Can computers read our dreams?

Scientists have developed the first system which is able to predict what objects are appearing in peoples' dreams.

The team, led by Tomoyasu Horikawa of the ATR Computational Neuroscience Laboratory in Kyoto, Japan, worked with three volunteers, who were invited to try to sleep while lying in a Functional MRI scanner.

Such scanners are able to map out which areas of a subject's brain are active, by monitoring the flow of oxygenated blood around the brain.

The research team fed the brain activity patterns into a computerised pattern-recognition system which searched for correlations between activity in particular brain areas, and reports that the subjects were dreaming about particular subjects.

A stumbling block for similar studies in the past has been that dreams are, by definition, internal to the subject's mind. Naturally, it is impossible to hold a conversation with the subjects while they are asleep. This makes it difficult to know when they might be experiencing dreams, or what those dreams might be about.

The volunteers who took part in the study were woken from sleep more than 200 times over a ten day period and asked to recount what they were dreaming about. As well as being subjected to these repeatedly reawakenings -- roughly every six minutes -- they were also being asked to try to sleep whilst surrounded by noisy MRI scanning equipment, though headphones were apparently provided to reduce the noise that they were subjected to as much as possible.

The research team were particularly interested to see how comparable the patterns of brain activity were when their subjects saw particular visual images during wakefulness, versus when similar objects were encountered in dreams, with the aim of identifying where in the brain dreams originate. To make this comparison, the team used internet search engines to find images of the kinds of objects that their subjects often dreamed about, and mapped out the brain activity of the subjects when they were shown these images while awake.

The team found striking similarities between the brain activity recorded in the visual cortices of their subjects between visual experiences had during wakefulness and dreaming, suggesting that the brain's visual processing circuitry is active during dreaming. Of perhaps even greater interest, they also found that by the end of their study, they could predict what objects that were appearing in the dreams of the subjects in around a third of cases.

From this, they conclude that they are still a little way away from having a reliable way to evesdrop on people's dreams -- though one reason why their predictions may not have been more accurate is the verbal descriptions given by the subjects may not have reported all of the objects appearing in their dreams: cars are usually associated with street scenes, for example, but the street may have seemed too obvious to mention.

However, their work brings together two poorly understood areas of neuroscience: how the brain processes visual information, and what is happening within the brain during dreaming.

As lead author Tomoyasu Horikawa points out, the most interesting aspect of this is that the majority of dreams are not remembered; in the future, techniques like this may reveal what people have dreamed about through the course of the night, and how the brain is functioning during those dreams.