The Tantalizing Possibility Of Locating Consciousness In The Brain

Daniel Freeman, a scientist at MIT Lincoln Laboratory, is developing transcranial ultrasound and AI methods to identify the neural basis of conscious experience, and he suggests that intelligence may be portable across physical substrates, as evidenced by AI chatbots exhibiting human-like cognition without biological neurons. Freeman highlights the work of neurologist Antonio Damasio, whose 1994 book 'Descartes' Error' argued that emotions are integral to rational thought, based on patients with prefrontal cortex damage who lost decision-making ability despite intact intellect.

Daniel Freeman is a scientist at MIT Lincoln Laboratory developing transcranial ultrasound and AI methods to identify the neural basis of conscious experience. He is also helping to build an interdisciplinary consciousness research initiative at MIT. It is past midnight in a fluorescent-lit office. A neuroscientist opens ChatGPT and uploads the seventh version of a paper, which is littered with critical notes from a peer reviewer. “Why is Reviewer 2 such an idiot?” the neuroscientist asks the chatbot. The reply comes instantly, witty and sympathetic, reframing the criticism in a way that puts everything into perspective. The neuroscientist laughs, then pauses. How can this computer seem so very human? It demonstrates abilities we usually associate with intelligent beings: language, abstraction, social inference, even a sense of humor. The design of modern chatbots was inspired by the neocortex, the pinnacle of mammalian evolution, so their human-like behavior should not have come as a complete surprise. But what few anticipated was that such complex behavior could emerge in the complete absence of biological neurons. For decades, cognitive neuroscience was organized around a grand challenge: explaining how electrical activity flowing through networks of brain cells could produce the remarkable behaviors associated with our intelligence. Planning, reasoning, problem-solving, creativity: These were treated as the crown jewels of the mind. Yet in that office that night, like on screens around the world, a machine exhibited many of those same abilities using nothing more than electrons bouncing around in a little block of silicon. A strange possibility has emerged. Maybe we have already solved a large part of what once seemed mysterious about cognition. Perhaps the ability to generate intelligent behavior never required anything beyond this clever trick of signal processing that we call “neural networks.” If so, intelligence may be more portable than we once imagined. The same computational architecture that operates in biological neurons can be implemented in silicon microprocessors, or even in hypothetical mechanical systems built from springs, gears and levers. The physical substrate may vary, but the computations, which produce human-like behavior, do not. The mainstream position in neuroscience is that cognition is a necessary ingredient of consciousness — that somehow, the capacity for complex thought goes hand in hand with this mysterious phenomenon of subjective experience. Yet now we have computers that can achieve such cognitive feats, and still, we seem no closer to understanding how the brain produces our conscious awareness. Most discussions of AI and consciousness ask whether machines can experience anything. But maybe the real mystery is: If consciousness is not fundamentally related to higher cognition, then what kind of phenomenon is it? The Neural Machinery Of Feeling Over the course of his career, neurologist Antonio Damasio became fascinated by patients whose brain injuries had left their intellect largely untouched, but had disrupted their ability to navigate the world. They could reason, hold conversations and solve complex problems, but their decision-making was impaired. One of Damasio’s most cited patients had damage to a region of the prefrontal cortex known to be critical for emotional valuation. When presented with two possible appointment times, the patient was paralyzed with indecision, spending hours considering the question. The problem was not a lack of intelligence. Rather, the emotional signals that normally assign value to different options had been disrupted. The patient lost the gut feeling that normally guides decision-making. This observation became the foundation of Damasio’s influential 1994 book, “Descartes’ Error.” The central idea was simple but provocative: Feelings aren’t peripheral to rational thought. They are part of the machinery that makes rational thought possible. Of course, human experience does not seem limited to feelings. We ruminate. We deliberate. We weigh evidence and reflect upon our own beliefs. Don’t these complex mental states involve something more than just a collection of primitive feelings? Maybe not. Inspired by Damasio’s insights, more and more researchers have been exploring https://www.scientificamerican.com/article/are-the-roots-of-consciousness-in-the-ancient-deep-brain/ the possibility that human thought may not constitute a separate category of experience from feeling. Instead, our stream of thought may emerge from combinations of a finite number of primitive feeling states. This proposal may seem improbable. But perhaps we have become too accustomed to thinking of consciousness as something inseparable from cognition. It may be that the computations occurring throughout the brain remain largely hidden from awareness, and what we actually experience consists of a highly processed, felt summary of their results. Emerging AI models offer an intriguing analogy: We type a question into ChatGPT and receive some text back, but we are not shown the trillions of calculations occurring beneath the surface. What reaches us is a compressed summary of an enormously complex process. Something similar may happen when the brain produces subjective experience. Beneath every conscious thought lies a vast ocean of subconscious computation. The cortex performs these calculations to support reasoning, planning, memory and decision-making, but what enters our conscious awareness may be an experiential readout of those processes. Subjective experience may be like the instrument panel of an aircraft, summarizing critical information from systems too complex to inspect directly. An Engineering Perspective Engineers attempting to understand a complex machine generally don’t begin with the most complicated component. Instead, they identify simpler building blocks and ask how those elements combine to generate more sophisticated behavior. Consciousness research may require a similar strategy. Let’s begin with the simplest case. It is not difficult to imagine that a particular pattern of neural activity could give rise to a simple subjective feeling. Hunger feels one way. Pain feels another. Unease, familiarity, anticipation and confidence all possess their own distinctive phenomenology. Each feels like something. Now suppose that complex thoughts are assembled from these primitive felt states like recipes in a cookbook. Let’s say that two parts familiarity and one part coherence produce the thing we call “understanding.” Three parts anticipation and one part motivation produce a thing we call “hope.” The individual ingredients remain simple, yet together they generate experiences that feel qualitatively different from any one ingredient alone. But what about the experience of thinking about something specific, like a bicycle? A bicycle need not correspond to a single primitive feeling. Instead, it may occupy its own location within a multidimensional landscape, defined by a distinctive combination of familiarity, anticipation, confidence, coherence, motivational pull and countless other experiential dimensions. A castle, a childhood friend and a mathematical theorem would each occupy different regions of that same landscape. A small number of ingredients can generate a very large number of combinations. Imagine that conscious experience is built from 30 independent experiential dimensions and that each can assume eight distinguishable levels. The resulting space would contain roughly 1 octillion 1027 possible configurations, an astronomical number. By comparison, if every human who has ever lived experienced a new conscious state three times every second throughout their entire lifetime, humanity would collectively sample only about 1 sextillion 1021 experiences. Even across the entire history of our species, we would have explored only one-millionth of the available experiential landscape. From this perspective, the richness of human thought no longer requires an endless catalogue of irreducible mental states. It may arise instead from an enormous combinatorial space built from a small set of primitive feelings. Can Consciousness Be Localized? Where exactly in the brain does the magic happen? If you ask neuroscientists to identify the brain structures responsible for subjective experience, they will likely point to the frontal areas of the cerebral cortex. This is the region responsible for many of our sophisticated cognitive abilities. Of course, the idea that consciousness requires intelligence might be pre-AI thinking. As that assumption has begun to weaken, researchers have become increasingly willing https://www.scientificamerican.com/article/are-the-roots-of-consciousness-in-the-ancient-deep-brain/ to consider the possibility that subjective experience arises from evolutionarily older structures deep within the brain. The question of whether a cortex is necessary to experience something is one of the most hotly debated https://www.radiologytoday.net/ultrasound-news-innocence-and-experience/ topics in the field. Yet we’ve known for a century that the brain can function without a cortex. Beginning in the 1920s, physiologists surgically removed https://psycnet.apa.org/record/1934-04911-001 the cerebral cortices from dozens of cats. Those that survived the complex procedure were surprisingly capable. Similar results have since been achieved with genetically modified mice born without cortices. They could eat, reproduce, swim, groom themselves and even learn simple mazes https://www.youtube.com/watch?v=hBuVPZND4-0 . Their capabilities were extremely impaired, but it is difficult to imagine that these animals experienced nothing at all. All of this raises a natural question: Could the neural circuitry that mediates subjective experience be located deep within the brain, rather than in the cortex? This possibility is consistent with the finding that deep-brain stimulation in humans can evoke intense subjective experiences https://www.nature.com/articles/s41583-022-00583-8 , whereas stimulation of large regions of the prefrontal cortex often produces little or no conscious effect https://www.jneurosci.org/content/41/10/2076 . It also fits with evolutionary history: Deep-brain circuits emerged long before https://academic.oup.com/book/37442 the cortex ever evolved, leaving a shared subcortical blueprint across all vertebrates. If consciousness arose early in that evolutionary story, it likely continues to depend on these ancient foundational structures today, from fish to humans. Here is where it gets especially interesting: Structures deep within the brain are more anatomically discrete than the expansive networks of the cerebral cortex. If subjective experience originates within these structures, then the search for consciousness changes from the sprawling landscape of the cortex to a relatively small piece of neural anatomy. Few discoveries would be more consequential than identifying the specific neural structures responsible for subjective experience. Why Feelings Exist Taken literally, the idea that conscious experience is like the instrument panel of an airplane carries an uncomfortable implication. Instrument panels do not fly airplanes; they simply report what the underlying machinery is doing. If consciousness merely summarizes neural computation, then it risks becoming little more than an epiphenomenon — a passive observer with no influence over behavior. That conclusion would be difficult to reconcile with evolution, which is remarkably efficient at preserving biological functions that improve survival outcomes while eliminating those that do not. If subjective experience has accompanied animal life for so long, then it is reasonable to ask what function it serves. Why should evolution preserve the feeling of hunger rather than simply generate feeding behavior? Why should fear be experienced rather than simply triggering actions to escape? The apparent contradiction disappears if we stop thinking of feelings as passive experiences and instead view them as the brain’s way of evaluating competing biological priorities. Long before animals evolved to demonstrate language, reasoning or abstract thought, they faced a much simpler problem: choosing among competing biological needs. Eat or flee. Drink or reproduce. Rest or explore. These drives can’t all be satisfied simultaneously. Somewhere in the brain, they must be compared, prioritized and translated into action. The ancient structures of the upper brainstem and hypothalamus are ideally positioned to perform this role. Rather than comparing countless physiological variables directly, evolution may have discovered a simpler solution: converting competing biological needs into a common language of subjective feeling. Feelings are not merely reports about the state of the organism; they are the mechanism through which motivation arbitration takes place. If this picture is correct, whatever physical mechanisms give rise to feeling must also influence the neural circuits responsible for selecting actions. Exactly how this interaction occurs remains one of neuroscience’s deepest puzzles. After all, it is hard to decipher how feeling drives behavior when we still don’t know how brain activity creates feeling in the first place. From this perspective, subjective experience is neither the computation itself nor a passive byproduct of computation. Instead, it may constitute the currency through which ancient brain circuits compare competing priorities and select actions. This is the kind of decision support that even an early vertebrate would have required hundreds of millions of years before the evolution of the cortex. When the cerebral cortex evolved, it did not replace the ancient deep-brain architecture, but it did start to supply it with much richer information through descending projections. Rather than generating subjective experience directly, the cortex became an extraordinarily powerful computational system, analyzing the external world, constructing internal models and predicting future outcomes. The results of those computations could then be fed back into the older systems responsible for motivational arbitration and translated into feelings of uncertainty, familiarity, confidence and countless other experiential states. To paraphrase neuroscientist Bjorn Merker https://pubmed.ncbi.nlm.nih.gov/17475053/ , the cortex may generate the contents of consciousness, while deep-brain structures generate conscious experience itself. Bringing Consciousness Research Into Mainstream Neuroscience Suppose someone finally identifies the neural basis of consciousness. Convincing the rest of the field may prove almost as difficult as the discovery itself because consciousness research has not yet converged on a common framework for evaluating competing ideas. Researchers disagree not only about the answer, but about the question itself. A walk across MIT’s campus illustrates the point. Even among researchers who study consciousness explicitly, there is remarkable diversity in both methods and assumptions. In one laboratory, anesthesiologist Emery Brown studies how consciousness disappears and re-emerges during anesthesia. Nearby, electrophysiologist Earl Miller investigates the cortical dynamics associated with cognition. Across the street, philosopher Matthias Michel examines the conceptual foundations of consciousness itself, asking how subjective experience should be defined and measured. Each approaches the mystery from a different direction, bringing a different set of questions, methods and assumptions about what consciousness is in the first place. Some of the researchers whose work may prove most relevant to consciousness do not study consciousness directly. Among scientists here at MIT, Ed Boyden develops technologies for recording, manipulating and mapping neural circuits. Fan Wang investigates pain and the neural circuits that generate aversive behaviors. Josh Tenenbaum studies the computational principles underlying cognition and intelligence. Satrajit Ghosh develops AI tools capable of extracting patterns from enormous neurophysiological datasets. These are the methods that drive modern neuroscience forward: identifying mechanisms, building models and generating predictions that can be experimentally tested. Yet few of these researchers would describe themselves as consciousness scientists. This creates an unusual situation. Neuroscientists can spend their entire careers uncovering the neural circuits responsible for pain, fear, memory or decision-making. But the moment they ask how those circuits give rise to subjective experience, the scientific ground becomes shaky. They find themselves steeped in conjecture. This situation may be changing. As theories become more firmly grounded in testable neural mechanisms, such as with non-invasive deep-brain stimulation https://news.mit.edu/2026/new-tool-could-tell-us-how-consciousness-works-0112 , the boundary between consciousness research and mainstream neuroscience may gradually dissolve. Rather than standing apart as a discipline defined by philosophical debate, consciousness research may become increasingly integrated with the broader effort to understand how the brain works. The Search Continues By demonstrating that many hallmarks of intelligence can emerge from artificial neural networks, AI has weakened one of the central assumptions that guided consciousness research for decades. Rather than searching for experience within the distributed computations that support language, planning and reasoning, researchers are asking a different set of questions. Which neural circuits generate primitive feeling states? How do those feelings combine to produce the rich phenomenology of thought? And how do ancient motivational systems interact with the extraordinary computational machinery of the cerebral cortex? The big questions surrounding consciousness are moving from philosophy to the laboratory bench. New tools are allowing researchers to test ideas about consciousness with the same rigor expected of every other branch of neuroscience. Artificial neural networks, originally inspired by the cerebral cortex, have become indispensable partners in that effort, helping investigators extract patterns from neural data that would have been difficult to recognize only a decade ago. The work ahead will require more than good ideas. It will require the time, talent and resources needed to transform decades of descriptive observations into mechanistic explanations. It will require building a scientific community capable of treating consciousness not only as a philosophical problem, but as a central question in neuroscience itself.