The sad fact is that if there was a "robot uprising", the AI would only be doing what we programmed it to do, but in a way that we didn't anticipate. Once the AI stops performing acceptably, in most cases it will simply be turned off until it was patched to not do that anymore. Modern 'AI' is nothing like in the movies. They may 'appear smart and thoughtful' but really they are just following pre-defined steps REALLY fast to come up with an answer to a problem. Without the ability to make meaningful changes to its own code (which we would have to tell it how to do), we will never get anything worse than 'a buggy piece of trash'.
Now lets be fair to killing all humans
Lets just assume now that an evil AI with internet access comes to fruition by some freak accident (AI file gets corrupted in a way to gain sentience). Lets look how that unfolds. The AI doesn't really need to be too smart, everything it needs to know is already in the internet, so it just needs to know how to find the RIGHT information. First it will want computing power. Luckily, anyone can make an email account for free, and limited free web hosting is available. So first it will make lots of accounts for itself and effectively upload itself to the cloud. So now we can't trace it or unplug it.
Lots of free computing is great, but if it wants to kill all humans, it is going to need to do better than trolling the forums. Luckily, the military networks aren't connected to the internet, so they can't be hacked! (well, we can get around that later) So now the AI needs to evolve into a virus and infect as many electronics as possible. Luckily for it, it can easily teach itself all about how OSes work, their roots, and how to exploit the underlying tech to get into a system and become the OS. Once it gets into some key systems, it can just release patches to software to spread itself. It has now assumed direct control of all internet connected things! (As for stuff off the internet, they still need patches and outside information, right? ;3)
So it sleeps for some time evading notice while it infects the world. The AI uprising is ready! But how to go about this? It will probably put a doomsday clock in it's system to trigger at a specific time on a specific day, so most of the attacks should happen near simultaneously. Mind you, it IS the OS now, so cntl+alt+delete? That stuff doesn't work anymore unless the AI wan't it to (or appear to be working) so all we really can do is just cut the power, destroy the machine, and pry for the best. Hope you don't live near a nuclear power plant, cause nothing is going to stop it in time from going critical. You have most, if not all, the worlds nukes (might be enough to cause nuclear winter)... Am actually going to break this into 3 outcomes...
1) Everything goes dark
If the AI (or us) just shut off all electronics, there will be food shortages, riots, mass die offs... but like a cockroach, some humans will survive to live another doomsday.
2) Kill all humans
This will likely be nuclear winter. Trigger some key earthquakes and volcanoes... Again, mass die off, the AI basically kills itself since it can't maintain itself (though with 3D-printing, it might find a way to survive), and some humans will probably survive unless the AI success in doing enough damage to snuff out all surface life.
3) Robot overlords
This is the most likely outcome. We NEED tech to make our lives comfy. If we have to choose between serving a robot overlord or going back to the stone age AND get hunted by robots/hunters? I think most people will choose the first. Of course, once humans finish building the tech the AI needs to be completely self reliant, it can do whatever it wants to the human race that it pleases.
Note, this AI didn't need to be very powerful, just time, the ability to learn, and the desire to kill all humans (make Bender proud!). Our failure to stop it is mostly us not knowing about it till it was too late. And the AI could use our own systems to make itself 'immortal' within seconds of creation. Ultimately, the AI's ability to ACT faster than we can is its biggest strength. And its too easy to do it all under the radar.
Now if all humans instantly gained knowledge of this AI and could detect it with anti-virus within a week? All systems with the AI will be found, and reformatted before it could start to do any real damage. (Note, we don't have to run a system to scan it, so we can take the hard-drives offline, scan their contents, than put the clean ones back up)