Let me start by addressing the lack of microphones that can do this. There isn't one. We have specialized highly directional microphones that can pick up very specific points at fairly long distances, however they require being very carefully aimed.
The area that a microphone picks up sound from is called the pickup pattern. Most consumer mics (and many professional ones) use either omni-directional (picking up in all directions) or cardioid (picking up mostly in the general front of the mic, but a little bit all around). There are also shotgun mics which are typically rather long mics which use super cardioid or hyper cardioid patterns that are much more highly targeted in picking up only sound directly in front of them. There are also specialized distance mics that use large reflectors to direct sound from a particular source.
The reason the microphone can't discern the sound is a combination of the wide pickup pattern (which means it gets sounds coming from all directions) and the fact that the intensity of sound falls off very quickly, so it doesn't take much distance for the actual "volume" of the sound to drop off to the point it is a similar level to that of the surrounding background noise.
So this only leaves the question why you, as a human, can understand what they are saying still when you are in the room. The answer to this is actually very similar to the pickup patterns of the mics, but much more complicated in practice. A microphone hears from a particular point, your ears hear from two points and compare between them. Your brain is very, very good at processing what it hears and can localize the source of sounds based on the delay between when sound reaches your ears and subtle ways that the shape of your head and ears impact the sound. The end result is that your brain can isolate the speaker and only pay attention to information coming from that source while filtering out the noise that doesn't matter to it.
There actually are forms of binaural recordings that use multiple microphones and sometimes even a mannequin head to achieve a recording that can be played back on headphones to produce something which sounds more natural as if you were physically in the space the recording occurred. HRTF (head related transform function) is a technique of modeling the impact to sound from the delays and changes your head and ears make and allows producing virtual surround sound with only a pair of headphones.
Both binaural recordings and HRTF sources use the same principles, it just differs in whether it results from an actual recording or from simulation of what a recording would capture. Either way, they are both utilizing the same property of our ability to localize sound with our brains very complicated signal processing, which is the same signal processing that lets you listen to someone in a noisy room.