In the effort to develop ethical frameworks for artificial intelligence, many conventional approaches fall short by treating humans as the only true moral agents and AI as inert tools under human control. Bruno Latour’s Actor-Network Theory (ANT) provides a radically different lens, one that sees ethics as something that emerges from the shifting, interconnected networks of human and non-human elements alike. Rather than imposing rigid, human-centered rules, ANT invites us to trace how agency, responsibility, and moral outcomes arise through associations among diverse actors. This perspective dissolves old hierarchies and opens up creative, adaptive ways to craft AI ethics—ones that are relational, experimental, and responsive to real-world contingencies.
Bruno Latour’s Actor-Network Theory: Core Ideas
Actor-Network Theory views the world as composed of constantly forming and reforming networks where humans and non-humans interact on equal footing. Key concepts include:
– Actants: Any entity—human or non-human—that can make a difference or exert influence counts as an actant. This includes people, algorithms, datasets, hardware, regulations, interfaces, and even physical environments. No inherent privilege is given to human intentionality.
– Symmetry: Humans and non-humans are analyzed with the same conceptual tools. An AI model is as much an actant as the engineer who trains it or the policy that governs its use.
– Translation: Actions occur through processes of negotiation, alignment, persuasion, or resistance. Actants translate one another’s interests, forming alliances that produce effects. Mediators (such as training data or decision interfaces) actively transform meanings and outcomes along the way.
– Networks and Black-Boxing: Networks are assemblages of relations that can stabilize into seemingly solid entities (“black boxes”) where internal complexity is hidden. Ethics emerges from these relations, not from isolated principles.
– Following the Actors: Inquiry begins by empirically tracing associations rather than starting with abstract categories or external forces.
ANT rejects dualisms—subject versus object, society versus technology, nature versus culture—and insists on a flat ontology where power and agency are enacted through connections. For AI, this means ethical questions are never about the machine alone but about how the entire network assembles and performs morality.
ANT and AI: Ethics as Emergent from Networks
Applying ANT to AI reveals that ethical issues are distributed across networks rather than concentrated in any single point. A recommendation algorithm, for example, is not merely code; it is a network involving developers, training datasets (drawn from historical patterns), user interactions, platform incentives, regulatory constraints, and feedback loops. Ethical harms—such as amplifying stereotypes—arise from translations within this assemblage: biased data mediating skewed outputs, corporate goals aligning with engagement metrics, users enacting certain behaviors that reinforce patterns.
By following the actors, ANT exposes how non-human elements actively shape outcomes. Datasets “act” by constraining possibilities; algorithms mediate by filtering and prioritizing; hardware enables or limits scale. This distributed view challenges simplistic blame (e.g., “the AI is biased”) and instead highlights how responsibility is co-produced across the network. It also shows AI as hybrid—neither fully autonomous nor purely instrumental—but as quasi-objects whose agency emerges relationally.
New Possibilities Opened by ANT for AI Ethics
ANT does not deliver ready-made ethical codes; it generates possibilities by encouraging empirical tracing, reflexive experimentation, and inclusive network-building. Several promising directions emerge:
1. Ethical Network Assembly and Participation
Ethics can be treated as dynamic assemblages rather than fixed doctrines. This opens the way for hybrid ethical committees or processes that include diverse actants: affected communities, bias-detection tools, simulation models, legal texts, and prototype interfaces. In medical AI, for instance, diagnostic networks could incorporate patient advocacy groups and data-resistance mechanisms (e.g., opt-out protocols) as active translators, helping realign the system toward equity and consent.
Such participatory assemblies distribute moral agency, reducing the risk of top-down imposition and allowing marginalized voices to mediate outcomes.
2. Controversy Mapping and Reflexive Inquiry
ANT thrives on tracing controversies—moments when networks destabilize and hidden relations become visible. For AI, this suggests building controversy-tracing features: systems that log and visualize actor interactions during decisions, making black boxes temporarily transparent.
In autonomous decision-making (e.g., content moderation or loan approval), controversy maps could reveal how certain actants (like skewed historical data) dominate translations, inviting negotiation and reconfiguration. This turns ethics into ongoing, empirical inquiry rather than pre-set compliance.
3. Pluralism and Localized Translations
By embracing multiplicity—multiple co-existing realities—ANT supports ethical pluralism. Instead of one-size-fits-all principles, ethics become localized translations adapted to cultural, contextual networks.
An educational AI might assemble differently in varied settings, with heuristics mediating between individual achievement and communal values. This flexibility allows AI ethics to respect diverse moral landscapes while remaining responsive to local resistances and alliances.
4. Resilience Through Reconfiguration
Networks are always precarious and open to reassembly. ANT thus inspires designs that anticipate disruption and enable quick ethical realignment. Fail-safe mediators (ethical override actors), redundant pathways, or self-auditing loops can make systems more resilient.
In high-stakes domains like policing or finance, networks could include whistleblower actants that detect manipulative translations and trigger reconfiguration, preventing irreversible ethical drift.
Addressing Challenges Within an ANT Framework
ANT’s openness can risk endless relativism or diluted accountability, especially when powerful actants (e.g., dominant corporations) shape networks disproportionately. To balance this, ANT-inspired ethics might combine relational tracing with normative anchors—justice, dignity, equity—used as orienting translations rather than absolute truths.
Implementation demands practical tools: mapping software, simulation environments, and documentation standards that make actor relations visible and contestable. These challenges themselves become part of the ethical network, inviting further translation and refinement.
An Open, Relational Horizon for AI Ethics
Bruno Latour’s Actor-Network Theory reframes AI ethics from a problem of control to an opportunity for creative co-creation. By seeing ethics as emerging from networks of human and non-human actants, we move beyond static rules toward living, adaptive practices: inclusive assemblies, controversy-driven reflection, pluralistic translations, and resilient reconfigurations.
This approach invites us to follow the actors, mediate relations thoughtfully, and assemble networks that perform better worlds. In an era of accelerating AI, ANT offers not closure but an open ethical horizon—one where humans and machines collaborate in ongoing, relational moral invention.
