Can intelligent agents have personalities and emotions?

Question

Can intelligent agents (and chatbots) have personalities and emotions, given a properly defined ontology?

Answer 1

It can depend on how you define "emotion" and "personality":

1. Personality

Wikipedia defines personality as "the characteristic set of behaviors, cognitions, and emotional patterns that evolve from biological and environmental factors." (This is a serviceable definition we can use.)

Since intelligent agents make decisions (take actions), it is safe to say they have a set of behaviors. Cognition is easy here as well--regardless of the structure of the algorithm, we can say it utilizes a logical process that serves the function of cognition.

Emotion is more difficult.

2. Emotions

How exactly human emotions work is still partly obscure, and it seems to be a safe bet that chemicals are involved. But, in a reductionist sense, going back to Ancient Greek Philosophy, we can use a binary "love/hate" (attraction/repulsion) and apply it generally, where these feelings are the prime motivators of natural forces:

And Hesiod says, “First of all things was Chaos made, and then/Broad-bosomed Earth . . . and Love, the foremost of immortal beings,” thus implying that there must be in the world some cause to move things and combine them.

_{Aristotle (Metaphysics 1.984b)}

The real issue comes with interpretations of emotion that involve consciousness, such as on Wikipedia, but even the wiki immediately issues the disclaimer: "Scientific discourse has drifted to other meanings and there is no consensus on a definition," so it may be more useful to look to definitions rooted in cognition and the presumed utility of emotions from an evolutionary standpoint.

If reinforcement learning is being used, an automata can be said to like/dislike (love/hate) certain outcomes, which can be expressed via methods such as reward functions.

Certainly this is not emotion as we humans understand it, but it can be seen as an analog. (Regarding humans chemical response to entities/objects and events/outcomes as reinforce future decisions.)

ANSWER:

In no way do contemporary bots and agents have personalities and emotions in the sense of humans--we are probably still very from this and don't know if it's even possible. It may never be possible validate artificial consciousness, regardless of how conscious a system appears to be.

But, depending on how we define these terms "personality" and "emotion", we can choose to regard automata as having some reduced form of analogous qualities and functions.

Example: If I engineer a game AI to have certain tendencies and behaviors (high-risk-tolerance, partisan, aggressive) that are distinct from other agents (low-risk-tolerance, cooperative, defensive), such agents are commonly referred to as having personalities.

Example: We actually do use terms like "goodness/badness" and informally refer to the automata as "liking" or "not liking" nodes in a graph. This is a function of anthropomorphization. (Whether you consider these more than analogies depends on the definitions of feelings, but, generally, there needs to be something to feel. So you also need to reduce consciousness/awareness to the most basic requirement of simply "receiving of input".)