The new frontier of brain science
We know today loosely that the neocortex is responsible for processing and relating sensory input (between adjacent senses) we know that "adjacent" is more than an abstract categorization of their hierarchy in the mind but is quite literal, when flattened out the surface consists of patches of cells devoted to processing various types of sensory input in several layers (5 or 6)...where their edged touch, synergistic processing can occur. New research is pointing out that these synergies may explain synesthesia but for the purpose of this discussion the most relevant discovery is that how we remember things involves a dance between the neocortical processing, the retrieval of long term memories for the associated sense and comparison of incoming sensation to stored experience and tied together with a coupling to an emotional import factor. Researchers that have illuminated these areas with their work include the neuroscientist V.S. Ramachandran. As an engineer I've always found brain science fascinating but the mysteries have found their most powerful tool in the fMRI imaging that is rapidly illuminating the ways in which information is processed in the brain and determining which regions are highly localized for specific processing activities experienced by a subject under testing. Many of the models are done using rats but it is safe to say given the similarity in our brains (which are different only in scale)will indicate similar mechanisms in humans.
The mechanism thus discovered
When analyzing the of emotional experience for the proper development of a primate mind we must accept that an experienced emotional event has a different effect on how the associated memory is recalled but there does not seem (at least from the current data) any adverse response to methods of learning that bypass the use of this emotional mechanism. The current evidence suggests that Emotion is tied in upon retrieval, the centers of this integration have been localized to the twin Amygdalae deep within the brain, the integration of the memory and the emotion happens apparently in the hippocampus. This method is also the one that from an engineering perspective is most efficient toward allowing emotion to be arbitrarily tied to experience, this could be accomplished by having retrieved memories stored with an "emotion" bit, a signal that triggers the amygdala to provide a response when incoming sensory data matches stored memory that is correlated with it. The emotion bit would trigger the required response (fear, joy, worry...etc.) from the amygdala as it is sent to the cortical mind for mediation. Thus the entire experience set of sensory data can be easily multiplexed with an emotional component upon retrieval eliminating the need for storing emotion with every memory bit (which would be inefficient) The study I refer to above was able to show that emotion could be subtracted from the memory recall using a specific timing in when a stimulus is provided. In short the emotional import tied to a given memory could be untied at will which support the "integration" theory. However, if emotion can be freely tied and untied it would seem to serve a positive purpose only when **higher cortical analysis is not present**.
We have to remember that the brain was built up over evolutionary time, we can see this evolution of the brain today in living terrestrial organisms. Some, like lizards have barely any cortical mass and have varied sizes in the underlying brain regions which we know now govern autonomic response and timings for the body and as well govern emotion. I am going to posit my theory here and this may in fact be incorrect but it is a hypothesis I put out for discussion. Prior to our building a neocortex where complex reasoning could be performed on acquired multi-sensory data in a cognitive space. The different memories retrieved from the pallet of sensory experience that painted them into our minds, coupled with the emotional contribution that may have been associated with those experiences was the only way to judge moment to moment sensory input, judge by emotion essentially. We have moved on, the neocortex serves as a proxy to this "judge by emotion" ancestral mechanism that is in use by the lower animals (just observe a Lion in action to be convinced of this) our neocortex allows us to presage the future based on projection and our abstract reasoning allows us to imagine an un-experienced pain without having to go through the ordeal...thus, I would assert that there is no longer a unique benefit for the process of learning to have an emotional component because we now have a neo cortical mind in which to navigate the probability space of possibilities without having to experience them. Emotional import serves as a marker to our memory that says "pay a little more attention to this" that existed before we could hold the abstract possibilities in the scratch board of our large fore-brain. We now can chose to either teach by full experience (let them get hurt to know what it means) or we can use this large fore-brain object handling by ensuring only a minimum of hurt is experienced (even if it is in a single sensory area we can mentally map it to others!..cool) and then using the cognitive mind to explain why new experiences in seemingly disparate areas may result in analogous emotional pains. In the context of human parenting of children it is obvious that different levels of effort on the part of the parents are required but I think this approach serves to build the critical reasoning that is pivotal to creating cognitively dynamic children as opposed to children that learn on reaction to new experience. It also avoid the dangers of learning by emotional import as all new experiences do not require personal experience to understand why they may be dangerous or painful.
What this means for AI
The current efforts to develop artificial intelligence I think will benefit from this new understanding. If emotion is critical to developing a sense of significance for various actions then it will be important to simulate it in the any created artificial intelligences. Such intelligences must be able to have the components of internal emotion as well as external neo cortical processing. Some researchers are taking approaches toward creating artificial minds that model only one or another aspect of the mind and I think these are incorrect and potentially dangerous routes. We know from watching how people change when they experience various brain injury that the personality of a person is fluid with how the cortical processing is modulated through the filter of low level emotional experience, when these experiences are obfuscated by injury changes can occur that significantly modify the persons essence. If we are to create AI we want to make sure that they are 'stable' cognitively in ways that will make them adjusted to the experience of life that we will be creating for them. An AI that acts purely on assessments based on processing inputs and outputs may deem it necessary to kill thousands to achieve some goal without an emotional compass to guide why such an act would be wrong in the context of our society. At the least, a sense of self and a desire for self preservation coupled with understanding that same desire in others should be a requirement of any advanced AI. Currently one promiment effort seeks to model an entire brain (including the lower brain) called the "blue brain" project, if done properly this effort may yield a stable AI. Another effort headed by Jeff Hawkins creator of the Palm Pilot also seeks to create advanced AI but plans to model only the neo cortex, I think this latter effort may succeed in achieving the singularity (the name given to the point where the first conscious AI is realized) but the consciousness may be devoid of the ability to gauge or process emotion or the importance of empathy another key emotion to ensuring human agents treat one another equitably even when calculation would dictate more base actions as being more personally beneficial.
(Interesting because it was revealed before the latest study that showed how to erase a connection to emotion. My assertion is that both cognitive and "implicit" memory are tied to the amygdala which serves as the factory that produces simply the response.)