The success of Deep Learning AI systems has leveraged neural networks to solve difficult problems such as self-driving cars, spotting fraudulent banking transactions, language translation, and more. These AI systems are remarkably good at analyzing huge amounts of data, spotting patterns and learning from them. They use a system of data layers, including hidden ones, to compute results. Hence it get’s a little tricky when such AI needs to explain the logic and reasoning behind a decision.

Consider an industry regulator that wishes to inspect the logic underlying the denial of a financial loan. Similarly, an investigation into the malfunction of an autonomous vehicle will throw up questions around causes. These questions point to the need for Explainable AI (XAI).

What powers robust XAI ? An answer lies in the field of Symbolic AI. When a system converts incoming information to symbols and then performs logic and reasoning, we obtain explainable decisions. This is how human intelligence generally operates.

Consider a system that examines human-generated operational text logs in conjunction with time-series data from equipment – it then applies reasoning, using in-built business process knowledge. Such a system converts complex textual information into structured form via its Natural Language Understanding (NLU) engine in conjunction with a business domain knowledge base. The system can now understand root causes of why an equipment failed, leveraging both text based and quantitative data. It also allows an analyst to interact with available information using a natural language interface. Importantly, it can explain any output fully and transparently.

XAI is likely to become very relevant as high-stakes AI applications such as complex decisions in finance, engineering, and medicine will require high accuracy and complete explainability. Knowledge-based AI – which is a form of symbolic AI – excels in XAI. It is rapidly scalable due to a flexible knowledge base.  AUI Systems provides a ready Knowledge-based AI.  We term it “Fully Explainable” or FXAI – where each output can be explained and audited.