When Is a Brain Organoid Ready to Learn and Think?

Abstract

Brain organoids, initially developed as models of human neurodevelopment, are increasingly being integrated with recording, stimulation, and computational systems to explore biological computing. However, most studies assess computational performance only after organoids have been selected and trained. A key unresolved question is whether pre-training electrophysiological features can predict an individual organoid’s capacity for computational learning. Here, we define computational learnability as reproducible and partially generalizable task improvement under a specified input, feedback, interface, and decoding protocol, with a demonstrable contribution from the biological system. We critically review the limited evidence for pre-training prediction and examine candidate electrophysiological predictors spanning signal quality, spontaneous network dynamics, evoked responses, plasticity, and stability. Their predictive value is likely to depend on the computational task and interface rather than converge on a universal measure. We further outline key requirements for validation, including prospective prediction in previously unseen organoids, rigorous separation of biological learning from maturation, repeated stimulation, interface drift, and decoder adaptation, and appropriate control of batch effects and uncertainty. Establishing robust predictors across organoids, batches, tasks, interfaces, and laboratories will be essential for developing electrophysiology-based functional screening strategies for brain organoid computing.