Mind over metrics: How can we tell if two brains (or AI models) are alike?
The ability to record from large populations of neurons has triggered the development of myriad methods for comparing them. But we’re still grappling with how to convert measures of likeness into a better mechanistic understanding.
Neuroscientists possess the ability to record from vast populations of neurons, which has led to the creation of numerous methods for comparing them. However, converting these measures of likeness into a deeper mechanistic understanding remains a challenge. Biologists have long used comparative analysis to understand the functioning of organisms, a method that Darwin famously employed to support his theory of evolution.
In neuroscience, comparative analysis is vital, as it aids researchers in building a mechanistic understanding of the brain. For instance, the correlation between renal medulla thickness and an organism's ability to produce concentrated urine in the 1960s provided crucial evidence for the countercurrent multiplication mechanism that underpins kidney function.
Comparative analysis is already a significant part of neuroscience, with neuroanatomical atlases being used to identify homologous brain structures across various species. Specific correlations can also be drawn, such as the relationship between hippocampus size and a species' spatial navigation ability. Recently, there has been a growing interest in developing new forms of comparative analysis between large populations of co-recorded neurons.
For example, comparing neural responses between the same brain region in two animals can help determine if the responses are similar or different.
Technical advancements in recording devices and artificial intelligence have made it possible to compare large populations of neurons in mammalian cortical systems. The field is overcoming obstacles related to the scarcity of neurons available in individual animals, thanks to cheaper, miniaturized, and standardized recording technologies.
Additionally, modern artificial intelligence models share some similarities with biological systems, despite their differences in spike-based versus analog communication modes. However, the question remains: how can we link diverse datasets through the principle of comparison, and what does it mean for two neural systems to be alike?
How can we rigorously quantify this likeness and convert these measures into a better mechanistic understanding of the brain?
Written by urgent.news from The Transmitter's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
