In an era where artificial intelligence permeates every facet of human life—from autonomous vehicles navigating city streets to algorithms curating our social media feeds—the question of ethics in AI has never been more pressing. Traditional ethical frameworks, often rooted in individual responsibility and clear lines of accountability, struggle to keep pace with the complexities of modern AI systems. Enter the concept of distributed agency, a paradigm that recognizes decision-making and action as spread across networks of humans, machines, and institutions rather than concentrated in a single entity. This article explores how we can craft robust ethics for AI by building on the foundation of distributed agency, offering a more nuanced approach to responsibility, transparency, and societal impact.
Understanding Distributed Agency in AI
At its core, distributed agency challenges the anthropocentric view of agency—the idea that only humans possess the capacity for intentional action. In AI contexts, agency is not solely attributable to the machine itself but is dispersed among various actors. Consider a self-driving car: the “decision” to swerve around an obstacle isn’t made by the AI in isolation. It’s the product of programmers who designed the algorithms, data scientists who trained the models on vast datasets, regulators who set safety standards, and even the users who input preferences or override systems.
This distribution arises from the interconnected nature of AI development and deployment. AI systems are built on layers of code, data, hardware, and human oversight, often spanning global supply chains and collaborative teams. Philosopher and AI ethicist Joanna Bryson has argued that agency in AI is “modular and distributed,” emphasizing that no single component holds full responsibility. Similarly, in fields like actor-network theory (ANT), scholars like Bruno Latour describe agency as emerging from assemblages of human and non-human elements.
By framing AI through distributed agency, we move beyond simplistic debates like “Is AI sentient?” to more practical questions: How do we trace accountability through these networks? And how can ethics be designed to address this diffusion?
Why Traditional Ethics Fall Short
Conventional ethical models, such as utilitarianism (maximizing overall good) or deontology (rule-based duties), often assume a centralized agent. For instance, Isaac Asimov’s Three Laws of Robotics presuppose a robot as an autonomous entity capable of adhering to predefined rules. But in reality, AI rarely operates in such a vacuum. A biased facial recognition system, for example, might discriminate not because of malicious intent in the code but due to skewed training data sourced from unrepresentative populations, influenced by corporate priorities and regulatory oversights.
This mismatch leads to ethical blind spots. When harms occur—like an AI hiring tool rejecting qualified candidates based on gender biases—the finger-pointing begins: Is it the developer’s fault? The data provider’s? The company’s? Distributed agency reveals that blame is rarely singular; it’s shared. Traditional frameworks risk oversimplifying this, leading to ineffective solutions like vague “AI principles” that companies adopt without meaningful enforcement.
Moreover, in an age of machine learning and neural networks, AI decisions can be opaque even to their creators. The “black box” problem exacerbates distributed agency, as agency flows through inscrutable processes. Ethics must therefore evolve to embrace this complexity, focusing on systemic accountability rather than individual culpability.
Crafting Ethics Based on Distributed Agency
To build ethics for AI on distributed agency, we need a multi-layered approach that integrates principles from philosophy, law, and technology. Here are key strategies:
1. Mapping Agency Networks
The first step is to visualize and document the distributed elements of an AI system. This involves creating “agency maps”—diagrams or audits that outline the roles of all stakeholders. For example, in developing a medical diagnostic AI, the map would include data annotators (who label images), algorithm trainers, clinicians providing feedback, and patients whose data is used.
Such mapping promotes transparency. Organizations like the AI Now Institute advocate for “impact assessments” that trace how agency is distributed and identify potential ethical risks. By making these networks explicit, we can assign proportional responsibilities: Developers ensure fair algorithms, while regulators enforce data privacy standards like GDPR.
2. Shared Responsibility Models
Distributed agency calls for ethics that distribute accountability accordingly. One promising model is “joint agency,” where responsibility is allocated based on control and influence. Legal scholar Ryan Calo proposes adapting tort law to AI, holding parties liable in proportion to their contribution to harm—much like how multiple defendants in a lawsuit share damages.
In practice, this could mean contractual agreements in AI supply chains that mandate ethical audits at each stage. For instance, cloud providers like AWS or Google Cloud could require clients to certify that their AI models undergo bias testing, creating a chain of accountability.
3. Designing for Emergent Behaviors
AI systems often exhibit emergent properties—unintended behaviors arising from interactions within the network. Ethics must anticipate these. Drawing from complex systems theory, we can incorporate “resilience ethics,” which emphasizes adaptability and feedback loops.
For example, in social media algorithms, distributed agency includes users who amplify content, platforms that prioritize engagement, and advertisers who fund it. To craft ethics here, platforms could implement “agency-aware” designs, like modular algorithms where human moderators intervene in high-stakes decisions, or decentralized governance where users vote on content policies.
4. Incorporating Diverse Perspectives
Since agency is distributed, so too should be the ethical deliberation. Inclusive processes— involving ethicists, affected communities, and interdisciplinary experts—ensure that ethics reflect broader societal values. Initiatives like the Partnership on AI bring together tech companies, academics, and NGOs to co-create guidelines.
A case study is the development of autonomous weapons systems (AWS). Groups like the Campaign to Stop Killer Robots argue for bans, highlighting how distributed agency (from military planners to AI engineers) could lead to diffused moral responsibility in warfare. Ethics here must prioritize human oversight to prevent agency from becoming too fragmented.
5. Technological Enablers for Ethical Distribution
Technology itself can support distributed ethics. Blockchain, for instance, offers immutable ledgers to track data provenance, making agency traceable. Explainable AI (XAI) tools, like LIME or SHAP, help demystify decisions, allowing stakeholders to understand their role in the network.
Furthermore, federated learning—a technique where AI models train on decentralized data without sharing it—embodies distributed agency by preserving privacy while distributing computational agency across devices.
Challenges and Critiques
Crafting ethics on distributed agency isn’t without hurdles. One critique is that it could dilute responsibility, allowing actors to evade accountability by pointing fingers elsewhere. To counter this, strong enforcement mechanisms, such as independent oversight bodies, are essential.
Another challenge is scalability. Mapping agency in global AI ecosystems is resource-intensive, potentially burdening smaller developers. Solutions might include standardized templates or AI-assisted auditing tools.
Cultural differences also complicate matters; what constitutes ethical distribution in one society may differ in another. Global frameworks, like UNESCO’s AI Ethics Recommendation, aim to bridge this by promoting universal principles while allowing local adaptations.
Toward a Networked Ethical Future
As AI continues to evolve, embracing distributed agency offers a pathway to ethics that are as interconnected and dynamic as the technologies they govern. By shifting from isolated accountability to shared, systemic responsibility, we can foster AI that benefits society without unintended harms. This isn’t just theoretical—it’s actionable. Policymakers, companies, and researchers must collaborate to implement these ideas, ensuring that ethics keep pace with innovation.
In the end, crafting ethics for AI on the basis of distributed agency reminds us that technology is a human endeavor. By acknowledging the web of agencies involved, we empower ourselves to weave a more just and equitable digital world. As we stand on the cusp of even more advanced AI, like general intelligence, this approach will be crucial in guiding us responsibly forward.
