Except that the claims of strong AI 'real soon now' have been coming since the '60s. Current AI research is producing things that are good at the sorts of things that an animal's autonomic system does. AI research 40 years ago was doing the same thing, only (much) slower. The difference between that and a sentient system is a qualitative difference, whereas the improvements that you list are all quantitative.
Neural networks are good at generating correlations, but that's about all that they're good for. A large part of learning to think as a human child is learning to emulate a model of computation that's better suited to sentient awareness on a complex neural network. Most animals have neural networks in their heads that are far more complex than anything that we can build now, yet I'm not seeing mice replacing humans in most jobs.
Given that most of this code was originally targeting systems from the 1960's and 70's, I can't imagine there being an insurmountable number of lines of code
According to Wikipedia, Gartner estimated about 200 billion lines of COBOL code in 1997. To put that in perspective, that's more than the total amount of open source C code tracked by OpenHub.net. Can you imagine persuading someone to rewrite all of that C code in a newer language?
Don't read a book. Go start a business. "Entrepreneurship" books are largely useless, in my opinion (as a successful entrepreneur).
While I can definitely respect the sentiment, I also like to do a bit of research on things before jumping in. Talking with entrepreneurs (both those are/were successful and those who weren't), I did like The Opportunity Analysis Canvas as a way to help one see the opportunity in the first place (something with which I continually struggle).
Having spent quite a bit of time over the last two years to re-implement in Java a system developed by the government in COBOL, I can tell you that COBOL will probably never die. For example, keeping precise, penny-perfect calculations of dollar amounts in Java is actually quite a pain. This is especially true when the calculations involve dozens of hundreds of steps. My solution in Java has been based around BigDecimal, which makes the code very difficult to read. Aside from that, I have spent the vast majority of the time writing very extensive tests and chasing down really small numeric discrepancies. Guess what, if you decide to replace a COBOL system that does any appreciable amount of math, you would get to do the same thing. Plus, you will never be sure that you found all the bugs.
The project actually considered the possibility of licensing a commercial Cobol runtime for PC-based platforms (e.g., Windows, Linux, etc.), but that was not feasible for several reasons.
COBOL is still remarkably good at quite a few things and leaves out lots of the bells and whistles that tend to become distractions in the hands of undisciplined programmers. My only complaint about COBOL (especially old COBOL) is that the control flow is a real pain. Aside from that, it is definitely a workhorse of a language. No need to go killing it off yet.
Then, almost by definition, it is worthless
And yet it works in exactly the way Libertarians are telling us things will work: companies put an agreed-on label on their products, they have an incentive to check unreasonable-sounding claims from their competitors as do consumer groups, and there is redress through the courts (and bad publicity) if anyone is caught cheating. For once, it's a free market solution that is working with a minimal amount of government intervention.
Frankly, Scarlett, I don't have a fix. -- Rhett Buggler