Fidelity

Functional Tests Guide Complex “Fidelity“ Tradeoffs in Whole-Brain Emulation

Charl Linssen, Randal Koene
Journal of Ethics and Emerging Technologies, 35(1), 1-14
DOI: 10.55613/jeet.v35i1.152

Abstract

The human brain can be understood as a vast network of neurons connected via synapses, the state of which is characterized by ion concentrations, phosphorylation patterns, receptor densities, etc. It is plausible that a mechanistic simulation at the scale of the whole brain (a “whole-brain emulation“ or WBE) will be made, raising questions about moral status and personal agency. Creating a dynamical model of the brain presents a complex tradeoff between better performance, and data collection and operating costs. To make informed scientific, engineering as well as personal decisions, a set of tests should be defined, that quantify the performance of the individual on a comprehensive repertoire of skills in a variety of domains. "Fidelity" can then be defined as a measure of how well the behavior of the model corresponds with the behavior of the original individual, or with respect to stereotyped brains. Models can subsequently be optimized to obtain the highest fidelity. Nevertheless, an overall measure of fidelity is the outcome of a complex, high-dimensional optimization problem (that of choosing the parameters for a WBE) and remains in and of itself (as a measure or index) challenging to define. Different people and organizations are expected to make different tradeoffs based on a diverse set of criteria. Consequently, there can be multiple variants on offer for the translation from an original, biological brain to a WBE. If some variants are deemed cognitively superior, but are available only at a high cost, then this could have undesired socioeconomic effects where only those who are wealthy can afford the higher-tier emulations. However, competition between different WBE providers attempting to achieve the highest fidelity at the lowest cost could help drive overall costs down. A framework of ethical standards pertaining to model fidelity should be defined, which should recommend a minimum set of standardized tests.

Full paper (open access): Linssen and Koene, JEET, 2025