Symbol grounding is a key part of classical AI and harbours significant potential in the ability to perform symbolic and causal reasoning on real-world data. In contrast, machine learning currently operates on raw sensor data and typically without consideration of any symbolic structure in the data. The recent emergence of powerful generative modelling approaches - and unsupervised, object-centric generative models in particular - provides an exciting opportunity to explore the intersection between symbolic AI and data-driven deep learning. Key considerations for this research will be how object-centric world-models can be used in the context of symbolic reasoning and causal inference and, conversely, how such inference can strengthen the inductive biases vital for effective and efficient learning of object centric scene decompositions and predictions.