Machine Learning (ML) and Artificial Intelligence (AI; also often referred to as “Deep Learning”) are two key tools of Data Science, and thus important levers to extract value from digitized data. In fact, AI is a subset of ML, using neural networks to mimic the capability of human brains to tackle a complex task through continuous learning.
While some people – rightfully – underline the risks of ML and AI (e.g., loss of qualified jobs, algorithms taking over control of our lives), others praise the new opportunities springing from this technology. Digitization will indeed destroy jobs, but many of these jobs are in essence highly repetitive, if not unpleasant. There are for instance professionals in many companies dealing with customer complaints. While it requires a good training and reasonable judgement to do this job, it is still boring and depressing. AI tools can already today handle 70…80 % of all written requests and complaints, and improve their capabilities every day. Professionals get relieved from these unpleasant tasks, and become available for more inspiring and value-adding activities.
Besides the ethical issue above, the major question is what ML and AI can do – and what they can’t do. As stated above, AI algorithms (e.g., neural networks) mimic the learning approach of human brains. In return, this means that if a human brain cannot solve a certain problem, AI is not able to do it either. Furthermore, human brain is a great generalist, capable of tackling a wide range of very diverse tasks – based on (mathematical) logic, acquired experience / background knowledge, and mere intuition. At this point in time, there is no such thing as a generalist AI – neural networks need to be developed and trained for each and any case. Not necessarily from scratch, but with considerable customization.
For a given logical problem, AI will never be smarter than an experienced human being, since it lacks experience and intuition. On the other hand, getting close to human capabilities is for many applica-tions more than enough to relieve professionals from a repetitive and boring task. Plus there is one exception from the above statement: if it comes to solving an issue by applying complex rules (such as a handbook for a certain process), an AI algorithm will always be perfect in knowing and applying the rules (once it has learned them), while human beings tend to get tired and oversee things. In any case, here are three reasons why certain issue cannot be solved at all by ML / technology1):
Having clarified what ML / AI can’t do, what is it that they cando? There are at least five major fields of applications: