20 years ago, or even more, I was a researcher at the CEETI (the Center of Studies on Electronics and Information Technologies) at the Central University of Las Villas, in Cuba, and at that time we were collaborating with the UNB (University of Newbrunswick). Our colleagues from the Institute of Biomedical Engineering (UNB) were working on several projects related to smart prosthesis with great accomplishments.
Of course we have access today to much more powerful digital signal processors than those we used at that "age" (like the old TMS320C10/C25/C32/etc.), but most concepts behind the hardware design remain the same, or almost the same.
The concept of deep learning (e.g. pattern matching is at the core of mioelectric prosthesis control) has gained a lot of traction and adepts and every day less people think of "data mining" as an occupation where you need a shovel and a safety helmet.
This video here (https://www.youtube.com/watch?v=ybFhDz-cMrw) introduces several key concepts and showcases some results (near time 23:00).
Probably in two hundred years or so we humans will come from the very beginning with a digital interface as a series feature for easy connection to IoT. And how the wheel works will still be there to be discovered, again.
Robots have always been very effective for precise, repetitive work, but for the most part they're also dumb as rocks. This is why robots are traditionally used only in carefully designed settings. It also explains why they cannot easily adapt to a new task, and cannot cope with an unfamiliar or uncertain situation.