Chapter 1: Challenges in Machine Learning
Many people believe that artificial intelligence (AI) is all about the idea of a humanoid robot or an intelligent computer program that takes over humanity. The shocking news is that we are not even close to this. A better term for such incredible machines is human-like intelligence or artificial general intelligence (AGI).
So, what is AI? A more straightforward answer would be a system that uses a combination of data and algorithms to make predictions. AI practitioners call it machine learning or ML. A particular subset of ML algorithms, called deep learning (DL), refers to an ML algorithm that uses a series of steps, or layers, of computation (Goodfellow, Bengio, and Courville, 2017). This technique employs deep neural networks (DNNs) with multiple layers of artificial neurons that mimic the architecture of the human brain. Though it sounds complicated enough, it does not always mean that all DL systems will have a better performance compared to other AI algorithms or even a traditional programming approach.
ML is not always about DL. Sometimes, a basic statistical model may be a better fit for a problem you are trying to solve than a complex DNN. One of the challenges of implementing ML is about selecting the right approach. Moreover, delivering an ML project comes with other challenges, not only on the business and technology side but also in people and processes. These challenges are the primary reasons why most ML initiatives fail to deliver their expected value.
In this chapter, we will revisit a basic understanding of ML and understand the challenges in delivering ML projects that can lead to a project not delivering its promised value.
The following topics will be covered:
- Understanding ML
- Delivering ML value
- Choosing the right approach
- Facing the challenges of adopting ML
- An overview of the ML platform