Have a look at this: https://www.vice.com/en/article/y3pezm/scientists-increasingly-cant-explain-how-ai-works

Here's a controversial one for you: https://arxiv.org/pdf/1503.02406.pdf and his talk: https://www.youtube.com/watch?v=utvIaZ6wYuw

Also figure 3 in https://arxiv.org/pdf/1703.00810.pdf

I think Batch Normalization is a great topic

How about tackling climate change with machine learning? It's not a new method, but would inspire people about great neural net applications.

https://m.youtube.com/watch?v=ppU_CYvrNjI

Generative networks are interesting and have fun results and therefore fun to show off.

Have a look at VAE, GAN or diffusion models that generate images.