On the magic of 10 bits/s: A surprise constraint on the channel capacity of machine and human intelligence.

• DOI: 10.31219/osf.io/gbn3c_v1
• Publication Date: 2025-03-07T15:29:07Z
• Authors: Vladimir Chituc
• Keywords: Surprise

A recent article by Zheng and Meister [Neuron, 113(2), 192–204. (2025)] reveals what seems to be a surprising constraint on the capacity for human mind process information: limit of 10 bits/s. In this short response, I aim reconcile compelling evidence presented Meister, with no-less demands common sense: how could our minds possibly that slow? My answer is twofold. First, regularity only counterintuitive insofar as we understandthis throughput (as quantified Shannon’s entropy, theoretical construct describing un-certainty reduction in terms minimum amount information required encode somemessage) comparison everyday data transmission literal number ofbits being sent over some channel). This claim becomes immediately more plausiblewhen explicitly framed context theory. Second, factmore general than contend, since demonstrate qualitatively similar (if notidentical) throughputs even most advanced AI models (which calculated using sameprocedure applied behavior). Even more, Ireport striking convergence: performance machine intelligence (either success competitive games or benchmark Large Language Models) reliably improves it approaches bits/s, whether means speeding up slowing down. Though speculative, believe unexpected convergence raises an exciting possibility: apparent bits/s not much speed solution optimization problem, involving e.g. inherent tension between seeking new (exploring) pursuing known reward (exploiting). finally contextualize finding relation one widely (but least well-understood) papers history psychological science.