Johannes Brinz
University of Osnabrück
The concept of artificial neurons has become increasingly prevalent in discussions on artificial intelligence and neural networks; however, the similarities they share with their biological counterparts remain opaque. The present paper attempts to shed some light on these issues by discussing whether artificial neurons are similar to biological neurons in that they can perform the same functions. First, I distinguish between two different notions of the term ’artificial neuron’, thereby clarifying its usage in different contexts. On the first understanding, the term refers to digital hardware implementations of artificial neural networks (ANNs), and on the
second to neuromorphic circuits that mimic neural electrochemical properties. By construing functional analyses as mechanism sketches, I argue that the functions of neurons are the roles they play in the relevant neural mechanisms. Depending on whether neural mechanisms are to be understood as computing or non-computing, those functions turn out to be either electrochemical signaling or changing their state in accordance with an appropriate computational description. It follows that ANNs implemented in digital hardware can perform the functions of biological neurons if neural mechanisms are computing and are best described by symbolic models. In contrast, neuromorphic artificial neurons can perform certain biological neuron functions, even if the brain is non-computing or better described by bio-plausible computational models.
Chair: Lucas Timmerman
Time: 03 September, 15:20 – 15:50
Location: SR 1.005
SOPhiA 2026 