Robots do NOT learn. They take data and do with it what they are programmed to do.
Also no robots will not have a conscience because that would require someone to program that into them and I see no beelnifit in that.
It would've been true for early-generation AI's. Computers acted on algorithms they were fed, which made them resemble enough human-like behavior to instill all sorts of fear in media (Terminator is a classic example). Back when Kasparov got defeated by deep blue, it was quite a discussion on whether that was really a program or not. But at best, that was a showcase that given enough options and calculation time, a computer could determine the best outcomes in the long term and act towards that.
The newer AI's (the neural networks that @Vego mentions) may start with the same principle, but their execution is..."more refined" isn't the proper term anymore. From what I gather, it would be like saying an adult is a "more refined" version of a child. They still seek of patterns and apply the best ones, but in the case of the world championship Go tournament, the AI sort of "invents" a new strategy to apply to win. The AI clones itself onto wherever calculation space it has to search for the best answers, and eventually loses/replaces those with the least favorable answers with the better ones.
You know how you have pets?
No...sorry, but I don't think that analogy is correct. Computers may will become smarter than humans, but we'll always remain conscious, aware and able to communicate. And because computers are always built with the idea that they serve us, that'll be a treat they won't "accidentally" forget (that stuff is more dark sci fi than likely).
The better analogy IMHO would be of the office world. A rule of thumb for managers is to always hire people that are smarter than him/her. This makes sense, because it'll increase the chances of reaching office goals, whereas hiring equaly smart or dumber people, it'll only serve to stay at the status quo.
What'll happen, though, is that intelligence will become overrated (more so than today, probably). Computers might (and will) learn to drive better than humans...but that doesn't mean humanity is going to give up their steering wheel.
I predict a worldwide discrimination against solutions brought forward by (or aided by) advanced AI. If a computer tells you something you don't like (say...how to lose weight), it'll be cool to dismiss it as coming from someone not human.
...but meanwhile, everyone will be using it. Take chess as an example: at this point chess computers beat professional players any time of the week. So why is there still a worldwide tournament going on? Because the computers don't matter. Even while smarter, they're more seen and used as interesting tools than anything else.