Tom Thomas is a writer with a career spanning forty years in publishing, technical writing, public relations, and popular fiction writing.
“My business now is to weave circumstance, happenstance, intention, and mischance into stories.”

The Machine Condition:

The Dreaming Brain – October 25, 2026

Recently I wrote about consciousness and how it might differ from the function and outputs of the large language models (LLMs). It now occurs to me that we have a model of the LLM processes in our own heads: the dreaming brain.

When we dream, the brain produces memory fragments, usually accompanied by visual images and auditory sequences, that are assembled in a story-like fashion, from one action to the next. But often there are dislocations, sudden changes in place or form. People, objects, and settings are stripped of character or assembled from disparate memory fragments. There is sequence but no coherent order or long-term continuity. It is almost as if you are awake, but something is missing.

Dreams often have an internal logic. The dreaming brain asks a question, and usually there is an answer, but not always the logical one. Another person in the dream makes a statement or takes an action, and usually the dreamer reacts, but not always in a consistent manner. And then the sequence quickly dissolves: the dreamer or the other people disappear, the scene and its implied concerns change, the dream itself goes on to something else.

That missing element may well be involvement of the prefrontal cortex during REM (rapid eye movement) sleep, which is associated with dreaming. This is the front part of the brain that makes decisions and directs our attention, our will, and our focus.

In the prefrontal cortex our brains have an internal mechanism that keeps us on task, completing one step before going on to something else, and planning in advance all the steps needed to complete the project. It directs and focuses our awareness by choice or necessity. The prefrontal cortex does let us notice distractions, evaluate them quickly for danger or opportunity, and then returns us to the work at hand.

This executive function is operative while we are awake. In fact, upon waking up, it sets about “clearing the cobwebs” of our nighttime imaginings and begins planning for the day ahead—what day of the week it is, what tasks or encounters we need to prepare for, what changes might be needed to the daily routine. All of this takes place in the first few minutes of breaking the sleep cycle.

During the 1940s and ’50s, hospitals used to perform lobotomies on patients with certain mental illnesses, like schizophrenia, psychosis, and manic depression. The doctors surgically severed the frontal lobe—the part I’m talking about—from the rest of the brain. The patients immediately became less agitated and disturbed. They also became emotionally blunted, less responsive, less able to function as decisive and creative human beings. We don’t do this anymore, because we now have medications that will alter the brain’s chemistry to quiet the psychotic urges and stimulate the depressed states.

My contention is that the large language models are like the dreaming brain because they lack this executive function, which is the driver of our conscious mind. They have no internal checks on how they go about a task, foreseeing the outcome and planning for it. They have no internal course correction.

This may be because, as I understand it, processing in an LLM is “once through.” It receives a task or prompt, begins linking tokens—that is, words and word fragments in the case of verbal output, or images and image fragments in the case of visual renderings—that seem to be related to the subject matter and to the next probable step in the formulation. The process only stops when the prompt seems to be satisfied. And if the human user is dissatisfied with the outcome and changes the prompt, then the process starts again at the beginning with the changed input parameters.

But along the way, other than by reference to the prompt, the LLM has no executive function to ask if the developing output makes logical sense, satisfies its understanding from comparative experience (other than the probabilistic linkage of one token with the next), or serves some purpose other than answering the prompt. The LLM has no such governing awareness. And the model may be biased toward completing the prompt rather than including only known or reliably proven facts and resources. The LLM can’t check those facts while it is proceeding, token by token, toward a complete response to the prompt.¹

Maybe some programmer or technician will find a way to make a computer follow two or more tracks at once: decide which token, thought, or image comes next in the stream; and determine whether or not the developing response is logical or reasonable based on the sum of the system’s collective experience. The human brain does this all the time because we don’t have just one stream of thought proceeding or neural circuit firing at any one time. While we are making breakfast, we are also listening to the news, deciding what to wear that day, and thinking about what we’ll say at the big meeting this morning. And other parts of the brain are simultaneously guiding our hands through familiar motions, monitoring the surrounding air temperature, governing our balance, directing our breathing and heart rate, and doing all the hundred and one tasks needed to keep us alive and moving forward.

Maybe computer chips with this kind of vast interconnection, running multiple functions while also checking on them, are possible. Maybe such things are already in development—I don’t know. But until they are, this thing we call “artificial intelligence” will remain as wild and woolly as our dreaming brains, disconnected from any kind of reality, and not “conscious” at all.

1. In other words, garbage in, garbage swirl around, garbage out. If you feed a database on a potpourri of true facts, doubtful assumptions, and artful lies—and most models are trained variously on fiction and nonfiction books, published newspaper and magazine articles, legal decisions, textbooks, and random sampling from social media—with no way to distinguish one category from the other, then you can’t trust the outcome.

Comment on this post at Blogspot.com.

Subscribe to my Facebook author's page by clicking “Like” at Thomas T. Thomas.

My Favorite Novels About Time Travel

Contact me.

This is the official home page of Thomas T. Thomas, the fiction writer. Member of The Authors Guild and Science Fiction and Fantasy Writers of America. As there are a number of other Thomas T. Thomases alive and active in the world, please see the Biography sidebar “What’s the Middle ‘T’ Stand For?” to make sure you have found the right one.