When I graduated from college twelve years ago, I was well aware of the Silicon Valley hype machine. Still, I thought that private tech company salesmanship was a world removed from the scientific facts of human biology that I had learned in neuroscience courses. I remember seeing neuroscientist Henry Markram announce in a TED talk that he had worked out how to replicate an entire human brain on supercomputers within ten years. This computer-simulated organ will enable us to test potential therapies for illnesses and diseases in real-time and noninvasively, allowing us to move away from animal experiments and delicate interventions on living humans and toward an “in silico” approach to neuroscience.
My 22-year-old brain didn’t recognize this as an overhyped bid. Instead, it felt thrilling and daring, the sort of moment that turns a distant science pipe dream into an instantly attainable target and inspires funders and colleagues to dream bigger. As a result, I embarked on a 10-year documentary project covering Markram and his Blue Brain Project, with the film’s premiere coinciding with the dawn of a new age of major neuroscience, in which Silicon Valley’s buzzing black boxes become the great new hope for making sense of the black boxes behind our ears.
My ten-year experience recording Markram’s vision has yielded no straightforward answers, except for one: that dazzling presentations and pure optimism are weak markers of progress in studying the intricate biochemical functions of the brain. Today, when we see a monkey mind-controlling a game of Pong as part of Elon Musk’s start-up Neuralink’s usually bombastic presentation, there is a greater need than ever to unwind the hysteria loops to grapple about what the future of brain technology and neuroscience has in store for humanity.
This hype often relies on selective amnesia for unfulfilled commitments in the past to rekindle hope for science and technological advancement. My initial enthusiasm for the Blue Brain Project’s promises remains, but it has been folded into a messier, more complex experience. Over the decade after Markram’s TED lecture, the critiques and technical shortcomings I’ve faced have turned each new neuro hype loop into a potent reminder of those early enthusiasms and the dangers of mishandling them.
About three years after my tenure recording the Blue Brain Project, I had my first epiphany. Things weren’t going as planned: in a visitor’s viewing room, there were stunning fly-through visualizations of the first square millimeter of artificial rat brain set to The Blue Danube, but there was a distinct lack of development along with the road map toward a human brain. Soon, the discussion turned to a more significant, necessary project known as the Human Brain Project, which would cost more money but have the tools needed to accomplish the objective. Proposals were sent, and the project was awarded a billion euros from the European Union, only to be dogged by controversy a year later after an open letter signed by over 800 neuroscientists disagreed with the project’s central concept on how to model a human brain and objected to its founder and director, Markram’s leadership style.
As the theoretical debate and interpersonal fallout steadily revealed that the 10-year attempt to replicate a human brain on a machine had been a pipe dream all along, I began to interview more skeptics of the project and delve further into what it meant to suggest you wanted to do anything like thisDuring our interview, Princeton neuroscientist Sebastian Seung asked me a question that had vexed me for some time, referring to science, legal, and moral traps that this research was heading for: “I’d like to ask you this,” Seung began, shifting the focus from the Blue Brain Project to my own time there and the essence of the artificial mouse brain that the project’s researchers had depicted in stunning visuals “Inside this, they gave you a simulation of any neuronal function.
What if it looked different? How would you know if it was correct or incorrect?” “Well, I wouldn’t know,” I answered from behind the camera. “Right, how does anyone know what was a wrong activity pattern or a right activity pattern?” Seung repeated.
The difficulties arise when one considers what “right” will mean in this case, since recreating a highly noisy biological mechanism within the circuits of a perfectly programmed computer seems to lead to a fundamental platform problem inevitably. A series of unpredictably “mistakes propel biology.”—also known as mutations—that cause variation in individuals within a species and interact with our surroundings to drive evolutionary change by natural selection. Neurons are also considered to be noisy components, producing action potentials that aren’t always predictable. Structure errors, known as “bugs,” in computers, on the other hand, are easily patched to make room for the ideal code for the job at hand.
Many human brain features will undoubtedly be modeled, probed, and their generalities removed, much as we have done with the human heart to develop a mechanism that could keep me or you alive. But how could a predetermined system of software operating on computers ever catch the utterly random errors seen at any stage of biological life, from mutations in our DNA to synapse function, as Markram had told me in our first interview?
Though the scientists I consulted over the years had a variety of viewpoints on the thorny problems of noise and disorder in computer simulations of biology, it wasn’t until I interviewed with a more junior neuroscientist at the Blue Brain Project that I heard a response that cuts through the positivist public-relations gloss. “That’s an interesting point because the best kind, we will never tell what’s the right kind of variability,” she said when asked if one can ever identify the “right” kind of variability in a simulation of a biological organism.
Suppose we’ll never know the right kind of variability. In that case, it seems that what we’re talking about while trying to replicate biological systems on machines is a computational device that does exactly what its designers want. Taking cues from AI, computational neuroscience is increasingly abandoning biological brains to pursue ideal algorithms, which, like its deep learning counterparts, can eventually result in more black boxes that perform tasks but are incomprehensible from the within.
The ethical responsibility for the simulation’s personality becomes a core concern in the halls of server farms and neuroscience laboratories bent on reproducing biological activity on artificial devices. Simulations of neural behaviour will eventually hold a mirror up to their authors’ prejudices, far from achieving an empirical recreation of “the human brain.” What would it mean when the device in question is the human brain, and tech firms conclude that knowing a system isn’t necessary for exploiting or reproducing a variant of it for profit?
The helplessness he felt after his son was diagnosed with autism was one of Markram’s reasons for trying to accelerate neuroscience toward a complete emulation of a human brain. Indeed, for many researchers in the area, the motivations behind their study can be intensely personal and, at least in principle, broadly beneficial—which is why I continue to be fascinated by the potential for improving the human condition that discovery and scientific ambition have. The distinction between fantasy and fact will begin to blur as technology advances, and a particular strain of technocratic salesmanship begins to command the collective human ear, resulting in cycles of speculation and disappointment that endanger long-term public trust in science.