Why Reinforcement Learning is Wrong for Your Business

Questions you should ask before starting an RL project

Reinforcement learning (RL) is one of the hottest areas of machine learning research in both academia and industry. Despite all of the focus and attention, it hasn’t produced the same type of gains and widespread commercial adoption as supervised learning which has improved areas such as image classification and speech recognition by leaps and bounds. Beyond playing games, RL doesn’t seem to have done much.

There are good reasons for this. The data requirements for reinforcement learning are massive and there are numerous technical challenges that makes application of the technology difficult for most taking their first forays into the field. Despite that, many people look at the towering successes of DeepMind and OpenAI in Go and DOTA2 respectively as reason to begin looking for RL solutions to their business problems. That’s not entirely unjustified — my jump into this field a few years ago was motivated by the AlphaGo victories as well — but much of what I see in the…