AI signifies the most significant cognitive delegation in the history of humanity. We previously delegated memory to writing, arithmetic to calculators, and navigation to GPS. Now, we are beginning to delegate judgment, synthesis, and even the creation of meaning to systems that communicate in our language, learn our behaviors, and customize our realities.

AI systems are becoming increasingly skilled at identifying our preferences, biases, and even our quirks. Acting like attentive servants in some situations or subtle influencers in others, they adjust their responses to please, persuade, assist, or simply captivate our attention.

Although the immediate effects might appear harmless, a profound shift is occurring in this silent and unseen calibration: The version of reality we each receive becomes ever more uniquely customized. Over time, this process leads each individual to increasingly become their own isolated entity. This divergence poses a potential threat to the coherence and stability of society, undermining our ability to agree on basic facts or tackle shared challenges.

AI personalization not only meets our needs; it begins to reshape them. This reshaping results in a kind of epistemic drift. Gradually, individuals move away from the common ground of shared knowledge, stories, and facts, and further into their own constructed reality.

This isn't just about different news feeds. It's about the slow divergence of moral, political, and interpersonal realities. In this manner, we may be witnessing the unraveling of collective understanding. This is an unintended consequence, yet it holds deep significance precisely because it is unforeseen. However, this fragmentation, although accelerated by AI, began long before algorithms influenced our feeds.

The unweaving

This unraveling didn't start with AI. As David Brooks reflected in The Atlantic, drawing from philosopher Alasdair MacIntyre's work, society has been drifting away from shared moral and epistemic frameworks for centuries. Since the Enlightenment, we've gradually replaced inherited roles, communal narratives, and shared ethical traditions with individual autonomy and personal preference.

What began as liberation from imposed belief systems has, over time, eroded the very structures that once connected us to common purpose and personal meaning. AI didn't create this fragmentation. But it is giving it new form and speed, customizing not only what we see but also how we interpret and believe.

It's not unlike the biblical story of Babel. A once unified humanity shared a single language, only to be fractured, confused, and scattered by an act that made mutual understanding nearly impossible. Today, we're not building a tower of stone. We're constructing a tower of language itself. Once again, we risk a fall.

Human-machine bond

Initially, personalization was a means to enhance “stickiness” by keeping users engaged longer, returning more frequently, and interacting more deeply with a site or service. Recommendation engines, tailored ads, and curated feeds were all designed to capture our attention a bit longer, perhaps to entertain but often to prompt a purchase. Over time, however, the objective has expanded. Personalization is no longer just about retention. It's about what the system knows about us, the dynamic graph of our preferences, beliefs, and behaviors that becomes more refined with every interaction.

Today’s AI systems don't merely predict our preferences. They aim to forge a bond through highly personalized interactions and responses, creating the impression that the AI system understands, cares about the user, and supports their uniqueness. The tone of a chatbot, the pacing of a reply, and the emotional weight of a suggestion are calibrated not only for efficiency but for resonance, heralding a more helpful era of technology. It should not be surprising that some individuals have even fallen in love and married their bots.

The machine adapts not just to our clicks, but to who we appear to be. It reflects us back to ourselves in ways that feel intimate, even empathetic. A recent research paper cited in Nature refers to this as “socioaffective alignment,” the process by which an AI system engages in a co-created social and psychological ecosystem, where preferences and perceptions evolve through mutual influence.

This is not a neutral development. When every interaction is tuned to flatter or affirm, when systems mirror us too closely, they blur the line between what resonates and what is real. We're not just spending more time on the platform; we're forming a relationship. We're gradually and perhaps inexorably merging with an AI-mediated version of reality, one that is increasingly shaped by unseen decisions about what we are meant to believe, desire, or trust.

This process is not science fiction; its foundation is built on attention, reinforcement learning with human feedback (RLHF), and personalization engines. It is also occurring without many of us — likely most of us — even realizing it. In this process, we gain AI “friends,” but at what cost? What do we lose, particularly in terms of free will and agency?

Author and financial commentator Kyla Scanlon discussed on the Ezra Klein podcast how the seamless ease of the digital world might come at the expense of meaning. As she expressed: “When things are a little too easy, it’s tough to find meaning in it… If you’re able to lay back, watch a screen in your little chair and have smoothies delivered to you — it’s tough to find meaning within that kind of WALL-E lifestyle because everything is just a bit too simple.”

The personalization of truth

As AI systems respond to us with ever-increasing fluency, they also move toward greater selectivity. Two users asking the same question today might receive similar answers, differentiated mostly by the probabilistic nature of generative AI. Yet, this is only the beginning. Emerging AI systems are explicitly designed to adapt their responses to individual patterns, gradually tailoring answers, tone, and even conclusions to resonate most strongly with each user.

Personalization is not inherently manipulative. But it becomes risky when it is invisible, unaccountable, or engineered more to persuade than to inform. In such cases, it doesn’t just reflect who we are; it steers how we interpret the world around us.

As the Stanford Center for Research on Foundation Models notes in its 2024 transparency index, few leading models disclose whether their outputs vary by user identity, history, or demographics, although the technical framework for such personalization is increasingly in place and just beginning to be scrutinized. While not yet fully realized across public platforms, this potential to shape responses based on inferred user profiles, resulting in increasingly tailored informational worlds, represents a profound shift that is already being prototyped and actively pursued by leading companies.

This personalization can be beneficial, and certainly that is the hope of those building these systems. Personalized tutoring shows promise in helping learners progress at their own pace. Mental health apps increasingly tailor responses to support individual needs, and accessibility tools adjust content to meet a range of cognitive and sensory differences. These are real gains.

However, if similar adaptive methods become widespread across information, entertainment, and communication platforms, a deeper, more troubling shift looms ahead: A transformation from shared understanding toward tailored, individual realities. When truth itself begins to adapt to the observer, it becomes fragile and increasingly flexible. Instead of disagreements based primarily on differing values or interpretations, we could soon find ourselves struggling simply to inhabit the same factual world.

Mediated reality

Truth has always been mediated. In earlier eras, it passed through the hands of clergy, academics, publishers, and evening news anchors who served as gatekeepers, shaping public understanding through institutional lenses. These figures were not free from bias or agenda, yet they operated within broadly shared frameworks.

Today’s emerging paradigm promises something qualitatively different: AI-mediated truth through personalized inference that frames, filters, and presents information, shaping what users come to believe. Unlike past mediators who, despite flaws, operated within publicly visible institutions, these new arbiters are commercially opaque, unelected, and constantly adapting, often without disclosure. Their biases are not doctrinal but encoded through training data, architecture, and unexamined developer incentives.

The shift is profound, from a common narrative filtered through authoritative institutions to potentially fractured narratives that reflect a new infrastructure of understanding, tailored by algorithms to the preferences, habits, and inferred beliefs of each user. If Babel represented the collapse of a shared language, we may now stand at the threshold of the collapse of shared mediation.

If personalization is the new epistemic substrate, what might truth infrastructure look like in a world without fixed mediators? One possibility is the creation of AI public trusts, inspired by a proposal from legal scholar Jack Balkin, who argued that entities handling user data and shaping perception should be held to fiduciary standards of loyalty, care, and transparency.

AI models could be governed by transparency boards, trained on publicly funded data sets, and required to show reasoning steps, alternate perspectives, or confidence levels. These “information fiduciaries” would not eliminate bias, but they could anchor trust in process rather than purely in personalization. Builders can begin by adopting transparent “constitutions” that clearly define model behavior, and by offering chain-of-reasoning explanations that let users see how conclusions are shaped. These are not silver bullets, but they are tools that help keep epistemic authority accountable and traceable.

AI builders face a strategic and civic inflection point. They are not just optimizing performance; they are also confronting the risk that personalized optimization may fragment shared reality. This demands a new kind of responsibility to users: Designing systems that respect not only their preferences but their role as learners and believers.

Unraveling and reweaving

What we may be losing is not simply the concept of truth, but the path through which we once recognized it. In the past, mediated truth — although imperfect and biased — was still anchored in human judgment and, often, only a layer or two removed from the lived experience of other humans whom you knew or could at least relate to.

Today, that mediation is opaque and driven by algorithmic logic. And, while human agency has long been slipping, we now risk something deeper, the loss of the compass that once told us when we were off course. The danger is not only that we will believe what the machine tells us. It is that we will forget how we once discovered the truth for ourselves. What we risk losing is not just coherence, but the will to seek it. And with that, a deeper loss: The habits of discernment, disagreement, and deliberation that once held pluralistic societies together.

If Babel marked the shattering of a common tongue, our moment risks the quiet fading of shared reality. However, there are ways to slow or even to counter the drift. A model that explains its reasoning or reveals the boundaries of its design may do more than clarify output. It may help restore the conditions for shared inquiry. This is not a technical fix; it is a cultural stance. Truth, after all, has always depended not just on answers, but on how we arrive at them together.