It’s not what AI can do for you

Virtually any company that I work with is exploring its data sets and business processes to identify opportunities for productivity improvements, higher accuracy or lower cost. The constant question that these companies struggle with is how can artificial intelligence, and specifically machine learning (ML) and deep learning (DL), support existing processes and ways of working. This results in a number of challenges.

First, the data available for training ML/DL models was typically generated for use by humans. This leads to data scientists having to spend significant amounts of time to preprocess and clean that data to make it suitable for use by machines. Second, the typical mindset is that AI solutions are there to assist humans to do a better job. In many cases, though, the best approach forward is to completely automate a specific process rather than creating an intermingling of automated and human actions. The primary challenge, however, is that the approach of looking to automate parts of human-oriented business processes will only allow for smaller, limited improvements rather than the order-of-magnitude improvements that we would like to see.