I agree completly with you. A machine client might know what link to follow (i.e. href), and what HTTP method to use (i.e. rel), but there is no way of consuming a service, even though it has a formal description (e.g. Swagger), without human assistance.
This problem is exactly what I’m working on for my Master’s thesis together with the rest of the RAD team (http://rad.ing.puc.cl/about-us/publications, it’s in construction). We aim to do REST API discovery with natural language queries, REST service composition, and the execution of such composition with as little human intervention as possible. The machine should only need to know that it needs to ‘buy’ and ‘apple’ (concepts and actions extracted from schema.org), and you would get an ‘apple’.
PD: I cannot excuse your English, as there’s nothing to excuse.