In which of those cases is permission from Lab A not supposed to be required in order to make the product?

There are two scenarios here. In the first case Lab A doesn't license the base patent because it wants to make the product itself or (equivalently) sells or exclusively licenses it to a single manufacturer. In that case Lab B is dealing with a monopsony buyer and is hosed. The holder of the base patent can offer Lab B some fraction of their research cost for their patent, secure in the knowledge that Lab B's only alternative is to eat the entire cost and get nothing.

In the second case Lab A is licensing the patent to lots of people and you're proposing that Lab B license the improvement to the same people. The trouble is Lab A still has all the leverage. The availability of the improvement increases the value of the base patent, which allows Lab A to squeeze the licensees harder and leave less for Lab B. Lab B can't respond in kind because the licensees can do business with the Lab A patent without the Lab B patent but not vice versa.

You're introducing an alternative scenario in which Lab B is an unscrupulous patent troll with no qualms about filing frivolous lawsuits for nuisance value. Sure, they can do that, but that isn't an incentive to improve the technology, it's an incentive to engage in patent trolling. And it's just as profitable whether Lab B makes a genuine improvement to the technology or not, so it provides no incentive to actually make the improvement.

Rather than B extending A's patent, the reality is a tangled web of patents, all of which require licensing all of the rest to produce a real product without infringement. Or more realistically, the solution involves licensing from those more powerful than you, staring down those at an equal level, and infringing with impunity on those below you. This reality is known as the "Patent Thicket"[1].

I agree with you on the second part, but the counter-argument would be that the potential future marketability of the patent incentivized Lab B to do the research, which they otherwise would not have done. I'm sure there is some effect in this direction, but personally I think it's small. The sad part is that Lab B is practically forced to abuse their patents merely as a defensive tactic, on the theory that if they don't defect first their opponent will.

(I'll quit at this point, overcome by my own cynicism. Likely I agree with all your points. I started out trying to be a genuine devil's advocate, but at this point I realize I'm veering toward trollhood. I apologize, and recommend the link below for a more balanced assessment.)

[1] Patent Thickets: Strategic Patenting of Complex Technologies (http://www.researchoninnovation.org/thicket.pdf)

Such patents are often granted because the standard patent lawyers have demanded is that a patent must issue unless the patent office can prove that every single detail and step ("limitation") is documented in the prior art. Also, patent officials have very little time to search for prior art, often just a few hours; the average time allotted to examiners in the US office is under nineteen man hours for all paperwork, documentation, and searching. That is why you see so many obviously redundant and unoriginal patents. There is an official requirement for non-obviousness but the patent court -- knowing where its judges' power, prestige, and lucre lies -- has made non-obviousness a dead letter by usually declaring nothing obvious except the exact prior art itself and raising the standards of proof sky-high.

So Lab B is probably not looking for substantive improvements. It's looking -- at least where computer, information, radio, video, software, and such patents apply, but probably with graphene also -- for a loophole where it can block Lab A's product by getting a patent on a necessary but incompletely documented part of the ordinary original product. If Lab B can accumulate enough of those, then Lab A will have to negotiate with Lab B.

Only then can Lab B's actual improvements have a chance in the market. And new patents on those improvements will then be promoted by patentability. But the cost of making such a system work is astronomically higher in the sciences HN usually deals with than any benefit it ever might offer.

You should hear the patent bar talk about how lost we used to be without them to regulate us and how software would collapse without them because patents are the only way to promote true innovation in the industry and how all research would shut down because the only reason it's viable is the patents you can acquire. And they're the ones that make the law.

This is not true, of course. The thing is (and I'm sure you know this) that the law defines "obviousness" as the combination of two or more prior art references, which differs from how we colloquially understand the word. This is for several reasons, the primary ones being objectiveness and avoiding hindsight. In fact, I theorize that the latter is precisely why people think they see "so many obviously redundant and unoriginal patents".