Agentic Self‑Determination Theory Extensions

Self‑determination, agency, and the buzzing of a hive may seem worlds apart. Yet the same psychological forces that drive a child to explore a new playground also guide a worker bee to a flower and a learning algorithm to a novel solution. This article weaves together the latest empirical work on agency with Deci & Ryan’s classic framework, offering a richer, “agentic” version of Self‑Determination Theory (SDT) that speaks to humans, insects, and artificial agents alike.

Self‑determination theory has been a cornerstone of motivation research for more than three decades. Its three universal psychological needs—autonomy, competence, and relatedness—have explained why people thrive in schools, workplaces, and sports teams. However, a growing body of research shows that the experience of agency—the sense that one is the originator of one’s actions—exerts a distinct influence on motivation, learning, and well‑being. Recent neuroimaging studies, large‑scale field experiments, and computational models suggest that agency is not just a by‑product of autonomy; it is a fourth, dynamic pillar that can amplify or dampen the classic needs.

Why does this matter for bee conservation and self‑governing AI? Bees exemplify collective agency: a single worker’s foraging decision emerges from, and feeds back into, the colony’s resource map, affecting pollination networks that sustain ecosystems and agriculture. Likewise, AI agents that can self‑determine their goals—while respecting human values—are central to safe, beneficial artificial general intelligence. By extending SDT to include agency, we gain a unified language for designing human‑centered policies, bee‑friendly habitats, and aligned AI systems.

Below we map the evolution from classic SDT to an Agentic Self‑Determination Theory (ASDT), grounding each step in concrete data, real‑world examples, and mechanisms that bridge biology, psychology, and technology.

1. Foundations of Self‑Determination Theory

Self‑Determination Theory was first articulated by Edward L. Deci and Richard M. Ryan in the early 1980s. Its core claim is that intrinsic motivation flourishes when three basic psychological needs are satisfied:

Meta‑analyses of over 200,000 participants across 128 studies confirm that higher need satisfaction predicts 30‑40 % greater well‑being, 20‑25 % higher academic performance, and 15‑20 % lower burnout (Ryan & Deci, 2020). The theory has been applied in education (e.g., self‑directed learning), health (e.g., adherence to exercise), and organizational design (e.g., job crafting).

Despite its breadth, classic SDT treats agency as an implicit outcome of autonomy. Autonomy is measured by items such as “I feel free to decide for myself,” but the subjective sense of being the originator of an action—the first‑person experience of agency—receives far less systematic attention. This omission becomes evident when we examine contexts where people choose but still feel controlled (e.g., following a strict schedule) versus contexts where they act without explicit choice but experience a strong sense of agency (e.g., improvisational jazz).

2. The Rise of Agency Research

2.1 Defining Agency

In cognitive psychology, agency refers to the judgment that one’s own intentions caused a particular outcome. It is measured with paradigms such as the Intentional Binding task, where participants perceive the interval between a voluntary action and its effect as shorter than when the action is forced (Haggard, Clark, & Kalogeras, 2002). The binding effect averages ≈ 30 ms for self‑initiated actions versus ≈ 0 ms for externally triggered ones, a robust physiological marker of agency.

2.2 Neural Correlates

Functional MRI studies consistently highlight the pre‑supplementary motor area (pre‑SMA) and inferior parietal lobule (IPL) as hubs for agency detection. A 2021 meta‑analysis of 45 neuroimaging experiments reported Cohen’s d = 0.78 for activation differences when participants experienced high versus low agency. Moreover, dopaminergic signaling in the ventral striatum correlates with agency‑related reward prediction errors, linking agency to reinforcement learning mechanisms (Schultz, 2016).

2.3 Empirical Findings on Motivation

A series of field experiments in 2022–2024 examined agency in workplace settings. In one study of 4,500 employees across three multinational firms, a simple redesign that gave workers “action‑choice prompts” (e.g., “You may choose to start this task now or later”) increased self‑reported agency scores by 22 % and led to a 12 % rise in quarterly productivity (Klein et al., 2023). Importantly, these gains persisted after controlling for autonomy, competence, and relatedness, suggesting agency adds unique predictive power.

2.4 Agency in Non‑Human Systems

Research on social insects shows that individual agents can exhibit a form of agency despite lacking consciousness. Honeybees (Apis mellifera) perform waggle dances that encode the distance and direction to resources. When a forager discovers a novel, high‑quality nectar source, it initiates a dance without external cue, altering the colony’s foraging allocation. Experiments using RFID tagging of > 10,000 bees demonstrated that individual agency events (first discovery dances) predict colony‑level foraging efficiency with a correlation of r = 0.61 (Seeley & Seeley, 2022).

3. Agentic Self‑Determination Theory (ASDT)

Building on the evidence, Agentic Self‑Determination Theory (ASDT) proposes a fourth, interacting need: Agency. ASDT retains the original triad but reconceptualizes autonomy as “volitional freedom” (the availability of choice) while agency captures “originative control” (the experience of being the cause). The model can be visualized as a tetrahedron where each vertex influences the others (see Figure 1 in the original PDF).

3.1 Core Propositions

1. Agency is a distinct psychological need that, when satisfied, enhances the quality of autonomy.
2. Agency interacts synergistically with competence: feeling capable amplifies the sense that one’s actions are self‑generated.
3. Relatedness can modulate agency; supportive social contexts increase the likelihood that individuals attribute outcomes to themselves rather than to external pressure.
4. Unmet agency can lead to “agency deficit” states—feelings of helplessness, learned helplessness, and reduced intrinsic motivation, even when autonomy is high.

3.2 Operationalizing Agency

ASDT recommends three measurable dimensions:

These items have been validated in a cross‑cultural sample of 12,342 participants across 15 nations (Zhou et al., 2024), yielding Cronbach’s α = 0.91 for the composite agency scale.

3.3 Theoretical Integration

ASDT can be expressed mathematically as a multiplicative function of need satisfaction:

\[ \text{Motivation}{\text{ASDT}} = (A{\text{auto}} \times C \times R) \times (1 + \beta \cdot \text{Agency}) \]

where \(A_{\text{auto}}\) = autonomy score, \(C\) = competence, \(R\) = relatedness, and \(\beta\) ≈ 0.45 (derived from regression analyses of 8,000 participants). This formulation captures the boosting effect of agency: a 10‑point increase in agency raises overall motivation by ≈ 4.5 %, even when other needs are held constant.

4. Empirical Evidence: Human Studies

4.1 Education

A longitudinal study in Finnish secondary schools (N = 2,108) introduced “student‑led project modules” that gave learners full agency over topic selection, methodology, and presentation format. Over two years, agency scores rose from 3.2 to 5.8 (on a 7‑point scale), while standardized math scores increased by 7.4 % relative to control classes (Kivinen & Salmi, 2023). Mediation analysis showed agency accounted for 38 % of the effect of the intervention on academic achievement.

4.2 Health & Well‑Being

In a randomized trial of 4,200 adults with type‑2 diabetes, participants receiving a self‑agency coaching protocol (goal‑setting, self‑monitoring, and reflective attribution training) demonstrated a 15 % greater reduction in HbA1c after 12 months compared with standard care (p < 0.001). Agency scores mediated 46 % of the improvement, highlighting the clinical relevance of agency beyond mere autonomy.

4.3 Workplace

Beyond the Klein et al. (2023) study mentioned earlier, a meta‑analysis of 27 organizational interventions (total N = 68,000) found that agency‑focused designs (e.g., “action‑choice dashboards”) produced an average effect size d = 0.63 for employee engagement, surpassing autonomy‑only interventions (d ≈ 0.38).

4.4 Cross‑Cultural Robustness

A comparative survey across East Asian, European, and Sub‑Saharan African contexts revealed that agency need strength varies modestly (mean scores: 4.9, 5.2, 5.0 respectively) but its predictive weight for life satisfaction is consistently high (β ≈ 0.41). This suggests agency is a universal driver, not limited to individualistic cultures.

5. Extending to Collective and Ecological Agents

5.1 Bees as a Model of Distributed Agency

Honeybee colonies function as superorganisms where individual foragers make autonomous decisions that are integrated into a colony‑level foraging map. Recent work using computer‑vision tracking of 12,000 bees in a controlled meadow demonstrated that first‑discoverer bees (≈ 5 % of foragers) trigger collective recruitment cascades that increase total nectar influx by ≈ 42 % (Schmidt et al., 2023).

From an ASDT perspective, each bee satisfies a micro‑agency need (initiating a dance) while also fulfilling collective competence (accurate distance encoding) and relatedness (social feedback from nest‑mates). The colony’s “collective agency” emerges when enough individuals experience successful agency events, leading to adaptive resilience during resource scarcity.

5.2 Ecosystem‑Level Agency

Pollination networks can be viewed as agency‑mediated flows: individual pollinators’ choices shape plant reproductive success, which in turn influences habitat quality. A 2022 meta‑analysis of 1,134 pollinator‑plant interaction studies found that higher pollinator agency (measured as the proportion of foraging bouts initiated without prior cue) correlated with greater plant species richness (r = 0.53). This suggests that fostering agency in pollinators—through habitat heterogeneity and floral diversity—can amplify ecosystem services.

6. AI Agents and Self‑Governance

6.1 Agency in Reinforcement Learning

In reinforcement learning (RL), an agent’s policy determines actions based on state information. Traditional RL maximizes expected reward but does not differentiate between externally imposed versus self‑generated goals. Recent work on Intrinsic Motivation Modules (e.g., curiosity‑driven exploration) introduces a self‑determination component: the agent chooses which internal prediction errors to minimize, effectively granting it a computational sense of agency.

A benchmark on the OpenAI ProcGen suite (500 levels) showed that agents equipped with an Agency‑Augmented Reward (AAR) term achieved 23 % higher sample efficiency and 15 % higher final scores than baseline agents (Wang & Lee, 2024). The AAR term is mathematically analogous to the ASDT multiplier, reinforcing actions that the agent originates rather than merely reacts to.

6.2 Alignment and Ethical Considerations

From an AI safety standpoint, self‑governing agents must balance agency with human oversight. The concept of “controlled agency” proposes a hierarchical architecture where lower‑level modules enjoy high agency (exploration, skill acquisition) while higher‑level supervisory layers enforce value alignment. Empirical simulations with 10,000 autonomous drones in a disaster‑response scenario demonstrated that controlled‑agency drones reduced unintended collateral actions by 68 % compared with fully autonomous drones, without sacrificing mission completion time.

6.3 Cross‑Disciplinary Bridges

Both bees and RL agents share a feedback loop: actions modify the environment, which in turn reshapes future action possibilities. In bees, the waggle dance updates the colony’s foraging map; in RL, the policy update revises the action‑value function. Recognizing these parallels allows researchers to transfer insights—for instance, using stigmergic communication models from entomology to design scalable multi‑agent coordination protocols.

7. Cross‑Domain Insights: Bees, Humans, and Artificial Agents

7.1 Case Study: Restoring a Pollinator‑Depleted Farm

A collaborative project in southern Spain combined ASDT‑informed farmer training with bee‑friendly habitat design and autonomous pollination robots. Farmers received workshops that emphasized personal agency—they set their own planting schedules and evaluated outcomes through reflective logs. Simultaneously, wildflower strips increased bee agency events by 31 % (more first‑discoverer foragers). Finally, robotic pollinators equipped with AAR algorithms adjusted flight patterns based on self‑generated coverage goals, improving pollination rates by 18 % over a conventional fixed‑path system. After two seasons, crop yields rose 12 %, and farmer satisfaction scores (including agency) improved from 4.1 to 6.3 on a 7‑point scale.

7.2 Mechanistic Synthesis

1. Feedback Amplification – Agency creates a positive feedback loop: successful self‑initiated actions raise competence, which in turn strengthens agency.
2. Error‑Correction – In both bees and RL agents, prediction error signals trigger agency adjustments (e.g., a bee changes its dance when the nectar source declines).
3. Social Calibration – Relatedness provides a calibration signal: peer acknowledgment reinforces the attribution of outcomes to the self, boosting agency.

8. Practical Implications for Conservation and AI Design

8.1 Designing Agency‑Friendly Environments

Human contexts:

• Choice architecture that offers action‑choice prompts (e.g., “You may start this task now or later”) rather than simple permission.
• Reflective debriefs that ask participants to attribute successes to their own decisions, strengthening agency‑recognition.

Bee habitats:

• Floral heterogeneity (minimum 3 species per 100 m²) increases spontaneous foraging choices, thereby raising individual bee agency events.
• Nest‑site micro‑climate diversity encourages workers to self‑select optimal brood chambers, enhancing colony competence.

AI systems:

• Implement Agency‑Augmented Reward (AAR) terms that reward self‑initiated sub‑goals.
• Use stochastic policy regularization to prevent over‑constraining agents, preserving a sense of originative control.

8.2 Policy Recommendations

1. Education policy should incorporate agency‑assessment tools alongside autonomy measures in curriculum standards.
2. Agricultural subsidies could prioritize farms that adopt agency‑enhancing pollinator practices (e.g., diversified planting, minimal pesticide use).
3. AI governance frameworks (e.g., the OECD AI Principles) can add a “Self‑Determination” clause, mandating that high‑risk systems include mechanisms for controlled agency and transparent attribution of decisions.

9. Future Directions and Open Questions

Addressing these gaps will solidify ASDT as a unifying framework for motivation science, conservation biology, and artificial intelligence.

Why it matters

Agency is the spark that turns freedom into purposeful action. By recognizing agency as a distinct psychological need, we can design schools that ignite curiosity, farms that empower pollinators, and AI that navigates the world with self‑determined goals while staying aligned with human values. The extensions presented here are not abstract theory; they are actionable insights backed by data, ready to guide policymakers, conservationists, educators, and technologists toward a future where autonomous beings—whether human, bee, or algorithm—thrive together.