The Evolving Taxonomy of Moral Agency

The Architecture of Agency: A Comparative Report on Moral Subjectivity and Responsibility

1. Introduction: The Evolving Taxonomy of Moral Agency

As autonomous systems, complex corporate structures, and new insights into animal ethology challenge our traditional ethical boundaries, the strategic importance of defining “moral agency” has never been greater. For centuries, the domain of moral participation was reserved almost exclusively for the human rational agent. However, as non-human entities increasingly occupy roles of significant social and ethical consequence—ranging from AI-driven healthcare diagnostics to the legal personhood of corporations—we must move beyond a species-centric model. To maintain a functional moral community, we must examine the specific functional requirements that allow an entity to participate in moral life, regardless of its biological or structural origin.The “Orthodox View” of moral agency traditionally rests on three pillars: the capacity to distinguish right from wrong, the ability to make autonomous choices, and the liability to be held accountable for those actions. Under this framework, an agent is not merely an object of concern (a moral patient) but a participant capable of being praised or blamed. As we shall see, this benchmark remains the foundation upon which all other comparative frameworks are built, though the boundaries are increasingly porous.

2. The Benchmark: Human Rational Agents and the Reflection Condition

Human agency serves as the traditional “gold standard” in ethics due to its perceived alignment between internal thought and external action. Understanding its hierarchical structure is vital for identifying where other entities, such as animals or artificial systems, diverge from the human model.The hallmark of human agency is the “Reflection Condition,” rooted in Kantian and Aristotelian thought. This condition suggests that to be moral, an agent must be able to reflect upon their motivations and evaluate them according to abstract principles. Harry Frankfurt’s influential hierarchical model further clarifies this by distinguishing between different levels of desire:

• First-Order Desires:  Immediate, base motivations or impulses (e.g., the desire to eat or the impulse to act).
• Second-Order Volitions:  A meta-capacity not just to have desires about desires, but to care about which first-order desire moves one to action. A “person” is defined by the capacity to want a specific desire to be their will.
• The “Miracle-of-the-Meta”:  Mark Rowlands critiques the assumption that reflection automatically grants control. He argues that the same questions of control arising at the level of base motivation are replicated at the level of evaluation.This hierarchical model faces significant philosophical hurdles. The  “Problem of Origins”  suggests that a second-order endorsement might itself be conditioned by external or biological factors, undermining true autonomy. Furthermore, the  “Infinite Regress”  problem arises: if a second-order desire requires a third-order desire for validation, and so on, agency risks dissolving into an endless chain. Frankfurt’s proposed solution— “Satisfaction” —is often critiqued as a state of “volitional passivity” or resignation, where the agent simply ceases to seek further validation, rather than exercising robust active control.Furthermore, we must evaluate the “Authenticity” and “Competency” conditions. While Authenticity requires identifying with one’s commitments, it can be “too strong” or even self-deceptive. As Peeters and Velleman suggest, what we call authenticity is often merely an “ideal self-image” or a “public persona” (a Jungian mask) rather than a manifestation of the true self. This reflective agent serves as the point of departure for Rowlands’ “moral subjects” in the animal kingdom.

3. Moral Subjects: Rowlands’ Animal Ethology

A strategic shift is occurring in ethology: moving from viewing animals as mere “moral patients” (objects of concern) to “moral subjects.” Moral subjects are beings that act for moral reasons but lack the cognitive machinery for reflective responsibility. This shift bypasses the need for intellectual reflection by focusing on “moral emotions.”Rowlands argues that emotions like empathy, compassion, or grief track “evaluative propositions” (e.g., “that being is in distress”) without requiring the subject to abstractly entertain the proposition. Normativity here is grounded in a “Wittgensteinian vein”—as participation in social practices. Morality is a matter of navigating a shared world through learned patterns of response, rather than internal meta-cognition.

Comparative Analysis: Moral Agent vs. Moral Subject

Feature,Moral Agent (Human),Moral Subject (Animal)

Primary Motivation,Reflective principles; second-order volitions.,”Moral emotions (empathy, sympathy, grief).”

Reflection,”Required; “”Reflection Condition”” via meta-cognition.”,Not required; tracks moral content via direct emotion.

Responsibility,Full; held accountable via praise/blame.,None; acts for moral reasons without responsibility.

Normative Force,Derived from reflective self-evaluation.,Derived from participation in social practices.

Rowlands’ argument rests on three pillars:

1. Evaluative Tracking:  Emotions track true moral content.
2. Moral Reasons:  Acting on a moral emotion is functionally acting for a moral reason.
3. Subjecthood vs. Agenthood:  One can act for reasons without being an agent who is held responsible.Crucially, the  “Understanding Condition”  distinguishes the two. In folk morality, we do not blame those who do not understand what they are doing. However, as critics note, this is a “folk intuition” that is often stipulated rather than justified. If we cannot justify why understanding is the prerequisite for responsibility, the line between animal subjects and human agents may be one of degree rather than kind.

4. Collective Actors: Corporate Moral Agency and CID Structures

While animals represent biological subjects, corporations represent a structural, abstract agency. The strategic necessity of corporate agency is driven by the requirements of legal liability and the “Stakeholder vs. Shareholder” debate.

CID Analysis vs. Legal Fiction

Peter French argues that “Corporate Internal Decision” (CID) structures—comprising organization charts and “rules of recognition” (bylaws)—create corporate intentions distinct from individuals. Conversely, Manuel Velasquez views the corporation as a “Legal Fiction,” where agency is merely a design of natural persons.

The Responsibility Paradox and Jurisprudential Weight

This creates a  Responsibility Paradox : a corporation can be found liable for an outcome that no specific natural person within the firm intended or caused. This has profound jurisprudential consequences, dictating two competing models:

• Firm-State (Stakeholder View):  The corporation is a polity where stakeholders have voice rights, similar to citizens.
• Firm-Contract (Shareholder View):  The corporation is a “nexus-of-contracts,” where the primary duty is the fiduciary care of equity owners.From a strategic standpoint, the CID view allows for the attribution of corporate “guilt,” whereas the Legal Fiction view insists that responsibility must always be traced back to a human “natural person.”

5. The Technological Frontier: Autonomous AI and the Responsibility Gap

As learning automata enter high-stakes domains like healthcare and warfare, we face the emergence of “Artificial Moral Agents” (AMAs) and a resulting  “Responsibility Gap.”  Traditional concepts of responsibility fail when machines move beyond human oversight through emergent learning.

Being vs. Taking Responsibility

A critical distinction exists between “being responsible” (causal or legal liability) and “taking responsibility.” In many strategic contexts, such as medical errors, humans “take the blame” for machine failures to maintain social trust. Tigard proposes a “workable notion” of artificial responsibility based on “Social Responsiveness”—a bottom-up approach that integrates machines into the moral community based on their interactions.

Requirements for Social Responsiveness

To ground artificial responsibility, an AMA must satisfy several interactive requirements:

1. Recognition of Human Sentiment:  The ability to track and respond to human emotional states and normative expectations.
2. Acknowledgment of Error:  The capacity to “admit” fallibility, allowing for a functional social response.
3. Interactive Reciprocity:  Features like the “pretty please” function in digital assistants, which facilitate polite, reciprocal engagement.
4. Moral Community Participation:  Integration into the “Reactive Attitudes” (Strawson) of humans, where our practices of holding the machine responsible (even if it cannot feel guilt) stabilize our moral ecology.This contrasts a “Consequentialist” view (reprogramming as “punishment”) with a social-relational view where the machine’s “responsiveness” allows it to function as a surrogate agent.

6. Synthesis: Comparative Architecture of Responsibility

A unified view is essential for future policy. Understanding these boundaries prevents the “abdication of human responsibility” by ensuring we do not misattribute agency while still closing “retribution gaps.”

Comparative Matrix of Moral Participation

Entity Type,Internal Decision Structure,Volitional Capacity,Accountability Mechanism,Key Representative Concept

Humans,Reflection & Meta-cognition,Second-order Volitions,Moral Blame/Praise & Legal,Authenticity & Autonomy

Animals,Moral Emotions,First-order Desires,None (Moral Patients),Evaluative Propositions

Corporations,CID Structures (Bylaws),Rules of Recognition,Legal Liability / Fines,Collective Actor

Autonomous AI,Algorithms & Learning Automata,Programmed / Emergent,Consequentialist / Functional,Social Responsiveness

The “So What?” Layer: Strategic Implications

The “Responsibility Gap” is the common thread across all four categories. Whether it is a corporation hiding behind a CID structure, an animal acting on “unthinking” empathy, or a “black box” AI, the risk remains that humans will use these complexities to avoid the hard work of moral accounting. By defining these architectures, we recognize that while humans alone may be capable of “taking responsibility” through reflective understanding, other entities function as vital participants in our moral life. The distinction between these types is increasingly a matter of degree in functional participation, rather than a fundamental difference in kind.