Improving the cross-cultural functioning of deep artificial neural networks through machine enculturation

Abstract

Artificial intelligence applications are being rapidly deployed around the world, where they need to interact with humans exhibiting different sociocultural values and related behaviors. Through machine enculturation, computers are supposed to assimilate these values so that they can better relate to humans. But artificial intelligence research has not yet fully managed that challenge. This article uses a deep artificial neural network to mimic the functioning of the human emotional brain by relating value priorities, opinions, and other factors with subjective well-being. It highlights that a network's hyperparameters configured to successfully train and perform with data from one country may not necessarily train and perform well with data from another country. Advancing our understanding of machine enculturation, the analysis demonstrates that the network's performance can be improved by pooling data across countries and coding binary country placeholder variables into the input vector.