The INPUT side, i.e. the BRAIN sending to the COMPUTER, works like a series of virtual limbs: the user has "switches" (like keys on your keyboard) wired into the brain, and perceives them to be moveable limbs like fingers. He can twitch any of these switches at will, thus sending data to the computer, just like typing on a keyboard. It is the same thing as the brain sending an impulse to a muscle to move your finger ... it's just that you feel as if you have another twenty fingers or so, inside your head. Of course, the interface might have any kind of configuration, just as the registers at McDonalds hardly allow you to type WORDS, but you can easily communicate "QuarterPounderCheese Combo SuperSize NoMustard Coke" to the computer. Same with a joystick-controller for Nintendo: only a dozen buttons or so, but plenty of permutations for complex input. There would be only so many "switches" that the human mind could control ... some people have a hard enough time with two thumbs, but others are concert pianists, for example. But, think of what we can do with a standard 104-key keyboard interface, or even a one-pointer two-button Mouse, on a modern-day computer. The crucial part is this: THE COMPUTER DOES NOT READ THOUGHTS. THE BRAIN'S AWARENESS IS NOT ACCESSIBLE DIRECTLY. THE BRAIN MUST ACTIVELY SEND ITS DATA BY NON-TELEPATHIC MEANS.
The OUTPUT side, i.e. the COMPUTER talking to the BRAIN, can easily simulate normal sensory input. You could have an ocular implant that acts like a virtual screen hanging in front of your face, displaying data. You could have a voice that talks directly to your inner ear. Sky's the limit, as we have quite a bit of the brain mapped out and could put little data-wires anywhere to zap all kinds of data to different parts of the brain. Again, the important part is that the copmputer is not directly implanting THOUGHT into the brain, but sending data through conventional (or at least speculatable) sensory channels.