Averaging is a convenient fiction of neuroscience

Averaging in Neuroscience and Signal Processing

• Several commenters liken averaging to a low‑pass filter: it smooths away high‑frequency structure that might contain meaningful temporal codes.
• Distinction is made between smoothing within a single spike train (uncontroversial) and averaging across many trials with variable timing (can erase real structure without being a true low‑pass filter).
• Some describe models where gamma‑frequency oscillations gate which neurons fire in brief windows, explicitly using high‑frequency timing rather than rate codes. Others note the mechanisms and roles of brain rhythms (e.g., gamma) remain uncertain and may not map cleanly onto “clock” metaphors.

Limits of Averages and Statistical Misuse

• Strong criticism of treating averages as complete descriptions, especially in skewed, bimodal, or fat‑tailed distributions.
• Examples: aircraft cockpits designed for the “average” pilot fit almost no one; policies aiming at the “middle” can fail both sides.
• Some argue problems arise less from normal‐distribution assumptions and more from practices like model post‑selection and weak statistical literacy.

Historical Life Expectancy Debate

• One side claims ancient adult modal lifespans were in the 70s–80s and that low averages mainly reflect child mortality.
• Others counter with historical data (e.g., population pyramids, UK records, Our World in Data) showing substantial adult mortality and improvements at all ages, not just infancy.
• Disagreement over the usefulness of mode vs mean for life expectancy; consensus that averages alone are misleading but no agreement on exact historical numbers.
• Some distrust specific data sources and highlight uncertainty in paleodemographic reconstruction.

Neural Coding vs Artificial Neural Networks

• Timing of spikes, not just firing rates, is emphasized as critical in biological neurons, unlike most machine learning models.
• Comparisons suggest ANN “neurons” more closely resemble aggregate firing‑rate functions (e.g., F‑I curves) than real spiking units.
• Multiple comments stress that brain computation involves synapse counts, glial cells, neuromodulators, astrocytes, and plastic, dynamic wiring, making simple neuron/parameter comparisons to ANNs inadequate.

Measurement Limits, fMRI, and “Scientism”

• Discussion that scientific consensus often follows what tools can measure; averaging may be a pragmatic “streetlight” strategy until better tools exist.
• Concerns raised about over‑interpretation of fMRI “areas lighting up” and about oversimplifying brains as pure spike‑based digital systems.
• Extended side‑thread critiques “scientism” (treating current scientific consensus as final dogma) versus science as creative, skeptical, evolving inquiry, with mentions of replication problems in some fields.