I agree. ML is best explained not from a CS angle but from a maths/physics/biology angle, starting with a Single Layer Perceptron. It's possible to explain the idea of feed-forward/backpropagation computation visually without having to dive into the math of matrix multiplication, and certainly in isolation from other CompSci theory.

Personally I've found this approach, in combination with a visual explainer like the Tensorflow Playground, really effective in getting people to 'get it'. I know it doesn't even begin to cover the breadth of today's ML techniques, but as a conceptual introduction that requires little to no CompSci foundation, it works and it's a start.