AI, ML and RPA A Comparison
Sorting out the differences among AI, RPA and ML
Tech enthusiasts have exhausted a lot of parallelisms between AI and human beings. So, we are resorting to compare AI with a government body. (Please don’t mind us being politically incorrect; PC is so last century after all-if you got that pun, we are in sync!)
Basic Analogies
Artificial Intelligence is the think-tank in the government; the brains of the organization. Machine Learning is an integral part of this think-tank.
RPA is the pencil pushing, red tape worshiping body of bureaucrats which merely follows the rules made by the think tank that is AI.
Basic differences
When you tell AI what to do, as in, feed it the essential data, it might start off with what you had in mind but come back to you with learnings, improvisations and suggestions. AI is an excellent self-learner; on the other hand, RPA will continue to do what you told it to, without question and without fatigue. In other words, AI is intelligence oriented while RPA is process oriented. AI does not necessarily need structured input. In fact, AI will make sense of large, unstructured data and process it, giving you actionable insights. RPA definitely needs structured input to help automate processes that are rigid and tedious.
Now that the basics are out of the way, let us get a blow by blow account of the differences existing among AI, ML and RPA:
Robotic Process Automation
Robotic Process Automation is just good old automation of tedious jobs that usually involve manual errors during execution. Customer Service is a case in point. To illustrate this claim, let us check this case study and the challenges it presented Enterprise Bot:
1-High customer service costs
2-Repeated questions from customers
3-Customer service not extending beyond working hours
among other issues.
Challenge #2 is a clear cut case for RPA. Customers are often asking the same questions to the customer service executives. If these were answered by RPA bots, this would be saving your customer executives a lot of time, which could be productively used for other tasks of high priority.
Challenge #3 is also a case for RPA, though it might not seem so on the surface. Beyond your working hours, you can limit the question suggestions to the frequent ones. This will help your RPA bots to fetch the necessary data as many times as needed without fatigue. This not only helps in customer acquisition, but also prevents the necessity of having employees on another shift, thereby reducing the customer service costs.
Note: Challenge #1 does not give you a clear picture of the factors behind these high costs. If you are not able to put your finger on the causes, then this might be a case for AI/ML, which will process a large data set and give you actionable insights.
Summarizing, RPA’s functions are:
1-Scrapping the web
2-Moving data from one system to another
(Especially useful for legacy systems)
3-Doing a said task repeatedly without fatigue
RPA’s advantages
1-Reduction of operational costs
2-Reduction of human errors
3-Reduction of labour costs
Machine Learning
Simply put, Machine Learning is the part of Artificial Intelligence that handles pattern recognition. The primary aim of Machine learning is to develop computer programs that can access large amounts of data and arrive at set patterns, which in turn can aid Artificial Intelligence’s predictive ability that can develop actionable insights for the end user.
While there are many machine learning algorithms that enable AI to recognize patterns, there are 3 main categories:
Supervised machine learning algorithms
The concept of machine learning is the same across all algorithms: using a large data set to draw and observe patterns. These patterns may help predict future events. In the supervised machine learning method, the system fetches data from a labelled data set and uses this learning to make predictions about output values. Say you are developing a program to generate a 5x5 sudoku, the system can initially fetch data (which ideally has all moves possible to solve the Sudoku set) from the Easy label and then define Medium and Hard through training.
Unsupervised Machine learning algorithms
Let us take the same example of a 5x5 Sudoku. There is no label on the data set this time. So, the system digs into a data set and then from what it gathers, labels if it is Easy, Medium or Difficult.
Semi Supervised Machine Learning Algorithms
If unsupervised Machine Learning algorithm is not returning the expected result. For example, when the system generates a Difficult level of 5x5 Sudoku when the user has chosen an Easy game, it is not the intended result. So, there is a bit of labelled data, nowhere close to the amount of data present in the Supervised method, to train the system to deal with unlabelled datasets.
Reinforced machine learning algorithms
This method of learning interacts with the environment and offers feedback and reward. For example, if the average time to complete the Easy level of a 5x5 Sudoku is 5 minutes, there might be a reward if you beat the time. This reward is called the reinforcement signal. Of course, the system should have a large data set which comprises data about time taken to solve the puzzle and data about the various levels of the puzzle.
When does AI come in?
Let us say Machine Learning algorithm can teach a system to generate or solve a 5x5. It can generate as many sets as the permutations let it. However, it cannot solve a 9x9 sudoku. It recognizes the pattern but lacks the consciousness or the abstract thinking ability to solve the puzzle. This is where AI takes over from ML.
Artificial Intelligence
This technology provides intelligent predictions and actionable insights to the user by processing large sets of unstructured and semi structured data. AI captures crucial data for analysis using:
Vision recognition
Sound recognition
Search: extracting data
Advanced data analysis: Pattern Recognition (Machine Learning)
There are two types of AI:
1-Weak AI: Executes basic functions of AI without being self-aware and conscious. For example, a weak AI system may mimic a human speaking French by providing French translation from English speech patterns.
2-Strong AI: This tool is conscious, self-aware and intelligent (apparently enough to scare the wits out of Elon Musk) The strong AI system actually learns the colloquialisms involved in speaking French and gains intelligence over time and uses data pertinent to the location and recognizes the accent variations. So, instead of mimicking a person, it becomes a person, assuming a personality.