My research addresses a range of issues in perception, language and memory, using behavioural and computational methods. In recent years I have focused on what Artificial Neural Networks (ANNs) can teach us about the mind and brain. Although it is widely claimed that ANNs are the current best models of vision and language, my colleagues and I have challenged this position.  The problem is that researchers often draw causal conclusions from correlational studies.  For example, many researchers compete in how well ANNs predict on behavioural and brain benchmarks, with better predictions taken to indicate that the ANNs are mechanistically aligned with biological intelligence.  However, when we manipulate variables to test hypotheses, we consistently show that ANNs fail to align with humans.  The main message from my research is that standard methods of manipulating independent variables to test hypotheses are needed in NeuroAI before drawing any strong conclusions.