Perceptual sensitivity is explained by optimization for ecological tasks
This paper demonstrates that human perceptual sensitivity and discrimination thresholds are best explained by neural representations optimized for ecological tasks like sound localization and recognition, rather than by internal neural noise.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
Human hearing is a marvel of biological engineering, capable of detecting the faintest whisper against a roar of city traffic or distinguishing the subtle pitch of a bird call from a distant engine. Yet, this incredible sense has hard limits. There is a smallest difference in volume, pitch, or timing that the ear can detect; below that threshold, two sounds become indistinguishable. For over a century, scientists have tried to explain why these limits exist. The traditional answer was simple: our biological machinery is imperfect. Just as a camera sensor might have grainy pixels or a microphone might pick up a static hiss, the human brain was thought to be limited by internal noise—the random, chaotic firing of nerve cells that blurs the signal. This view suggested that our sensitivity is capped by the unreliability of our own neural hardware.
However, a different idea has been gaining ground: that our hearing limits are not set by how noisy our brains are, but by what our brains need to do. This perspective, known as efficient coding, proposes that the sensory system is a highly tuned instrument shaped by evolution to solve specific problems in the real world. If the brain is optimized to recognize a friend's voice in a crowd or locate a predator in the bushes, it might not waste energy representing tiny differences that don't help with those tasks. In this view, the limits of hearing are not flaws, but features—boundaries drawn by the demands of everyday life.
To test this idea, researchers built a computer model designed to mimic the human ear and brain, but with a crucial twist. Instead of programming the model with rules about how hearing works, they trained it to perform natural tasks: finding where sounds are coming from and recognizing words, voices, and environmental noises like a barking dog or a crashing wave. The model started with a simulated version of the auditory nerve, the bundle of fibers that carries sound signals from the ear to the brain, and then learned to solve these ecological problems through trial and error, much like a child learning to listen. The researchers then asked a simple question: if a system is only optimized to survive and navigate the real world, will it naturally develop the same hearing limits as a human?
The answer, they found, was a striking yes. The researchers measured the model's ability to detect tiny differences in sound across hundreds of different scenarios, covering everything from the faintest tone to the most complex rhythm. They compared these results to thousands of measurements taken from human listeners in decades of psychological studies. The model, which had never seen a single human hearing test and was never told what a human threshold should be, produced results that closely matched human performance. It could not hear differences smaller than a human could, and it struggled with the exact same types of sounds. This suggests that the limits of human hearing are not primarily caused by the random noise of our nerves, but by the fact that our brains are optimized for the specific tasks of everyday life.
To be sure that this match wasn't just a coincidence, the researchers ran several control experiments. They tested the model before it had learned anything, using only the raw input from the simulated ear. This untrained version was far less sensitive than a human, missing many differences that people can easily hear. They also tested a version of the model that was trained specifically to pass the hearing tests themselves, rather than to navigate the real world. This version became superhuman, detecting differences far smaller than any person could. This proved that the model had the capacity to be much more sensitive, but that its "human-like" limits emerged only when it was trained for natural behavior. The fact that the model matched human limits so closely without being told to do so suggests that the brain's sensitivity is determined by what is useful, not by how noisy the hardware is.
The study also looked at where in the model these human-like limits appeared. The researchers peered into the different layers of the computer network, which roughly correspond to different stages of processing in the brain. They found that the ability to distinguish sounds improved as the information moved deeper into the network, eventually settling into the human-like pattern at the same stage where the model began to specialize in different tasks. This indicates that the limits of perception are not fixed at the ear but are shaped as the brain processes information to make sense of the world.
There were, of course, some differences. The model was slightly better at some things and worse at others than a human. For instance, it struggled with certain complex patterns of sound masking, where one sound hides another, and it did not fully capture how humans hear very high-pitched tones. These discrepancies likely point to specific biological details the model missed, such as the complex, non-linear way the inner ear compresses sound or the specific timing mechanisms used for pitch. But these were the exceptions; for the vast majority of hearing tasks, the model's behavior was a near-perfect mirror of human performance.
The implications of this work are significant. It suggests that the boundaries of human perception are not arbitrary or merely the result of biological imperfection. Instead, they appear to be the result of a system that has been finely tuned to the demands of the environment. The brain does not try to represent every possible detail of the acoustic world; it represents what is necessary for survival and interaction. By optimizing for the tasks of everyday life, the system naturally arrives at the same limits that humans have evolved to possess. This challenges the long-held belief that internal noise is the primary reason we cannot hear everything. Instead, it suggests that we hear exactly as well as we need to, and no better, because that is what is required to navigate the world we live in.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.