Dear AI … can you explain your logic?

When a friend recommends a restaurant, usually he explains the logic for such a recommendation. When a doctor recommends a medical procedure, usually he presents the medical records to the patient and explains the rationale for the recommendation. When a retail sale recommends something to customers, usually he explains the base of such recommendations. When a recommendation is proposed, it is one of your rights to know the logic behind it; this will give the comfort that there is a base, and opens the eyes (and the minds) to the inputted data, and the used process, which sets the floor for more testing, and evaluation. Actually, it leads to more recommendations/ scenarios, ensuring the final result will be more solid & robust.

The story above … is not usually valid in the case of #AI recommendation systems, AI simply gives recommendations without sharing how it was produced. This is because most AI algorithms are #blackbox, and sometimes even the developers don’t know how the AI reaches such recommendations, AlphGo game was a famous example in this case! This is not a fault of AI technology or the developers, it is the nature of the mathematical equations’ underneath.

Such AI behavior, make people scared to apply its recommendations, even if the accuracy is 99%, who guarantees that this exact recommendation is not the 1% error? Not knowing the logic creates barriers to introducing AI technology, especially in operation-like domains, such as energy, power, manufacturing, and petrochemical.?

One of the #AI #ethical #principles addressing this AI limitation is “#Transparency & #Explainability”, which aims to present clearly the decision-making stages. This includes the used data, algorithm, and processes; the AI should justify the rationale, and logic behind its decision, and behavior.

The good news, AI developers are working toward more transparent AI recommendation systems, which will adopt this AI ethical concept, leading to more acceptance of AI systems. So, if you are a developer try to explore Whitebox AI technology, if you are an end-user and the AI could not explain its logic … I assure you, AI is very kind, and honest … but it has a little bit of alexithymia!