There's an interesting discussion on the LinkedIn AI Research forum surrounding a 2005 PhD paper by Afshin Ganjoo, "Designing Emotion-Capable Robots, One Emotion at a Time" (PDF format). Afshin is hoping generate some interest among robot builder in trying to actually implement his model. While there was general agreement that true intelligence is impossible without emotion, there were some differing ideas on how to get there. Anne van Rossum, a researcher at Almende B.V., suggested an evolutionary neuroconstructivist approach to emotion is necessary. The argument being that humans could not select appropriate categories of emotion whereas an evolutionary approach might result in more natural emotions. Sindhu Joseph, a researcher at IIIA-CSIC, offered a third option, that we first needed robots with a level of cognition that allowed them to evaluate past experience and form beliefs, which would lead to the natural evolution of emotions. In response, Afshin makes this observation:
When we design a robot with two legs, or six legs, we are borrowing concepts from nature that has already evolved. When we give robots sensory devices such as cameras and heat sensors, we are borrowing from creatures which have already evolved. Even the hard supporters of producing intelligence out of neural nets like the ones we humans developed, and allow those neural nets to evolve or learn their own connections and structures, we are borrowing intelligence structures from things which evolved outside the realm of robotics. So, it is not so much “define vs. evolve” but rather “at what point do you want to break away from what has already evolved, and allow your creation to start evolving on its own.”
If you're a robot builder interested in implementing Afshin's ideas for emotion-capable robots, you can contact him through LinkedIn.