Yeah, this is the real question. I'm sure I could program an AI agent to respond to (dis-)incentives in a fuzzy way. Is it consciously experiencing pain and pleasure at that point? The answer would be "no" unless we could prove it.
But that isn't the question. The question it possible that a sufficiently complex program could ever be able to have the experiences in the same way that humans have them (AKA "qualia").
Present and future anger, present and future pain, predicting whether future social cohesion will be maintained based on how a response is made to someone within a social graph. Understanding and assuming that nodes (other people) within a social graph can communicate to each other to update the state of a social interaction that they weren't present for.
I think it is possible.
If pleasure was just a variable on a scale from 0 to 10, and not its own co-processor, and an assumption that some forms of pleasure can influence parts of the social graph, it maybe isn't even that complicated anymore.
I think it would become less distinguishable from consciousness.