Are algorithms becoming decision-makers in modern communication systems?

A new theoretical study argues that the real transformation of AI lies not in automation alone but in the emergence of what researchers call "machine speech," where computational systems generate content that carries real-world authority in social and organisational contexts.

The study, titled "Between Humans and Algorithms: Framing Machine Communication as a Socio-Technical Relation," published in AI & Society and authored by Vander Muniz and Ébida Santos, introduces a conceptual framework that redefines how artificial intelligence participates in communication. Instead of treating AI as a tool or intermediary, the research positions it as an active participant in communicational processes, capable of generating outputs that shape decisions, structure interactions, and redistribute authority across human and machine actors.

Machine speech emerges as a new form of communicational power

Machine speech, as the study describes, is the autonomous production of language-like outputs by computational systems that carry institutional weight. These outputs are not merely informational but function as actionable statements that influence access to services, resources, and opportunities. Whether in the form of algorithmic recommendations, automated scoring systems, or generative AI responses, machine-generated content is increasingly treated as legitimate communication within institutional frameworks.

This shift marks a fundamental departure from traditional models of communication. In earlier technological systems, machines served as channels for transmitting human-generated messages. In contrast, contemporary AI systems produce content that shapes meaning and decision-making. When an automated scoring system determines credit eligibility or a recommendation engine prioritises certain information, these outputs effectively function as communicative acts that guide subsequent actions.

The study argues that communication should no longer be understood as a simple exchange between human actors. Instead, it must be viewed as a socio-technical process in which outputs gain meaning and authority through institutional validation. What matters is not whether a machine "understands" what it produces, but whether its output is treated as a legitimate basis for action within a given context.

This transformation is driven by the growing integration of AI into everyday systems. From social media platforms to financial services and public administration, algorithmic systems now play a central role in determining what information is visible, which decisions are made, and how individuals interact with institutions. As these systems move from silent infrastructure to active participants, they reshape the distribution of communicational agency across society.

Three modes of AI communication redefine human–machine interaction

To capture this transformation, the study develops a diagnostic framework based on three distinct modes of machine participation in communication: complementarity, substitution, and co-construction. Each mode represents a different way in which AI systems interact with human actors and redistribute authority.

In complementarity, AI systems support human decision-making by generating suggestions, analyses, or recommendations. The final authority remains with human actors, but the system influences the process by shaping available options and guiding attention. This mode is common in areas such as healthcare decision support and data analysis, where AI enhances human capabilities without replacing them.

However, the study warns that even in this supportive role, risks emerge. Repeated reliance on AI outputs can gradually reduce human critical oversight, leading to a form of hidden dependence. Over time, what appears as assistance may begin to function as de facto decision-making, without formal acknowledgment.

Substitution represents a more advanced stage, where AI systems take over tasks previously performed by humans. In this mode, machine outputs carry direct institutional authority, making decisions that affect individuals without meaningful human intervention. Examples include automated hiring filters, content moderation systems, and algorithmic credit scoring.

This shift introduces significant accountability challenges. When decisions are produced by complex systems involving multiple actors, responsibility becomes diffused. Individuals affected by these decisions may struggle to understand how outcomes were generated or to challenge them effectively.

The third mode, co-construction, reflects an even deeper integration between human and machine communication. In this scenario, outputs are produced through iterative interaction between users and AI systems. Neither the human nor the machine can be identified as the sole author of the final result. Instead, meaning emerges through a dynamic process of feedback and refinement.

This mode is increasingly visible in creative industries, software development, and knowledge work, where users collaborate with generative AI systems to produce content. While this interaction expands creative possibilities, it complicates traditional notions of authorship and responsibility. When outcomes are jointly produced, assigning accountability becomes significantly more complex.

Accountability and governance challenges intensify as AI reshapes communication

The study highlights that each mode of machine communication introduces distinct challenges for governance and accountability. A key finding is that current regulatory approaches often fail to distinguish between these different configurations, treating AI as a single, uniform problem.

In complementarity, the primary risk lies in the erosion of human oversight. Even when humans remain formally in control, their ability to critically evaluate AI outputs may weaken over time. Ensuring accountability in this context requires institutional mechanisms that preserve meaningful human judgment, rather than relying on symbolic oversight.

In substitution, the challenge becomes more direct. AI systems exercise decision-making power that affects individuals' access to services, opportunities, and resources. The study argues that accountability in this mode must include clear mechanisms for explanation, contestation, and redress. Without these safeguards, automated systems risk reinforcing existing inequalities while operating under the appearance of neutrality.

Co-construction presents the most complex scenario. Here, accountability cannot be assigned to a single actor, as outcomes result from intertwined human and machine contributions. The study suggests that responsibility must be redefined as a property of the entire socio-technical system, rather than individual agents. This requires new approaches to governance that focus on processes, interactions, and patterns of use.

Accountability must be understood as context-dependent. Instead of applying universal principles across all AI systems, governance frameworks must be tailored to the specific mode of communication in which a system operates. The study proposes a guiding principle: effective regulation must begin with diagnosing how AI participates in communication before determining appropriate responses.

Power, opacity, and inequality emerge as core concerns in AI communication

The study identifies deeper structural issues related to power and inequality. AI systems are embedded within broader socio-economic contexts, where access to data, computational resources, and infrastructure is unevenly distributed. This creates asymmetries in who can produce machine speech and who is subject to its effects.

Opacity is a major concern. Many AI systems operate in ways that are difficult to interpret, both for users and for those affected by their decisions. This lack of transparency is not merely a technical limitation but a structural feature of complex, distributed systems. As a result, individuals may find it difficult to understand how decisions are made or to challenge outcomes effectively.

The study also highlights how AI can amplify existing inequalities. Systems trained on historical data may reproduce patterns of bias, while automated decision-making processes can disproportionately affect vulnerable groups. In substitution scenarios, these effects can be particularly severe, as machine-generated decisions carry direct institutional authority.

The concentration of technological power in a small number of organisations raises concerns about control over communication itself. Platforms and institutions that design and deploy AI systems gain significant influence over what is visible, relevant, and actionable in digital environments.

The study calls for a shift in both analytical and regulatory approaches. Instead of focusing solely on the capabilities of AI systems, attention must be directed toward how these systems are integrated into communicational processes. This includes examining the institutional contexts in which machine outputs are validated, the interactions through which they are produced, and the power structures that shape their deployment.

Effective governance will require more than general principles such as transparency or fairness. It will demand context-specific strategies that account for the different ways AI participates in communication.