You spent months fine tuning your neural network, constantly tweaking its architecture, training it until it achieves 99.5% accuracy, but when you deployed it in the real world with less than stellar results. Why? It’s probably because you managed to hit the limits to what our current machine learning formulas and approaches can achieve when training computers to make sense of our messy and inconsistent world. On this week’s episode we dive into some of those limits and what they mean to the humans AI is meant to help or replace.