My takeaways from some recent academic talks I gave at the QUT Centre for Data Science, the The University of Queensland, and the QUT Centre for Robotics:

We AI scientists like to discuss the intricacies of representationlearning, e.g. to learn better worldmodels: why for example actions need to be taken into account alongside observations to learn truly extensible, interpretable representations. How this could structure the learnt latent space respectively to be actionable, rather than just compressed and predictive. How this helps industrial and medical applications, or robotic navigation. Cool stuff!

But then also the big picture: How the technology we build interacts with and affects society and us humans. What the risks are (e.g., voluntarily giving up too much of our freedom). How to prevent those from happening.

I am optimistic that we can build a future worth living in, seeing so many researchers being commited to building it right (and ask the right questions in the first place).

Link to original article