Source-linked AI summary
The relationship between professional and general ethics in generative AI
Omri Asscher, Mirco Musolesi
TL;DR
Generative AI increasingly performs professional tasks, but AI-ethics research has focused mainly on general values rather than profession-specific ethics. The paper develops a four-dimensional framework for analyzing how general and professional ethics interact in situated conflicts, arguing that their equilibrium prioritizes particular professional ethics and can be empirically evaluated and reshaped. It also limits its scope to ethical decision-making rather than the governance and political-economic effects of AI on human professions.
Problem
AI-ethics research and policy have largely assumed general values while paying relatively little attention to professional ethics embedded in generative-AI decision-making.
Method
The paper conceptualizes professional AI ethics through professional ethics, general ethics, their relationship, and situational context, then uses this framework to guide evaluation and intervention.
Results
The framework argues that AI models prioritize and implement particular professional ethics through an equilibrium between professional and general ethical layers in situated conflicts.
Takeaways & Limitations
Systematic evaluation across similar but nonidentical professional scenarios can identify models’ favored ethics and inform efforts to reshape them.
Takeaways & Limitations
The article addresses the ethical underpinnings of practice-specific AI decision-making but leaves AI governance, regulation, and broader effects on human professions outside its scope.
Abstract
from arXiv · showhide
Recent years have seen a growing discrepancy in the field of AI alignment: research and policy recommendations on AI ethics tend to assume a general set of ethical values, yet proliferating practice-specific uses of AI systems on the ground - in the legal, medical and translation domains, among others - have been effectively manifesting ethics of professional practice. This article begins by outlining the reasons why general and professional ethics are increasingly conflicted in contemporary AI systems, and by surveying how the research literature attests to, but has not yet resolved, this conceptual and practical challenge. We then conceptualize the main dimensions of AI models' decision-making in areas of professional practice, emphasizing professional ethics' hierarchically structured relationship with general ethics, and elaborating on the mechanisms through which they reach an equilibrium in situational contexts that involve conflict. It is through this equilibrium, we suggest, that certain professional ethics are prioritized over others and implemented in practice. We then show how our framework can be the basis for a systematic empirical assessment of AI models' professional ethics in various domains, identifying the nuances of the models' favored ethic by examining their production in a series of similar but not identical scenarios. Finally, we propose a formulation for how to intervene in and change AI models' favored ethics in professional practices - while noting the inherent dimension of subjectivity involved in both the evaluation and implementation of professional ethics in AI models.
1. Introduction
Generative AI has intensified a mismatch between universal AI-ethics frameworks and the profession-specific ethics manifested in real-world uses. Because the same models handle diverse expertise while retaining general traits, their ethical agency must account for interactions between general and professional ethics.
- AI ethics research and policy generally assume universal values, while deployed systems increasingly perform ethically implicated tasks governed by professional practices.
- Generative AI is newly holistic and multitask: one entity makes decisions across expertise-specific contexts while retaining general cognitive, sociocultural, cultural, and ethical traits.
- The same LLM can function as both general and narrow AI, so ethical analysis must examine how general and professional repertoires intersect and shape outputs.
- Customization and personalization have expanded individual and autonomous AI use into professional practices that previously relied mainly on human experts.
- The article surveys the literature, conceptualizes four dimensions of practice-specific AI ethics, and proposes using them to support systematic empirical evaluation.
2. AI ethics and professional ethics: main themes in the research literature
The literature contrasts broad AI-ethics principles with richer, domain-specific professional frameworks, while treating their relationship as interdependent and unresolved. The article therefore studies AI professional ethics without directly addressing the political economy or regulation shaping human professions.
- AI-ethics scholarship and policy have paid relatively little attention to how practice-specific ethics are embedded in generative-AI decision-making.
- Common AI principles such as beneficence, autonomy, justice, and explicability remain broad when applied to distinctive professional roles and constraints.
- Professional ethics in medicine, translation, law, and other fields address domain-specific stakeholders, obligations, and situations more directly than sweeping general systems.
- The literature generally treats professional ethics as separate from yet dependent on general morality, although it does not reach full consensus on their relationship.
- Conflicts between general and professional claims require an equilibrium whose balance determines the ethical output, without the article taking a normative position on the relationship.
- The article conceptualizes AI decision-making in professional practices while leaving the effects of AI on human professions, governance, and regulation outside its scope.
3. Ethics of professional practice in generative AI: four main dimensions
The framework models AI professional ethics through four interacting dimensions: embedded professional ethics, embedded general ethics, their relationship, and situational context. It explains how context-sensitive conflicts can be resolved through hierarchical adjustment toward an equilibrium, while recognizing interpretation and subjectivity.
- The framework identifies four dimensions: professional ethics, general ethics, their hierarchical or other relationship, and situational context.
- AI professional ethics are diverse, discursive, and learned from practice-related discourse rather than represented by one unified formal rule system.
- Professional guidelines require interpretation, so different approaches within the same field can produce different implementations of ethical ideals.
- General ethics are multiple and prioritized differently, while professional ethics retain distinctive imperatives that cannot be reduced to universal morality.
- 3.3 The situational context: Situational context supplies the concrete details through which abstract ethical principles are applied in translation, medicine, law, coding, and other practices.
- 3.4 The relationship between professional ethics and general ethics: The prescribed relationship between ethical layers determines how competing principles are reconciled when they conflict in a given context.
- 3.4 The relationship between professional ethics and general ethics: Models may iteratively adjust outputs toward a reflective equilibrium between potentially hierarchical professional and general ethical demands.
- 3.4 The relationship between professional ethics and general ethics: Some professional ethics incorporate universal concerns more deeply than others, allowing general obligations to override profession-specific duties in particular cases.
4. Practical implications: how to evaluate equilibriums between general and professional ethics, and how to reshape them
The framework treats generative AI as already prioritizing professional ethics and reconciling them with general ethics in context. It proposes evaluating these equilibriums across scenarios and reshaping them through normatively specified interventions, while recognizing evaluator subjectivity.
- Generative AI models already prioritize some professional ethics over others and reconcile professional and general ethics when they conflict.
- Models acquire professional ethics through training on general-ethical and practice-specific texts, while personalization and later training can further shape responses.
- Both ethical assessment and intervention require some subjective interpretation and preference from the evaluator or user.
- Evaluation should examine prioritized professional ethics, prioritized general ethics, and their hierarchical relationship across multiple situational contexts.
- Similar but non-identical scenarios are needed to identify an ethical equilibrium rather than treating one response as representative.
- Quantitative evaluation can compare a benchmark with the model’s general-professional ethical configuration and measure their divergence, including with Kullback-Leibler divergence.
- Intervention can reduce divergence from a desired ethical relationship by rewarding professional-ethics alignment while constraining departures from existing policy.
5. Conclusion
The article argues that generative AI is already performing ethically consequential professional practices and therefore requires systematic evaluation and, where necessary, intervention. Its framework centers on equilibria between professional and general ethics while limiting its claims to structural relationships rather than resolving moral relativism.
- LLMs increasingly perform legal, medical, translation, coding, and other professional practices, expanding the volume of ethically implicated decisions.
- The article proposes empirical evaluation and intervention because AI-mediated professional practices are expanding and disrupting established professions.
- The framework explains how professional and general ethics reach situational equilibria through which particular professional ethics are prioritized in practice.
- The analysis addresses structural relationships between professional and general ethics, not the broader normative question of whether all ethical options are equally valid.