The truth crisis: How AI is reshaping knowledge and power worldwide

A new study warns that algorithmic systems and generative AI are fundamentally altering the foundations of knowledge, shifting authority away from traditional institutions and toward opaque computational infrastructures.

The study, titled "Artificial Truth: Algorithmic Power, Epistemic Authority, and the Crisis of Democratic Knowledge," published in Societies, presents a theoretical framework explaining how digital platforms, recommendation systems, and AI-driven content generation are reorganizing truth regimes.

The research claims that a deeper transformation is underway. Truth is increasingly determined not by correspondence with reality but by visibility, engagement, computational plausibility, and platform logic. This shift signals the rise of what the study defines as "Artificial Truth," a new form of epistemic governance driven by algorithmic systems.

Algorithmic systems are replacing institutions as gatekeepers of truth

The study identifies a major structural shift in how societies determine what counts as credible knowledge. Traditionally, institutions such as journalism, science, and academia acted as primary gatekeepers, relying on professional norms, verification processes, and peer validation to establish authority.

In digital environments, these structures are being replaced or bypassed. Platforms now act as intermediaries that rank, recommend, and amplify content based on algorithmic criteria. These systems determine which information becomes visible and which remains hidden, effectively shaping public reality.

This transformation is not simply about access to information but about control over attention. Algorithms prioritize content that generates engagement, meaning that popularity, emotional resonance, and visibility increasingly function as indicators of credibility. As a result, institutional expertise is no longer the dominant source of authority.

According to the study, this shift represents a restructuring of symbolic capital. Instead of credentials and expertise, influence is now tied to metrics such as follower counts, likes, and shares. Actors who understand platform dynamics and optimize content for visibility gain epistemic authority, often outperforming traditional experts.

Users are also increasingly delegating judgment to algorithms. Rather than evaluating sources or arguments directly, they rely on recommendations, rankings, and summaries generated by systems they do not fully understand. This phenomenon, described as algorithmic trust, reflects a broader shift in how knowledge is consumed and validated.

When visibility becomes the primary driver of credibility, truth is shaped by system design rather than by established standards of verification. This reconfiguration of authority marks a fundamental change in the structure of knowledge production.

Generative AI produces 'synthetic truth' that feels real but lacks grounding

The study argues that generative AI represents a second major transformation in epistemic systems. Unlike earlier technologies that organized or filtered information, large language models actively generate new content, functioning as epistemic actors rather than passive tools.

These systems produce responses that mimic expert discourse, using technical language, structured arguments, and coherent narratives. However, their outputs are based on probabilistic pattern recognition rather than actual understanding or verification. This leads to the emergence of what the study calls synthetic truth. Instead of being grounded in evidence or correspondence with reality, truth claims are generated based on statistical plausibility. The system predicts what is likely to sound correct, not what is necessarily true.

One of the most significant findings is that fluency itself becomes a marker of authority. Users tend to equate clarity, coherence, and confidence with accuracy, even when they are aware that the system operates without genuine knowledge. This creates a powerful illusion of expertise. The conversational design of generative AI further reinforces this effect. By simulating dialogue and responsiveness, these systems encourage users to attribute intentionality and understanding to them. This shifts epistemic agency, as machine-generated outputs become reference points in everyday reasoning.

The study also highlights the risks associated with this transformation. Generative AI can produce false or misleading information that appears credible, making it difficult for users to distinguish between accurate and inaccurate content. These outputs are not random errors but structural features of systems optimized for plausibility.

Over time, repeated interaction with generative systems normalizes their authority. Users begin to rely on them as default sources of information, reducing the role of critical evaluation and independent verification. This process gradually reshapes how knowledge is produced and consumed.

Automated fact-checking turns truth into a technical output

A third key transformation identified in the study is the rise of automated fact-checking systems, which translate complex and contested judgments into computational outputs. These systems classify information as true or false using machine learning models trained on labeled datasets.

While often presented as neutral tools, the study argues that these systems embed specific assumptions about what counts as truth. Decisions about which sources to trust, how to interpret evidence, and how to resolve disagreement are encoded into algorithms and datasets.

This process, described as computational veridiction, reduces truth to measurable outputs such as scores, labels, or rankings. Instead of engaging with the complexity of arguments, context, and interpretation, systems produce simplified classifications that may overlook ambiguity and nuance.

The study highlights that this transformation introduces significant risks. By forcing complex issues into binary categories, automated systems can marginalize legitimate perspectives and reinforce dominant viewpoints. This creates the potential for epistemic injustice, particularly for marginalized communities whose knowledge may not align with dominant frameworks.

Another critical issue is the construction of ground truth. The datasets used to train these systems are shaped by human decisions, including which sources are considered authoritative and how labels are assigned. These choices reflect specific cultural, institutional, and geopolitical contexts.

The study notes that much of this work is performed through distributed labor, often involving workers in different regions who contribute to data annotation. At the same time, control over system design and decision-making remains concentrated in a small number of institutions and corporations.

This asymmetry reflects broader patterns of power in the digital economy. Algorithmic systems do not simply process information but reproduce and amplify existing inequalities, shaping whose knowledge is recognized and whose is excluded.

Artificial truth signals a shift in democratic knowledge systems

The study brings these transformations together under the concept of Artificial Truth, describing a new regime in which truth is governed by algorithmic systems rather than by traditional institutions.

In this regime, three dynamics converge. First, the restructuring of trust shifts authority from institutions to platforms. Second, generative AI produces synthetic knowledge that prioritizes plausibility over accuracy. Third, automated systems translate truth into computational outputs that appear objective but embed specific assumptions and biases.

Together, these changes reconfigure the public sphere. Instead of open deliberation shaped by identifiable actors, knowledge is increasingly mediated by proprietary systems optimized for engagement and efficiency. This introduces new forms of power that are difficult to observe and challenge.

One of the most notable implications is the privatization of epistemic authority. Platforms now control the infrastructures through which information is produced, distributed, and validated. This gives them a central role in shaping public knowledge without corresponding levels of democratic accountability.

The study also highlights a transformation in epistemic citizenship. Users are increasingly positioned as consumers of pre-selected information rather than as active participants in knowledge production. Algorithmic curation reduces the need for independent evaluation, leading to a form of passive engagement with information.

This shift has broader democratic consequences. When truth becomes a function of algorithmic systems, the space for disagreement, debate, and critical inquiry may be reduced. Issues that were once contested through public deliberation are reframed as technical problems to be solved through optimization.

The consequence is what the study describes as a potential erosion of democratic knowledge systems. As algorithmic mediation becomes more pervasive, the criteria for truth become less transparent and less open to contestation