Development works like a pattern triggered on and off by various genes. Certain parts of the program code for certain things so things like arms and legs are triggered by limb buds and follow very similar programs. The primary bone is a single bone be it the femur or humerus then a joint and then a lower limb of two primary bones (tibia and fibula or ulna and radius). The same pattern is followed with modifications.
For example the knee cap is actually an ossified tendon that evolved early within the placental mammals. You won't find them in marsupials or any more diverse tetrapods. Likewise other tetrapods like some of the hoofed ungulates have modified limb ends, or sloths have fewer digits, or pandas have heavily modified wrist bones to make a thumb (look at their hands they have 6 digits because their thumb is not a digit).
So biology and genetics is more like a toolkit and the various cells and different cell types follow a decentralized build scheme to invoke specific organs in specific places. So it only takes one mutation to make limbs into hooves or to make humans largely hairless (by changing the growth rest cycles) rather than coding for individual parts.
This evo-devo scheme of developmental biology combines the understanding of evolution, development, genetics, and embryology into a single coherent field. So we know interesting things like octopuses, like squid, have ten legs but, unlike squid, have two of their legs suppressed by other genes. So you'll encounter an occasional "mutant" octopus with nine or ten legs, but never eleven. Or that the gene used in mice and fruitflies to start the program to build an eye is the same, even though mice have a lensed eye and fruitflies have a compound eye. You can simply move the gene from one to the other and it works just as well. Or that snakes still have functional leg genes that don't get activated and are just invoked elsewhere for non-leg purposes.
To properly change a lot of the patterning you need to change a lot of the structure, you'd need to make another upper torso section or otherwise cause the patterning to start the arm creation program multiple times. But, as the program only really takes small tweaks it would be hard to cause such a duplication and have a viable resulting phenotype.
It's like asking if there's some step you could tweak in making a chocolate cake that would actually end up making cookie dough icecream instead. If you want to gradually make a larger cake or tweak the coloring or flavor, that's easy, but you can't make a small tweak to the recipe to make a cake into cookie dough. Even if cookie-dough icecream were a great thing to make, using slight modifications of a cake recipe you'd likely end up with a cake that mimics cookie-dough icecream rather than the real thing. You can't really change the plan you already have down in the books, that's the reason why animals have vestiges. It's why humans have an eye with a blind spot. It's why aquatic flightless penguins have wings. You can only tweak the previous recipes, rather than change them wholecloth. And the recipe for land animals with four limbs was set down during the evolution of fish prior to migration on to the land. Tweaking that bit is going to cause serious problems for every step that comes later, which is pretty much every step.
We can tweak the expression of genes a lot easier than the pattern. So longer arms or more hair or to change the degree of anything is easy. However to actually change the recipe itself is not so simple. Which is why things typically evolve gradually from very small nearly inconsequential traits to large robust developments. Rather than encountering freakish but awesome new organs like sharp spikes on one's knuckles or extra arms or whatnot.
Humans like all the tetrapods are really just a jerryrigged fish.
Sorry if I'm cryptic at times, but evo-devo is very impressive science and somewhat cutting edge biology. And is actually the correct answer to why if extra arms might be evolutionarily advantageous could humans never evolve them. A lot of organisms get locked into certain patterns. For example humans have seven neck vertebrae, want to venture a guess as to how many a giraffe has? How about a mouse? A cat? A dog? A cow? -- The answer to all of these is always seven. Giraffes and humans have the same number of neck vertebrae they just get longer and shorter rather than more common like one would find in long necked birds, because, for some reason, nearly all mammals are locked into this pattern and the recipe is really hard to change.