The way it worked was approximately this: Each subject was shown a batch of movie clips. Their brain activation patterns were recorded each time. Then, when the person watched a new, previously unseen clip, activation patterns were once again recorded. Then the original clips were mixed into a new clip proportionally, according to how similar the activation pattern for the new clip was to each original clip. With enough previously seen clips mixed together, a fuzzy new clip emerges that does look like what the subject is watching.
