References
Academic sources and theoretical foundations for ProEthica
9-Component Framework D = (R, P, O, S, Rs, A, E, Ca, Cs)
Foundational Works
- McLaren, B.M. (2003). Extensionally defining principles and cases in ethics: An AI model. Artificial Intelligence, 150(1-2), 145-181. DOI
- Berreby, F., Bourgne, G. & Ganascia, J.-G. (2017). A Declarative Modular Framework for Representing and Applying Ethical Principles. AAMAS 2017.
- Tolmeijer, S. et al. (2021). Implementations in Machine Ethics: A Survey. ACM Computing Surveys, 53(6), Article 132. DOI
Pass 1: Situational
- R Roles
- S States
- Rs Resources
Pass 2: Normative
- P Principles
- O Obligations
- Cs Constraints
- Ca Capabilities
Pass 3: Temporal
- A Actions
- E Events
Click each component below for detailed academic references and definitions.
Professional roles generate distinctive ethical obligations tied to professional goods. This is the foundational basis of ProEthica's approach: roles determine which obligations apply.
Primary Framework: Oakley & Cocking (2001)
Oakley, J. & Cocking, D. (2001)
Virtue Ethics and Professional Roles
Cambridge University Press. ISBN: 978-0-521-79305-6. DOI
Core Thesis: Good professions serve key human goods (health, justice, public safety). Professional roles generate distinctive obligations grounded in commitment to these goods.
Key Concepts:
- Regulative Ideals. Internalized conceptions guiding professional action (e.g., "public safety paramount" for engineers)
- Professional Virtues. Character traits for excellence: competence, trustworthiness, humility, justice
- Role-Generated Obligations. Duties arising from professional commitment, not general morality
- Founding Value Limits. Cannot violate the good the profession serves (e.g., engineer cannot approve unsafe design even if client demands it)
Why Foundational: Explains why ProEthica's 9-component framework begins with Roles (R) - professional roles determine which obligations apply and provide ethical authority. Engineering obligations differ from Medical obligations differ from Legal obligations.
Relational Role Categories (the role_category shown in extraction)
Each extracted role can carry a relational
category naming the duty-relationship it establishes (orthogonal to the
occupational kind of agent). The four categories are literature-grounded and map to
proethica-intermediate archetype classes under RelationalRole:
provider_client→ ProviderClientRole: service-delivery relationship with duties of competent service, confidentiality, and client welfare. Kong et al. (2020) DOIprofessional_peer→ ProfessionalPeerRole: collegial relationship with peer review, mentoring, and standard-maintenance obligations. Kong et al. (2020) DOIemployer_relationship→ EmployerRelationshipRole: organizational relationship balancing loyalty with professional independence. Wendel (2024) DOIpublic_responsibility→ PublicResponsibilityRole: societal obligation that can override client or employer interests. Thornton et al. (2017) DOI
The participant and
stakeholder categories are generic fallbacks (mapped to the neutral
RelationalRole), not part of the Kong identity-role framework.
Additional References
- Kong et al. (2020). Identity roles and virtues across engineering, law, accounting
- Doernberg & Truog (2023). Spheres of morality framework for role-based contexts
- Dennis et al. (2016). Roles as filters determining which obligations apply
- Cointe et al. (2016). Multi-agent frameworks for role-based principle interpretation
Abstract ethical principles require extensional definition through precedents to achieve concrete meaning.
Key Literature (Table 2.1)
- McLaren (2003). Extensional vs intensional principle definition
- Prem (2023). Abstract nature of principles - too vague for concrete designs without operationalization
- Taddeo et al. (2024). Constitutional interpretation - principles require balancing and specification in context
Additional References
- Hallamaa & Kalliokoski (2022). Context-sensitivity of principles
- Anderson & Anderson (2018). GenEth: learning principles from expert examples
- Benzmuller et al. (2020). LogiKEy higher-order logic formalization
Obligations specify what, when, by whom, and under what conditions: the "moral core" of professional ethics.
Key Literature (Table 2.1)
- Dennis et al. (2016). Specification requirements - obligations must be fully specified to be computationally verifiable
- Anderson & Anderson (2006, 2007, 2011). Duty quantification - tuple representation for degrees of satisfaction
- Almpani et al. (2023). Dynamic priorities - obligations can have changing priorities based on environmental conditions
Additional References
- Ganascia (2007). Answer set programming for conflicting obligations
- Dennis & del Olmo (2021). Defeasible deontic logic
Context determines which principles activate and how they apply; identical actions receive different moral evaluations.
Key Literature (Table 2.1)
- Jones (1991). Moral intensity - six characteristics that affect ethical salience (magnitude, consensus, probability, immediacy, proximity, concentration)
- Almpani et al. (2023). Event Calculus - temporal predicates for current conditions and historical information
- Berreby et al. (2017). Fluents - time-varying properties that affect ethical assessment
- Sarmiento et al. (2023). Causal chains - volitional actions produce events that initiate new states
Professional ethical knowledge is largely encoded in cases and codes rather than explicit rules.
Key Literature (Table 2.1)
- McLaren (2003). Precedents - professional ethical knowledge exists primarily in accumulated cases, not abstract rules
- Davis (1991) & Frankel (1989). Codes - hierarchical structure from aspirational principles to regulatory procedures
- Harris et al. (2018). Decision procedures - step-by-step techniques for analyzing ethical dilemmas through code application
- Anderson & Anderson (2018). GenEth learning - principles can be learned from expert examples using inductive logic programming
Additional References
- Ashley & McLaren (1995). Case-based reasoning foundations
- Guarini (2006). Neural network case classification
- Kong et al. (2020). Computational linguistics for code analysis
Professional actions require multi-criteria assessment across deontological, consequentialist, and virtue dimensions.
Key Literature (Table 2.1)
- Sarmiento et al. (2023). Volitional nature - systems must separate Actions from Events through attribution of responsibility
- Bonnemains et al. (2018). Multi-framework evaluation - identical physical actions require evaluation through different ethical frameworks
- Govindarajulu & Bringsjord (2017). Intentional status - same action carries different ethical weight based on agent's intention and knowledge
Ethical evaluation depends on event sequences; appropriate responses depend on historical context and anticipated future.
Key Literature (Table 2.1)
- Berreby et al. (2017). The Event Calculus distinction between agent-caused outcomes and exogenous / automatic occurrences (events that fire when preconditions hold without agent input). This is the basis of the event-type classification described below.
- Event Calculus formalism. Mathematical foundation for modeling temporal dynamics - events modify fluents and trigger state transitions (see the Fluents and OWL-Time entries below for how this is materialized)
- Arkin (2008). Emergency overrides - an emergency can temporarily override normal constraints reflecting prima facie duties
Additional References
- Anderson & Anderson (2018). Event sequence evaluation
- Govindarajulu & Bringsjord (2017). Temporal modal logic
Implementation note: classification vs. severity
An event carries two distinct labels. Event type
(outcome, a consequence of an agent action; exogenous, external
and not caused by a case agent; automatic_trigger, fires when preconditions
hold) is the literature-grounded Event Calculus classification above (Berreby et al.
2017), and it carries weight for responsibility attribution. Severity
(critical / high / medium / low / routine) is a deliberately heuristic triage indicator,
not a formal ontology category. Arkin's emergency-override is realized not as a
standalone event flag but through the fluent path: the event initiates a State
(for example an elevated public-safety risk) that in turn activates the overriding
obligation or constraint, so the override is a queryable consequence of the state
transition rather than a property asserted on the event.
Competence transcends technical skill: includes communication, knowledge, reasoning, values, and reflection.
Key Literature (Table 2.1)
- Narvaez & Rest (1995). Four Component Model - moral sensitivity, judgment, motivation, and character
- Tolmeijer et al. (2021). Four requirements - norm competence, situational awareness, learning/adaptation, explanation/justification
- Berreby et al. (2017). Action Model - represents physical and analytical competencies determining available actions
- Epstein & Hundert (2002). Domain-specific judgment - professional competence integrates communication, knowledge, reasoning, values
Additional References
- Cervantes et al. (2020). Capacities/capabilities/competencies taxonomy
- Stenseke (2022). Functional competence vs moral agency
- Dennis et al. (2016). Formal verification for capability sufficiency
Constraints establish inviolable limits (boundaries) while obligations specify requirements.
Key Literature (Table 2.1)
- Ganascia (2007). Defeasible logic - constraints as default rules that tolerate justified exceptions under extraordinary circumstances
- Dennis et al. (2016). Hierarchical management - constraints assigned ethical order for evaluation, selecting least unethical option when multiple compete
Additional References
- Taddeo et al. (2024). Legal/regulatory constraint interaction
- Arkin (2008). Architectural "ethical governors"
Transformation Classification
Source
Marchais-Roubelat, A. and Roubelat, F. (2015)
"Designing a moving strategic foresight approach: ontological and methodological issues of scenario design"
Foresight, Vol. 17 No. 6, pp. 545-555. DOI
Transformation Types
From Table II (p. 550): "Steering rule" evaluates process transformations to challenge scenario duration.
| Type | Definition (from paper) | Application to NSPE Cases |
|---|---|---|
| Transfer | "Shifts from a scenario set to a new one" | Resolution transfers obligation/responsibility to another party. The ethical burden moves from one stakeholder to another. |
| Stalemate | "Stakeholders cannot quit the scenario, as they seem to be trapped in the set of rules" | Competing obligations remain in tension without clear resolution. Both duties are valid but incompatible. |
| Oscillation | "Stakeholders go to and fro between different sets of rules" | Duties shift back and forth between parties over time. Responsibility cycles as circumstances change. |
| Phase Lag | "Some stakeholders do not follow the set of rules of the scenario where they are engaged, performing parallel scenarios" | Delayed consequences reveal obligations not initially apparent. Hidden defects or future harms emerge later. |
Context
Marchais-Roubelat and Roubelat's framework comes from strategic foresight methodology, specifically their "action-based scenarios" approach. ProEthica applies these transformation types to classify how NSPE Board of Ethical Review cases resolve competing ethical obligations. The classification helps identify patterns in ethical reasoning that can inform precedent discovery.
Precedent Discovery
ProEthica uses sentence embeddings and case-based reasoning to find relevant prior NSPE Board of Ethical Review decisions.
Foundational Text
Richter, M.M. & Weber, R.O. (2013)
Case-Based Reasoning: A Textbook
Springer. ISBN: 978-3-642-40166-4. SpringerLink
Textbook covering CBR theory: similarity measures, retrieval, and adaptation. Provides the foundation for experience-based problem solving applied in ProEthica.
Legal Case Retrieval
Sun, Z., Zhang, K., Yu, W., Wang, H. & Xu, J. (2024)
Logic Rules as Explanations for Legal Case Retrieval
Proceedings of LREC-COLING 2024, pp. 10747-10759. ACL Anthology
Neural-Symbolic Legal Case Retrieval (NS-LCR): combines embeddings with logic rules for explainable retrieval in legal contexts.
Wiratunga, N., Abeyratne, R., Jayawardena, L., et al. (2024)
CBR-RAG: Case-Based Reasoning for Retrieval Augmented Generation in LLMs for Legal Question Answering
Proceedings of ICCBR 2024, LNCS vol. 14775, pp. 445-461. Springer. DOI | arXiv
Integrates CBR with RAG for legal QA. Compares domain-specific vs general embeddings.
Professional Ethics Analysis Methods
ProEthica uses structured argument mining and practical professional ethics methodologies to compose decision points, compare cases to precedents, and generate entity-grounded arguments.
Toulmin Model of Argumentation
Toulmin, S.E. (1958, 2003)
The Uses of Argument
Cambridge University Press. Updated edition 2003. DOI
Foundational framework decomposing arguments into six components: Claim, Data/Grounds, Warrant, Backing, Qualifier, and Rebuttal. ProEthica maps extracted entities to these components for structured ethical argument analysis.
| Toulmin Component | ProEthica Entity | Role in Ethical Argument |
|---|---|---|
| Claim | Board Conclusions | The ethical determination |
| Data/Grounds | Actions, Events, States | What happened in the case |
| Warrant | Principles, Obligations | Professional duties justifying the claim |
| Backing | Code Provisions | Authority supporting the warrant |
| Qualifier | Capabilities, Constraints | Limits on the claim |
| Rebuttal | Conflicting Obligations | Counter-arguments |
Argument Validation Score
Oakley, J. & Cocking, D. (2001)
Virtue Ethics and Professional Roles
Cambridge University Press. DOI
ProEthica validates generated arguments using a three-tier test derived from role ethics theory. Each test uses a specific computational method:
| Test | Weight | Computation (from extracted entities) | Theoretical Basis |
|---|---|---|---|
| Entity Grounding | 40% |
Computed: validity_rate = valid_refs / total_refsEach argument component URI (warrant, backing, data) checked against extracted entity URIs from temporary_rdf_storage. Passes if warrant exists AND validity_rate ≥ 70%. |
Structural validity: arguments reference extracted facts |
| Founding Value | 40% |
Computed: Keyword search against domain violation dictionary Claim and warrant text searched for violation keywords (e.g., "endanger", "harm" for public_safety domain). Passes if no violation keywords found. |
Oakley & Cocking "regulative ideal": no action can violate founding good |
| Professional Virtue | 20% |
Computed: Virtue-to-Capability entity matching Required virtues (from warrant text triggers) matched against extracted Capability entities. Passes if any virtue present or has capability support. |
Oakley & Cocking: virtues require capabilities for enactment |
Data sources: All computations use entities extracted during Steps 1-4 (stored in temporary_rdf_storage).
Entity URIs, Capability labels, and argument text are all derived from extraction pipeline output.
Ethical Tensions
Definition: Ethical tensions document conflicts between obligations and constraints that create ethical difficulty. ProEthica identifies tensions during Stage 4 (Narrative Construction) and characterizes each through two complementary frameworks.
| Framework | Description | Source |
|---|---|---|
| Conflict Type Classification | Categorizes tension as Obligation vs Obligation or Obligation vs Constraint | Ganascia (2007); Dennis et al. (2016) |
| Moral Intensity Scoring | Assesses severity using six intensity dimensions | Jones (1991) (see below) |
Conflict Type Classification
Ganascia, J.-G. (2007)
Modelling Ethical Rules of Lying with Answer Set Programming
Ethics and Information Technology, Vol. 9, pp. 39-47. DOI
Formalizes conflicting obligations through defeasible logic, showing how constraints can operate as default rules that tolerate justified exceptions under extraordinary circumstances. These default rules with exceptions enable systems to satisfy a general prohibition while recommending an exception when adherence would violate other duties. ProEthica adopts this framework for Obligation vs Obligation tensions where two prima facie duties pull in opposite directions.
Dennis, L.A., Fisher, M., Slavkovik, M., & Webster, M. (2016)
Formal Verification of Ethical Choices in Autonomous Systems
Robotics and Autonomous Systems, Vol. 77, pp. 1-14. DOI
Establishes hierarchical constraint management that assigns constraints an ethical order for evaluation purposes. This structure allows systems to distinguish available choices by how badly they violate principles and to select the least unethical option when multiple constraints compete. ProEthica uses this for Obligation vs Constraint tensions where a professional duty conflicts with an inviolable boundary.
Application: Each extracted ethical tension displays its conflict type (Obligation vs Obligation or Obligation vs Constraint) alongside moral intensity scores. This dual characterization helps practitioners understand both what is in conflict and how serious the tension is.
Decision Point Salience / Moral Intensity
Jones, T.M. (1991)
Ethical Decision Making by Individuals in Organizations: An Issue-Contingent Model
Academy of Management Review, Vol. 16, No. 2, pp. 366-395. JSTOR | AOM
Provides actionable criteria for prioritizing which professional decisions deserve most attention. ProEthica uses these six intensity components to score decision point salience:
| Component | Definition | Professional Application |
|---|---|---|
| Magnitude | Impact on founding professional good (safety/health/justice) | Severity of Events (death > injury > property damage) |
| Social Consensus | Professional consensus via codified standards | Explicit Code provision citations |
| Probability | Likelihood of violating professional obligations | Professional foresight requirement |
| Temporal Immediacy | When consequences affect key goods | Professional duty timing (immediate vs. long-term) |
| Proximity | Role-relationship binding strength | Direct professional relationships (client, public) |
| Concentration | Scope of professional impact | Number of affected parties (Roles involved) |
Harris et al. Line-Drawing Method
Harris, C.E., Pritchard, M.S., & Rabins, M.J. (2018)
Engineering Ethics: Concepts and Cases (6th ed.)
Cengage Learning. Chapter 2: A Practical Ethics Toolkit. Cengage
Engineering-domain methodology for Step E3 (Decision Point Composition). The line-drawing approach determines ethical boundaries by comparing cases to paradigms:
- Identify paradigm cases. Clear-cut examples at extremes (e.g., clear bribe vs. clear non-bribe)
- Identify paradigmatic features. Characteristics that make cases paradigmatic (e.g., gift size, timing, reason, responsibility)
- Compare test case to paradigms. Place "X" on continuum for each feature
- Draw the line. Determine where ethical boundary lies based on feature distribution
Box 2.1: Components of a Moral Problem: Factual Issues (what happened), Conceptual Issues (term meanings), Application Issues (whether concepts apply), Moral Issues (balancing conflicting considerations).
Causal-Normative Links
Definition: Causal-normative links connect professional actions to the obligations they engage. Each action engages an obligation in one of three ways: it fulfills the obligation (directly satisfies it), violates it (directly breaches it), or raises it (puts it in force, or at stake, to be resolved by a later action). The fulfills and violates assignment is grounded in causal responsibility analysis; the raises assignment is grounded in the Event Calculus treatment of an obligation as a fluent.
Sarmiento, C., Bourgne, G., Inoue, K., & Ganascia, J.-G. (2023)
Action Languages Based Actual Causality in Decision Making Contexts
LNCS vol. 14276, pp. 247-262. Springer. DOI | arXiv
Uses the NESS test (Necessary Element of a Sufficient Set) for causal responsibility attribution. An action is causally responsible for an outcome if it was a necessary element in a set of conditions jointly sufficient to produce that outcome. ProEthica applies NESS to determine which actions fulfill or violate specific obligations.
Grounding for raises (at stake): Berreby, F., Bourgne, G., & Ganascia, J.-G. (2017), A Declarative Modular Framework for Representing and Applying Ethical Principles (AAMAS 2017, pp. 96-104); and Dennis, L., et al. (2016), Formal Verification of Ethical Choices in Autonomous Systems, with Dennis & del Olmo (2021), A Defeasible Logic Implementation of Ethical Reasoning.
In the Event Calculus an obligation is a fluent that a happening initiates: the
action puts the obligation in force rather than statically satisfying or breaching it, and a later
action resolves it. The defeasible account treats obligations as coming into force under conditions
and remaining defeasibly in play until discharged. This is the action-side analog of the core State
linkage activatesObligation / defeasibleUnder: a happening
raises an obligation just as a State activates one.
Application: The Analysis and Timeline tabs display causal-normative links showing, per action, the obligations it fulfills, violates, or raises (puts at stake), enabling systematic evaluation of professional conduct across the case timeline.
These are not free-text annotations. Each is materialized as an object-property edge from the
Action to the actual extracted Obligation individual (fulfillsObligation /
violatesObligation / raisesObligation), and the guiding principle as
guidedByPrinciple to the extracted Principle, resolved to the case individuals rather
than restated as prose. This closes the loop the fluent layer begins: the Action initiates a
State, the State activates the real Obligation or Constraint, and the Action fulfills, violates,
or raises that same obligation individual, so the whole chain is queryable.
Fluents and Fluent Transitions (Event Calculus)
Definition: A State is a fluent, a time-varying property that holds over an interval of the case. A happening (an Action, volitional, or an Event, non-volitional) changes what holds: it initiates a fluent (brings it into force) or terminates it (ends it). Inertial fluents persist by default until terminated. The fluent layer is what connects the temporal components to the normative ones: a happening initiates a State, and the State activates obligations and constraints.
Kowalski, R., & Sergot, M. (1986)
A Logic-Based Calculus of Events
New Generation Computing, 4(1), pp. 67-95.
Berreby, F., Bourgne, G., & Ganascia, J.-G. (2017)
A Declarative Modular Framework for Representing and Applying Ethical Principles
Proceedings of AAMAS 2017, pp. 96-104. ACM | PDF
Kowalski & Sergot introduce the Event Calculus: a happening initiates or terminates a fluent, and a fluent holds at a time if it was initiated and not since terminated. Berreby et al. apply this to machine ethics, distinguishing inertial fluents (persist until terminated) from non-inertial ones (hold momentarily) and agent actions from automatic events.
Application: Step 3 records, per Action/Event, the States it brings into or
out of holding, materialized at commit as proeth-core:initiates and
proeth-core:terminates edges (Action/Event to State). These generalize the
State-side activatedByEvent / terminatedByEvent to both happening
kinds in the canonical happening-to-fluent direction.
Temporal Position and Ordering (OWL-Time / Allen)
Definition: Where the Event Calculus says what holds and why it changed, OWL-Time says when. Each happening is anchored to a temporal entity, and happenings are ordered by the thirteen Allen interval relations, standardized as OWL-Time object properties.
Allen, J. F. (1983)
Maintaining Knowledge about Temporal Intervals
Communications of the ACM, 26(11), pp. 832-843.
Cox, S., & Little, C. (2017)
Time Ontology in OWL
W3C Recommendation (namespace http://www.w3.org/2006/time#).
W3C
Allen's interval algebra defines the thirteen relations between intervals (before, meets, overlaps,
during, starts, finishes, equals, and inverses); OWL-Time provides them as
time:intervalBefore, time:intervalMeets, and so on, over
time:ProperInterval, plus time:Instant for point occurrences and
time:hasTime to anchor any entity to a temporal entity.
Application: Each Action/Event carries a proeth:temporalExtent
(instant or interval) and is anchored via time:hasTime to a minted
time:Instant or time:ProperInterval; the Allen relations between
happenings are stored as time:interval* object properties, and
proeth:temporalSequence gives the discrete chronological order shown in the Timeline.
Question Emergence (Toulmin Warrants)
Definition: Question emergence analysis identifies how ethical questions arise from triggering conditions and competing normative claims.
Toulmin, S.E. (1958/2003)
The Uses of Argument
Cambridge University Press. DOI
In Toulmin's model, Data (facts/events) trigger ethical consideration when they invoke Warrants (principles/obligations that license conclusions). Competing warrants arise when the same data supports different conclusions through different normative pathways.
Application: The Analysis tab shows question emergence with triggering events/actions (data) and competing warrants (obligations that pull toward different answers).
Ethical Frameworks (Research Exploratory)
Status: The ethical framework categories used in theoretical question generation (deontological, virtue ethics, utilitarian, etc.) are exploratory research features and are not yet formally defined in the ProEthica ontology.
Questions labeled as "Theoretical" are generated by prompting the LLM to consider the case from different ethical perspectives. These labels indicate the intended analytical lens rather than a formally modeled ethical framework.
Future Work: A formal ontology of ethical frameworks (deontology, virtue ethics, consequentialism, care ethics, etc.) with their relationships to principles, obligations, and reasoning patterns is planned for future development.
Additional References
Additional academic references will be added as the system develops.