Is Machine Learning same as Machine driven Automation?

Buzzwords like “Analytics” and “Machine Learning” are very much trending in all sectors of the industry. IT solution vendors are transforming themselves to be ready and be able to cater to the demand of “Analytics” in the market.

“Machine Learning”, “Deep Learning” neural networks, various types of CNN are being matured in the market and are becoming common commodity to be used for Machine driven approach for solving business priorities.

Machine learning demand in the industry and advancements are happening at a speed of light. However, there are still lots of flux and confusion in the industry towards “Machine Learning”. This is causing obstacles in leveraging real value of Machine learning.

Is Machine Learning is same as Machine Automation? “Automation” is being referenced loosely as “Machine Learning” in the community. There is a distinct difference among both. It is very important that business solution creators should clearly understand and distinguish between the two and make effective use of both.

Way back in year 1794 Robert Street patented an internal combustion engine, which enhanced later by various scientist and industry experts. Today internal combustion engine used in locomotives and transportation vehicles. This automates the human activities of quickly traveling from one place to other with significantly less human involvement. Can we qualify this as “Machine Learning”? If not, then we need to be careful of not considering every automation in the industry that are achieved using technology/machine/computers/processors as “Machine Learning”. By misinterpreting automation as machine learning, we significantly underplay capabilities of machine learning and creates a major hurdle in the way of building effective business solutions.

Automation history is very long and it goes back to Stone Age of 2.6 million years ago where various tools developed to automate the human activities. In year 1771 Richard Arkwright developed fully automated spinning mill driven by water power. We have fully automatic washing machines at our homes now days. There are many examples of  “Machine Automation” but these are not necessarily “Machine Learning”.

Business process automation may involve Machine Learning but not necessary. Many legacy business process automation are defined based on pre-defined algorithms where machines simply follow the instructions and execute those. It does not involve automatic pattern discovery by machine and take decisions based on cognitive intelligence. Many a times we received automated emails from airline about flight booking confirmation or flight schedule change etc. No human is manually writing such emails and sending across. This is the automated process based on certain fixed algorithms but cannot be qualified as “Machine Learning” in all the cases.

The primary distinction is “Pattern Discovery”. Simple automation does not involve pattern discovery. Rather it simply uses the pre-defined patterns to take the necessary actions.  For instance, if credit card payments are defaulted, machine simply uses the pre-defined threshold of number of days to qualify as defaulter and generates automated email communication to customer. This is the pre-defined pattern that machine algorithm encompasses.

Machine learning is a paradigm shift from automation perspective. It adds cognitive aspect to automation making it more intelligent and deliver true value. The independent input features determines the outcome of dependent output features. For instance, consider a case of electricity supplier organization. They need to constantly monitor their supply network and make sure that there would not be any significant downtime in electric supply. They cannot afford to have their supply infrastructure get impacted or interrupted due to unforeseen reasons. They cannot define algorithms to consider for such unforeseen scenario simply because patterns of such kind of unforeseen interruptions are not “visible” beforehand for human to consider in algorithm,. However, they can use Machine Learning to discover the patterns based on history of electric supply interruptions and correlate it with multiple independent input features. The input independent features and output dependent features correlation along with cause-and-effect relation is not visible to human unless machine would do pattern discovery. Subsequent to pattern discovery, Machine can further remember the various such patterns and can use later to do predictions of dependent features. Independent input features could be numerous including weather history in typical seasons, electric supply demand history, any specific events that cause temporary increase in demand, social parameters and any specific environmental conditions like hurricane or cyclone that could damage electric infrastructure and many more such independent features. Such independent features along with their history data can be used to train the Machine. In this scenario, Machine does “Learn” and discover the pattern through iterative process. Once Machine “learning” is complete, Machine can take the decisions as how much infrastructure materials like electric poles, cables, connectors etc need to be stocked to ensure uninterrupted electric supply in case of unlikely event of interruptions.  

Whenever cause-and-effect patterns are know in advance and such patterns can also be directly correlated, simple algorithms are best suited without Machine Learning capabilities. However where such cause-and-effect patterns not visible, Machine learning plays a significant role to discover the patterns and enable decisions.

It is critical to understand the possibilities and limitations of automation and machine learning and apply in best applicable scenarios. It would be devastating if Automation were considered as ML. This would block all possible innovations that Machine Learning actually can bring in to solve the business challenges.