Here's a bit of interesting news regarding a human-machine interface from MIT: Brain-controlled Robots.
A new system from MIT CSAIL uses EEG brain signals to detect if a person notices robots making a mistake.
They're a long way from a symbiotic relationship between humans and machines, but this is a good start down that path. This technology strikes me as a clever way to train your robot without the need to sort your dataset up front. The EEG methodology is crude and requires its own customization (a major impediment to general purpose application), but it's easy to imagine a future where we have implants that will give us better information about our thoughts.
Most adults will instinctively correct the behavior of babies and toddlers, and children have an equally native response that cues them to our sentiments. The type of technology that MIT is working on will let us do the same for our future not-so-human helpers. The next question in my mind: how do we make sure our robot toddler doesn't go through a "No!" phase?