Artificial intelligence(AI) agents are all about acting “rational” and maximizing “utility”. Not surprisingly, AI theory borrows a lot of principles from economics and specifically game theory which are centered around maximizing utility value. However, utility (or market value) optimization is hardly enough to replace human interaction in many environments. As humans, our behavior and decisions are not only dictated by the benefits of its market value but also about their social impact. Let’s call that metric social value.
Social norms typically involve the friendly requests that people make one of another. Could you help me with the homework? I am paying for dinner tonight. According to psychologists, we live in parallel worlds: one regulated by transactional value and another one governed by social value. While AI has made great inroads in models that maximize market value, there is still a lot of work to be done in terms of social value.
Sex and Social Value
To understand the difference between market and social value (and to spice things up a bit ;) ) , let’s take sex as an example. From a social perspective, sex value is related to the emotional connection that creates between the parties involved. However, there is also another context in which sex has a market value and cost money. Typically, bad things happen when somebody tries to optimize both the social and market value of sex simultaneously. That’s why prostitutes are not necessarily always looking for eternal love (contrary to what some customers might believe ;) ) or why husbands and wives don’t expect money in return for intimacy.
Gifts and Social Value
Gifts are another example of social value. Many gifts don’t bring any intrinsic relevant monetary value but they produce a direct emotional value. Gifts are an important element of human interaction because they are a good example of social norms and the parties involve receive immediate emotional gratification from it. In that context, gifts are a direct producer of social value.
AI and Social Value
AI algorithms are great optimizers of market value but they struggle when comes to creating social value. However, I think this is slowly starting to change. Can we create AI algorithms that operates in a social context and efficiently maximize social value. I believe we can. Let’s imagine a scenario in which multiple AI agents collaborate and compete against each other in a social context. In those environments, maximizing social value could be an important asset to AI agents.
In my opinion, social value in AI can be other in two fundamental ways: learning social norms and maximizing social utility. There is a good change that well-established algorithms such as inductive learning can be used to learn social norms in the way they learn rules in large datasets. In the context of social value, there are some AI models that can be adapted to obtain and maximize social utility. Techniques such as sentiment analysis can be used to evaluate and even quantify the reaction to social actions and they can be effectively used to maximize social value.