Let’s admit it; there is no company or individual who doesn’t suffer from a little bit of FOMO (Fear Of Missing Out) syndrome. So when the world is buzzing with AI and deep learning breakthroughs daily, engineers and executives get anxious and don’t want to miss the train. We often get calls from companies asking us to start projects about putting AI initiatives in motion. Caught on buzzwords like GPT-3 and BERT etc., they are asking us how they can get them into their product.

There are many issues with a request “I want our company to do Machine Learning or Artificial Intelligence.” The very first question that any company should ask is, “How can I make the product better?”. Yes, AI can most likely improve your product, but only if you have the right foundation in place. It is always a good idea to see what the leaders in AI like Google do. If we look at the history of Google, we see that there were two pillars of its early success before the rise of AI. The first one was its ability to collect data and do business intelligence at scale. The next step was to clean the data at scale in order to either train ML models or populate huge knowledge graphs like Google Maps. There is a misconception that this is done completely automatically with AI. The truth is that AI is used heavily, but human annotators have the last word. The extent of human crowd work is described really work in the book “Ghost work” by Mary Gray. The point is that the human-in-the-loop AI process is made possible by efficient and well-designed User Interfaces that multiply annotators' throughput of generating clean data at scale.

In other cases, UIs are designed so that the users are empowered to provide the desired data without the use of AI at all. A simple question like “Was that helpful?” can be better answered directly versus an AI guess.

At last, don’t forget that in order to train an AI model, you need a rather complex pipeline for cleaning data and creating features that are based on BI systems.

If you want to build a roadmap for AI, make sure that you set up your operations properly to optimize UI first, then see how far you can go with BI. Then you are ready to start thinking about implementing AI. We have been part of this journey several times, and we always enjoy it!