Reader Ella asks: “I read a theory that while dreaming, the brain cannot invent new people out of nowhere. Instead, the brain shows people we've seen while awake, or combines a mix of previously-seen physical features to create a "new" person. How would you prove/disprove this theory? Why does the brain do this?”
This is a really interesting question, but unfortunately, it’s an impossible question to answer experimentally. To test this theory, we would need an accurate image of the unknown dream person and a reliable and accurate way to know if the dreamer had ever seen the person in their waking life. This is difficult for a number of reasons. First, our dreams are typically not vivid enough to distinguish the individual facial or body features that would be required to get a precise image of a dream person. Think about your own dreams -- when you dream about someone you know, do you recognize that person because you can see their face clearly or do you just know that it’s them?
Second, our memories about our dreams are extremely fleeting. We start to forget our dreams as soon as we wake up, making the precise recollection of a person or face extremely difficult. To make the problem worse, when we can’t remember something clearly, our brains have the habit of filling in the gaps with people/places/things from previous experiences or similar events, so even if we did dream of a unique person, if we weren’t able to recall that person while awake, our brains would likely fill in the details with people that we have seen before.
Third, an individual person could encounter dozens or even hundreds of human faces on a daily basis from commuting to and from school or work, or from seeing people on the news or in television and movies. Most of these people will remain strangers to our conscious selves, but their faces and figures will still be perceived and processed by our brains. Because of this, it would be impossible to say for certain that you had never seen a person or face outside of your dream before.
Because of these issues, among others, this theory cannot be tested in a meaningful or accurate way. However, it is an interesting question, so if we speculate based on what we do know about the nature of dreams and our brains, it is more likely that our sleeping brain recycles previously seen faces rather than creating new ones. This has to do with what dreams are made of: during wakefulness our thoughts are influenced by input from both the external environment - i.e., the people and things we see, hear, and interact with - as well as our internal environment, i.e., our memories; during sleep, however, our brains receive very little input from the external environment, which leaves our memories as the source for most, if not all, of the material that makes up our dreams.
How do we know this? Let’s first consider what function sleep and dreaming actually perform. While there have been many reported functions of sleep, one of the most important and widely studied is the consolidation of new memories. Memory function involves three different phases: acquisition, consolidation, and retrieval. During acquisition, new information is initially stored as weak memory traces in the hippocampus. During consolidation, most of which occurs during sleep, these weak memory traces are strengthened and transferred to the cortex for integration into pre-existing networks and long-term storage [1-2]. Because the same neural circuits are active during acquisition (while awake) and during consolidation (while asleep), some researchers hypothesize that dreams – particularly those dreams in which the dreamer relives events from their waking life – are part of the memory consolidation process. This is not universally accepted, however, and other researchers have argued that dreams are merely an unintended side effect of the cortical activation that occurs during consolidation and have no biological function. Either way, there is mounting evidence suggesting that dreams, whatever their function may be, are a consequence of the cortical-hippocampal activation that occurs during sleep.
As newly acquired memories are being transferred to the cortex for long-term storage, existing memory circuits are also being activated as the brain integrates this new information into existing contexts. For example, one day you and two friends attend a Flaming Lips concert and are driven to the venue in a yellow Volkswagen beetle. As this new memory is being consolidated, related memories -- such as the time you and the same two friends watched The Avengers, or the time you got carsick and threw up while riding in a different yellow beetle – may also be activated. As the new and existing memory circuits are being co-activated, the people, places, experiences, and emotions (from real life, fiction, and fantasies) stored in these circuits can be combined to create novel situations, for example, you might dream about going to a concert with Iron Man. As sleep researcher Corrado Cavallero explained, “dreaming is not “creating” but merely recombining, possibly in original ways, what has been previously stored in long-term memory.” 
This hypothesis is supported by studies dating back to the 1960s and ‘70s that have consistently found that the content of our dreams simulates everyday life to a large degree [4-5]. In one such study , published in 1971, the researchers analyzed the dreams of 16 participants. The participants rated the novelty of the physical surroundings, characters, activities, and social interactions in each dream on a scale of 1-6, ranging from an exact replication of a waking life experience, to something not previously experienced and extremely unlikely to happen in waking life. They found that more than half of all the elements in a given dream (physical surroundings, characters, and activities/interactions) were either exact replicas or slightly varied versions of something that occurred in the dreamer’s waking life. Conversely, only ~15% of physical surroundings and ~5% of characters were unknown to the dreamer and either unlikely or extremely unlikely to occur in real life (one example they give is a dream in which “a nine-foot man appeared out of nowhere”).
So the majority of dreams are about people, places, and experiences taken from the dreamer’s waking life, but what about those unfamiliar elements, especially those unfamiliar people? Where do they come from? Another finding from the study gives us a clue: the researchers also found that ~42% of physical surrounding elements and ~37% of character elements were unknown to the dreamer but could very easily occur in waking life (example: “A couple, whom I don't know, came down the street dressed in winter clothes since it was cold out”). The authors explain the importance of this last category as such: “Conceivably, the occurrence of apparently novel but otherwise unremarkable elements in dreams could represent memories of previously experienced elements which have been lost to waking recall.”
So in the above example, the dreamer may have seen the unknown couple in a different context, but because the event was unremarkable, the memory of this previous experience did not rise to the level of conscious awareness or was forgotten. The weak memory trace of this couple then combined with the memory trace of the dreamer walking down the street on a cold day, giving rise to a novel experience.
So answer your question: because the vast majority of our dreams involve mundane elements from our waking life, it stands to reason that the strangers in our dreams also come from waking life, even if we don’t recognize them in our dream. Certainly our brains are capable of inventing a unique person (although even a “unique” creation would be composed of facial and body features that we’ve seen before), and there is nothing that would necessarily prevent a sleeping brain from doing so. However, based on what dreams are and where dream content comes from, it is more likely that the strangers in our dreams are a version of someone we’ve seen in our waking lives.
 Vallant R, Chatard B, Blagrove M, and Ruby P. Characteristics of the memory sources of dreams: A new version of the content-matching paradigm to take mundane and remote memories into account (2017). PLoS ONE 12(10): e0185262.