Humans and ML Models: Are They Really That Different?

October 11, 2019

Would you argue that the brain is the most important organ in the human body because it is the body’s control center? Perhaps, you would! Might you try to convince somebody that the human heart is, in fact, the most important organ in the body given its ability to beat independent of the brain? You surely might! Would you present the holistic view that the human body is a complex system comprised of many parts, each with its own identity, working together in an organized manner for the benefit of the total being? You probably should!

Like the human body, Machine Learning is a combination of several component parts. First, there is raw or unstructured data, then there are tags to that raw data or annotation methods, then there is the “algorithm” or supervised learning algorithms. Like all systems, in Machine Learning, different parts come together to form a single entity. When you look at the human body you see a single outcome; you do not see the five vital organs and 11 body systems that comprise it. Similarly, when you look at a Machine Learning model you see a single output; you do not see the multitude of components and processes that go into preparing it.

The Machine Learning process is complex and comprised of many subprocesses that shape the actuator input to achieve a specific application output. The different Machine Learning processes are all of tremendous value. First, there is the data collection and preparation process. Only after this process has begun can the process of feature extraction begin. After structure is brought to the data, the process of training the algorithm begins. The algorithm training continues unabated after the model is deployed and for the duration of the engagement. In Machine Learning, the model is always being evaluated and depending on what is being learned through the evaluation process a practitioner may change the way the data is prepared, adjust the features that are being extracted from the data or fine-tune the algorithm.

The complexity associated with Machine Learning suggests that no one component part and no one subprocess is more important than another. As a result, innovative business leaders who are relying on Machine Learning to help them plot courses in the AI age will need to invest heavily in data, heavily in data labeling and annotation, and heavily in the building of Machine Learning algorithms and models. — Mark Papia

(Mark Papia is the Director of Content and Communications at iMerit.)