For an artificial agent to be perceived as a "social actor" rather than a mere tool, it must move beyond direct manipulation scripts. Early work in the ALIVE system by Maes et al. established that believable interaction emerges when agents possess internal motivations, needs, and sensory fatigue. By mimicking animal behavior—an approach rooted in ethology—agents can prioritize their own simulated states over user commands, transforming the user experience from controlling a puppet to interacting with a lifelike entity.
The Architecture of Intrinsic Drive
The transition from reactive scripts to autonomous behavior requires a framework that translates internal values into external actions. Chen et al. propose a three-module architecture to achieve this: an internal module for core values, a pursue module for generating goals, and a desire module for fine-tuning actions to meet urgent needs.
This value-driven approach is increasingly grounded in established psychological theory. For example, Lin et al. utilize Maslow’s Hierarchy and Social Value Orientation (SVO) to model complex dynamics like bullying and cooperation. By grounding Large Language Model (LLM) agents in these frameworks, researchers can create simulations where behavior is not just a response to a prompt, but an expression of a stable, underlying personality.
Affect as a Social Signal
If internal values provide the "why" behind an action, emotional expression provides the "how" for social communication. Clark Elliot argued that agents do not need biological complexity to be effective; they need a symbolic mapping of situations to emotional states. This "affective reasoning" allows agents to engage in social dynamics like empathy or negotiation.
Humans are naturally attuned to these signals. de Melo et al. found that humans are more likely to cooperate with agents perceived to have both Agency (the ability to plan) and Experience (the ability to feel). Crucially, these effects are additive: even an agent with limited rational capabilities can build trust by displaying appropriate affective cues.
The Necessity of Action Consistency
The mere presence of emotion is not enough to foster cooperation; in fact, it can be counterproductive. Becker et al. observed that in high-stakes scenarios, human-like emotional expressions can increase participant anxiety and decrease trust if the emotions introduce a sense of unpredictability.
The solution lies in action-consistency. As Antos et al. demonstrate, trust is driven by the alignment between an agent's emotional expression and its behavioral strategy. A "smiling hypocrite"—an agent that acts selfishly while displaying cooperative emotions—is perceived as less trustworthy than an "angry egoist" whose negative emotions match its selfish actions. Consistency makes the agent’s intentions predictable.
Gratch and Carnevale further show that when an agent's emotions are aligned with cooperative goals, humans perceive the agent as more trustworthy and considerate of their welfare. This alignment is what allows agents to function as effective proxies in complex ecosystems. For instance, Ma et al. found that emotionally grounded agents in Metaverse simulations achieve significantly higher task acceptance rates because their decisions feel "human-like" and explainable.
By integrating internal psychological needs with consistent external signaling, this research direction moves AI from being a predictable machine to a predictable social partner.
Key Foundational Work
ALIVE: Artificial Life Interactive Video Environment (1994) — contributed 2 concept elements to this direction.
Research Problems in the Use of a Shallow Artificial Intelligence Model of Personality and Emotion (1994) — contributed 1 concept elements to this direction.
Modeling Interactive Agents in ALIVE (1995) — contributed 1 concept elements to this direction.