This is a really old idea that turned out to be harder than it looked, and require more computational power to be available to programmers than has generally been available. It has also required that this be available for long enough that it could build up a head of steam as people built up libraries that could be used to do real work and build runtimes that were strong in practice as well as theory.

Past failures can not always predict future failures properly in a Moore's Law regime. For instance, as everyone knows, tablet computing is a totally stupid and repeatedly failed idea, except, iPad. Computer vision is a complete waste of time, unless you have GPUs sitting around that can chew through billions of operations for cheap. Etc.

This "old idea" is getting somewhere now. It's only early days.