Singularity. A few months ago, Sam Altman, the CEO of OpenAI, published a short essay about the future of artificial intelligence. His central message was a gentle role for AI—a vision in which technology supports us quietly in the background rather than staging some dramatic takeover of human life. What caught my attention, however, was not the word “gentle” but the word “singularity.” For science fiction readers, this term carries weight. It evokes images of runaway technology, accelerated futures, and the moment when machines surpass human intelligence. Yet in the world I inhabit, working with rare diseases and clinical genetics, the reality is far more modest. AI is entering our lives in practical, incremental ways. And despite its advances, one telling detail remains: it cannot draw a synapse. This small but persistent limitation says something important about where we are—and where we are not.
Figure 1. The Ice Neurons of Delaware County. When I moved to the United States in 2014, I first encountered a phenomenon on frozen ponds that I had never seen in Europe: neuron-like dendritic patterns etched into lake ice after snowfall. These “ice neurons” arise when wet, heavy snow falls on thin ice, followed by rapid thaw and refreeze. Meltwater seeps through cracks and imperfections, radiating outward in branching channels reminiscent of axons and dendrites. In Delaware County, with its many shallow ponds and frequent freeze–thaw cycles, these conditions align perfectly—creating striking natural structures at the intersection of weather and biology.
Concepts. Let’s start with the singularity itself. In science fiction, it is the hypothetical point when artificial intelligence evolves past human control. But it is not alone in this category of grand inevitabilities. Computronium, the idea that matter itself can be transformed into optimal computing substrate, carries the same aura of certainty. So does the Dyson sphere, a vast shell encasing a star to harvest its full energy output, or the Accellerando, the accelerating rush toward technological transcendence. Each concept, when described, seems not just possible but destined. Yet when we look closer, their contradictions quickly emerge. If a Dyson sphere harnesses all energy, where is the room for human life? If computronium consumes all matter, what is left for biology? Singularity and its cousins remind us that inevitability is often an illusion, and that even the most obvious visions remain theoretical.
Synapse. After this reality check on inevitability, let’s turn to something far more tangible: the synapse. The synapse is the junction between neurons, with intricate pre- and postsynaptic structures, neurotransmitter vesicles, receptors, and scaffolding proteins that form the foundation of brain function. For me, synapses are not just abstractions—they are where my clinical and research focus resides, in disorders such as STXBP1-related and SYNGAP1-related disorders, conditions we have written about extensively in prior posts. And here is the curious thing: ask an AI system to generate an image of a synapse, and it often goes hilariously wrong. Vesicles appear in the wrong place, dendrites morph into mechanical tubes, and receptors look more like spare parts from a hardware store. It is here that the idea of “tacit knowledge” becomes important. We humans often know when something is “off” without needing to formalize why, while AI can assemble components without truly understanding their relationships. The faulty synapse drawings serve as a reminder of this gap.
Tacit knowledge. Tacit knowledge is the understanding that cannot be easily codified, the lived experience that complements formal rules. In medicine, tacit knowledge is everywhere—recognizing the subtle tone of a child’s cry in the emergency room, or knowing when a patient’s expression signals that a treatment side effect is worse than their words suggest. In rare disease medicine, tacit knowledge becomes even more critical, as so many of our decisions rely on pattern recognition across tiny patient cohorts. For now, AI struggles here. It can parse text, synthesize information, and even propose hypotheses, but it cannot replicate the lived intuition that comes from sitting with families or examining patients over time. It reminds me of the eager fellow who has read every paper on the topic but has not seen a single patient. Their knowledge is expansive yet incomplete. Similarly, AI may know the literature but lacks the practical sense-making that clinical care requires.
A path forward. I do not write this to sound pessimistic. On the contrary, I find the idea of a “gentle singularity” both appealing and realistic. Artificial intelligence will not replace the messy, imperfect, deeply human elements of medicine anytime soon. But it will help us gently overcome barriers of data access, decision-making complexity, and knowledge dissemination. That vision is more promising than the imagined takeover by omniscient AI overlords. By complementing human strengths rather than competing with them, AI highlights where human experience, pattern recognition, and ingenuity truly excel. We will still need to draw synapses ourselves, but AI can help us contextualize and interpret what we see. Perhaps the inability to draw a synapse is not a failure, but a reminder of where collaboration, rather than replacement, is the real frontier.
What you need to know. For this post, I resisted the temptation to show AI-generated synapses with their amusing distortions. Instead, I chose to include the “Ice Neurons of Delaware County,” a figure that captures how nature itself can generate neuron-like patterns in unexpected places. The larger point is this: inevitability is rarely inevitable, and the singularity is no exception. Science fiction teaches us that grand concepts often falter in the details. In medicine, tacit knowledge and human judgment remain irreplaceable, even as AI continues to evolve. The gentle singularity that Altman describes resonates with me because it reflects our lived reality: technology as partner, not ruler. And in this partnership, we can remain humble, knowing that even the simplest synapse cannot yet be convincingly drawn by a machine.
