The current incarnation of machine learning is nothing more than an ultra high dimensional two year old. "Here's a new item. Which pool of millions of items is this new thing like?" "Is the red square more like these other red squares or the green triangles?"
You can get pretty far extracting the math behind the meaning, but lack of meaning is still deathly apparent.
I think most people are interested in practically solving problems. If the black box can solve it, there's no reason to care about "meaning". Once the machines have taken over all jobs, you can sit around and understand the meaning of things to your satisfaction at your own leisure surrounded by the luxury the machines have created for you.
Imagine how many more problems we could solve with meaning. Every secretary in the world could be fired. Software could actually know you. It would be a huge psychological problem for people.
You think MMOs kill brains in south east asia now? Wait until software acts like a boyfriend. Every site (read: website/app/software) could be unique, well designed, and work cross platform by telling the system what you want (remember: in this delusional scenario the system understands meaning to avoid the unintentional vengeful djinn problem).
You think multivariate testing is omgballz amazing? Imagine if ad networks share a sophisticated personality model (instead of: likes cats, doesn't like monster trucks) of you. Every site you visit could be rearranged to appeal to your individual design sense, buying patterns, and social expectations. A/B testing basically tries to suss out your average user so you can appeal to them. You leave your non-average users in the dirt by optimizing for the common moron.
It's possible and I think the incumbent giants will miss it whenever it happens (5-10-50 years out).
You can get pretty far extracting the math behind the meaning, but lack of meaning is still deathly apparent.
We should fix that sometime.