Explainable AI
Explainable AI (XAI) is a class of emergent technology that is defined by the set of methods which humans can comprehend, that improves their trust of the output provided by the AI1
AI adoption by people increases with its explainability and transparency in functioning. People ought to know what decisions an AI is taking and how it has calculated something. Presently, a lot of AI models are black boxes2 (They’re called Opaque models). This means that using AI in high-stakes environments such as autonomous driving, healthcare, criminal justice, and finance may be costly and prohibitive2. If something happens, it’ll be hard to trace back and find out the precise cause
XAI techniques
- Pre-modelling - Analysing the data being used to train an AI model
- Explainable modelling - Adding interpretability to the AI model’s architecture
- Post-modelling - Producing post-hoc explanations of the AI’s behaviour
XAI websites
Current limitations of XAI
- There doesn’t seem to be a consensus on various key definitions and processes. If there’s no consensus, then XAI systems may fail to provide information that everyone involved in the ecosystem may understand or agree with
- Although the number of papers on XAI is increasing, real-world applications of it are scarce. Non-AI experts are yet to embrace XAI because of this
- The chances of making mistakes explaining an opaque AI system’s workings is high. Instead of attempting to make opaque models explainable, it may be better to make a shift to building inherently interpretable models