Approach Used by Most AI Systems
In general, there are 2 types of machine learning algorithms that can be used to create an AI systems. To cut story short, one of them is "analyse data to get a model, where the model is used to propose solution to a problem at hand", which focuses on producing highly accurate solutions for its users. On the other hand, another type of algorithms "do not create any model, it proposes its solution based on reasoning with the data every time it is given a problem to solve", which concentrates on proposing explainable solutions to the users.
Most of AI systems today use the first type of algorithms, this includes the most popular algorithm called Neural Network (NN). However, despite its success in a number of areas, in many real-world cases, NN alone cannot really solve a problem, especially when it comes to making a system that can explain why a certain solution is proposed by the system. The explanation is crucial for businesses if the AI systems expect them to consider what is proposed by the systems. In fact, without explanation, for businesses, what proposed by AI systems (or even human) is useless because it will be considered as a guess. This is one of the most talk-about issues that has been discussed widely in NN communities around the world.
Explainable AI Systems
One of the types of machine learning algorithms focus its strength on being able to explain. Human make decision mostly by referring to his/her experiences. In other words, in many cases, we solve our problems by using similar experiences that we have in the past. In this case, we can explain why we make a certain decision by referring back to those previous experiences that we use to derive at the new solution.
This type of machine learning also has its own weakness. Yes, it can explain why a certain decision is made, but the problem is that it needs a very good set of experiences (or well-analysed experiences) to back it up. The problem is that it is not easy to find the quality set of experiences from data of day-to-day operations of businesses. A number of large AI systems of this kind has to be shutdown because of unable to find such experiences for the systems.
Ditto: Hybrid AI Engine
Our approach at Innocop is quite simple, just combine goodness of both words. Since both types of machine learning algorithms has their own strengths and weaknesses, we combine both types of algorithms when design or customize an AI system for a specific question of business. However, unlike most of the hybrid approach used by others, even for the same type of algorithm, Ditto allows you to get a number of algorithms to compete each other for the best result.
Why we our engine has various algorithms?
AI is not new. There are hundreds of algorithms available for dealing different problems/conditions. Even the same type of algorithm, everyone of them has their own strengths and weaknesses. As a result, Ditto contains a number of machine learning algorithms that can be customized to race against each other in order to get the best solution for its users. The final result is an engine that can provide users with well balanced accuracy and explainability for their problems.