If you hard-code effective learned distributions from a trained model, I suppose that could be described as an 'AI algorithm', even though the final output is a flat algorithm.