Historically, biological plausibility was an important component of neural network and connectionist approaches to artificial intelligence and modelling cognition. Despite this, the approach has been frequently criticised for being too divorced from biological detail and neuroscientific research. Recently, however, there appears to be a revival in interest in trying to build more biologically realistic neural networks. This renewed emphasis on biological realism raises several questions about these models: How do we determine which biological features are relevant for inclusion in these models? How much detail is needed for a model to be “biologically realistic”? And why is a greater degree of biological detail desirable for neural network research in the first place?

In this talk, I will focus on two kinds of approaches to building “bio-inspired” computational models and describe examples of these approaches from research on vision. On the one hand, some researchers take a ‘biology-first’ approach where they focus on incorporating relevant features from biological systems into their computational models. For example, this approach is used in the Chittka lab (such as Roper, Fernando, & Chittka (2017) and Vasas & Chittka (2018)) to build simple neural networks that draw on the neural circuitry of bees. On the other hand, some researchers have a ‘function-first’ approach where they prioritise the performance of their models and then make comparisons to biological systems. For example, this approach is used by the DiCarlo and Kriegeskorte lab who use representational similarity analysis and the Brain-Score platform to allow researchers to compare computational models by how “brain-like” they are. (Schrimpf et al., 2018) I will argue that these approaches have different goals and thus consider different features to be biologically relevant to their models. Finally, I end by discussing some general conclusions that emerge from comparing these two approaches. This is work in progress so I look forward to hearing from the CNBC community about their perspectives on this research.