Explainable AI
Cheat Sheet
Explainable AI
Want to know why AI gives a particular answer to your question? Darien Graham-Smith explains
@PCPRO
FACEBOOK.COM/PCPRO
What’s the difference between “explainable” and regular AI?
When we talk about explainable AI, we’re not referring to a specific service. Sometimes shortened to XAI – but not to be confused with Elon Musk’s xAI company – explainable AI is a collection of techniques used to analyse AI functions, with a view to understanding why a particular input generates a particular response. These techniques can be applied to either mainstream cloud AI services or local frameworks.
But aren’t modern AI systems inscrutable “black boxes”?
Practically speaking, yes they are – and that means explainability isn’t an exact science. Because of the complex, non-linear way AI engines work, it’s effectively impossible to say for sure why a particular response was generated. So explainability tools don’t try – instead, they might create a model of how the AI approaches a particular operation, or weight outcomes to map out what makes its reasoning tip one way or another. They can then use this information to analyse a real-world problem and suggest why the AI gave the answer it did – subject to certain limitations.