AI solution types and their applicability

This article tries to segment AI solutions based on inputs available to solve a problem. The article sees inputs to develop an AI solutions as:

1.      Only Data

2.      Knowledge of Subject matter experts (with/without data).

And hence categorizes AI solutions in to two segments.

The first kind is Data driven AI, which is used when there is large volume of data from numerous sources about customers, sales, market, machines, asset performance, health, energy etc. and the goal of the solution is to enhance interaction, performance or effectiveness by getting descriptions, predictions or recommendations from the solution.

The available data when combined with good and efficient statistical modelling techniques (data science approach), provides insights and knowledge about the source systems. These gained insights and knowledge help organizations to make wiser, more informed and quicker decisions based on descriptive, preventive, predictive or prescriptive recommendations.

However, there remains a limitation, this data science driven system and solution is static, which means as data format, process flow, business rules etc. changes, the result becomes unreliable and new statistical models should be deployed to work on changed data.

Machine learning algorithms provides answer for this static behavior of systems developed with only data science. Machine learning process learns data types, responses, actions and processes over time and makes itself mature on its own. This combined solution (Data Science + Machine Learning) has now become an AI solution. This type of AI solution helps in preventive, prescription or predictive recommendations along with continuous learning and maturing of itself. This kind of solution is very helpful in augmenting and enriching the decisions to take actions with these recommendations, prescriptions or predictions. This type of AI solution is created with and is made to learn with data only. And hence, this kind of AI solution falls under the category of Data driven AI. This kind of AI solution is majorly used as adviser, recommend-er or predictor.

There is another type of situation when organizations that have business processes in house for operations ranging from Health care to Customer Services, Supply Chain to Digital Marketing etc., want to increase the process output in terms of efforts and accuracy. These organizations have with them a wonderful knowledge base, in the form of Subject Matter Experts, of the operations and processes they do.

There has been a continuous wave of Robotic Process Automation (RPA) to automatize the actions done by these Subject Matter Experts. By automating a few actions within the processes, we are saving average handling time and maintaining accuracy of the fixed rules process.

However, this solution has two limitations, first the success is only partial, which means that these RPAs generally, are doing some parts of the full process and not end to end. Second, this solution is static, which means that if business rules, process flow etc changes, the results become unreliable.

Here again, if we combine machine learning algorithms over the actions performed by the SME based on his/her rational or reasoning and acceptance of those actions by Audit functionary, we can come with a solution that focuses on the knowledge substrate of the SME rather than his/her actions only. The solution developed with this approach is always learning and is updated and reliable. This kind of solution is categorized as Knowledge driven AI systems. In Business operations, Knowledge driven AI systems are most useful for end to end processing. This kind of AI solution is majorly used as some task processor.

So we can say that:

1.

Data + Mathematical and Statistical Models = Data Science; (Static)

Data Science + ML = Data Driven AI; (Dynamic)

Used primarily for recommendations, prescriptions and predictions.

2.

Process flow + Automation Tools = Robotic Process Automation; (Static)

Subject Matter Expertise + Mathematical and Statistical Models + ML = Knowledge Driven AI; (Dynamic)

Used primarily for end to end actions and tasks completion.