Mirror neurons are cells in monkey frontal area F5 that respond both during the execution of action and during the perception of action. Explaining why these cells respond during action execution is easy and uncontroversial: they are motor cells in a motor area -- they respond during action execution because they are involved in the coding of actions. The perceptual response is more difficult to explain. Think first about "canonical neurons", neighbors of mirror neurons in F5. Like mirror neurons, these cells respond during action execution, e.g., grasping, and they also have sensory properties, e.g., responding to the presentation of graspable objects. The sensory responses of canonical neurons have a fairly intuitive and standard explanation: the grasping of objects needs to be informed by the shape of the object (you grasp a paperclip differently than a grapefruit) and so the sensory input is used to drive appropriate grasping gestures. Importantly, canonical neurons are not assumed to be responsible for visual recognition, they just receive relevant input from areas involved in the processing of visual features.
But what about mirror neurons? Why would the percept of someone else performing an action such as grasping a piece of food help guide the monkey's own food-grasping action? One thought is that mirror neurons support imitation, but apparently macaque monkeys don't imitate so that can't be right. So the theory that was proposed early on and completely dominates (suffocates even) thought on mirror neuron function is that these cells support action understanding. According to this view, the sensory response of mirror neurons is not relevant to the monkey's own actions, unlike canonical neurons. It is rather a mechanism for understanding what other animals are doing via motor simulation. The logic is, if I understand what I'm doing when I reach for a peanut, then if I can simulate another's peanut-reaching action in my motor system, I can understand what s/he's doing.
I have argued that the action understanding theory of mirror neurons has never actually been tested in monkeys and where it has been tested in the "human mirror system" it has been proven wrong: damage to the "mirror system" does not necessarily cause a deficit in action understanding (Hickok, 2009). I have yet to see a strong empirical refutation of the evidence I discussed, but a common response that I do hear is, "you propose no alternative theory of mirror neurons."
Although I've never been fond of it's-the-only-game-in-town arguments (a theory can be demonstrably wrong even if we don't yet have a better theory) I think the point is worth taking seriously even if it is only partially true. I did propose that mirror neurons reflected a form of motor priming, but didn't develop the idea in any detail.
In response to the only-game-in-town argument, here is what I'd like neuroscientists to do, just for fun. Rather than obsessing on the idea that the sensory response of mirror neurons has no relevance to action execution, I'd like folks to at least consider the possibility that mirror neurons, like their canonical neighbors, take sensory input for a motor purpose.
I'll start. This is speculative and unsubstantiated, but so is the action understanding theory and you have to start somewhere. Consider this a jumping off point for discussion...
Can we learn something from the behavior of dogs? If you've played fetch with a dog you may have noticed that it quickly learns to anticipate the consequences of throwing actions. For example, it is not hard to fool a naive dog who plays a lot of fetch with a fake throw. Even though the ball isn't flying through the air the dog may nonetheless take off in chase. Presumably, the animal has learned to recognize throwing actions. This is interesting because dogs can't throw and so can't have throwing mirror neurons. This is also interesting because somehow the action observation, throwing, is triggering an action execution, chasing, in the dog. This tells us that and action observation-execution sensory-motor circuit exists in the animal. There may even be "chase" cells in the dog's motor cortex that fire both during action observation and action execution. This is the same sort of circuit by which the pairing of a tone (cf., the throwing action) with an airpuff to the eye (cf., the ball flying) can eventually lead to an eyeblink response (cf., chasing) just to the presence of the tone (cf., the throwing action).
But dogs are smart. Try fake throwing without the ball, e.g., while the ball is still in the dog's mouth. You don't get much of a response (if your dog responds anyway, retry the exercise with a larger ball such that the dog can easily see whether or not you have something in your hand). This is interesting because it is kind of like pantomime and mirror neurons don't respond to pantomime. You can imagine how our non-mirror action observation-execution circuit might start to behave like a mirror neuron.
The point here is that it is not hard to imagine sensory-motor circuits that take observed actions as input and use these actions as triggers for any number of executed actions via regular old sensory-motor association. The cells underlying these circuits would probably behave like canonical neurons responding both to the execution and observation of the (non-mirror) actions.
That's all fine but how could the observation of an animal reaching for a piece of food trigger a similar action in the observer? In other words why would some actions trigger mirror actions? Here's where I need the help of any readers who know primate behavior (I certainly don't). But again reasoning from dog behavior, I've noticed that if you place a ball or toy in front of a moderately well-trained dog, it may watch you and wait (an untrained dog might just grab it). If you start to reach for the object the dog may suddenly lunge for it, trying to beat you to it. The dog isn't imitating you (in fact it can't), it learned to recognize that your reaching results in ball-possession and this triggers an action with a competitive goal. I can imagine this happening naturally with food for example, where one animal's movement towards a piece of meat triggers a competitive counteraction on the part of the dog. The trigger action could be the reach of a human, a forward movement of another dog, or even the looming flight of a bird or the mechanical action of a tool. Importantly, recognition of these actions is not being carried out in the motor system of the dog via motor simulation (at least not for the human, bird, and tool actions).
Presumably monkeys can also learn to recognize actions and respond with appropriate actions themselves. Observing an aggressive posture might trigger a flee or hit action. Observing a grasp toward a piece of fruit might trigger a competitive "mirror grasp" for the same piece of fruit. Maybe watching an experimenter reaching for a raisin that the monkey really wants triggers exactly this kind of competitive motor response and maybe this is what mirror neurons really reflect. Or maybe it is just another wrong theory about mirror neuron function.
Hickok, G. (2009). Eight Problems for the Mirror Neuron Theory of Action Understanding in Monkeys and Humans Journal of Cognitive Neuroscience, 21 (7), 1229-1243 DOI: 10.1162/jocn.2009.21189