoconsciousness’ hypothesis about dreaming has recently been
extended using ideas from predictive processing developed by theoretical neu-
roscientist Karl Friston. The function of sleep has long been hotly disputed, with
theories ranging from maintaining neurotransmitter function to consolidating
new memories, from driving metabolite clearance to promoting neural plasticity
(Assefa et al., 2015). Hobson and Friston (2012) propose a new function. During
sleep, the brain’s ‘virtual reality generator’ (p. 85) simplifies its model of the
waking world, so improving our ability to make reliable predictions. This idea is
supported by various physiological observations. For example, pontine-genicu-
late-occipital (PGO) waves are involved in conveying eye-movement information
within the visual system and might allow the brain to carry out predictive work
during sleep. This could encompass both eye-movement command signals and
the corollary discharge that allows us to predict the visual consequences of mov-
ing our eyes.
‘REM sleep is a state of
the brain that enables
essential housekeeping
functions, upon which
waking consciousness
depend[s]’
(Hobson and Friston, 2012,
p. 87)