When you open a streaming service, a social media feed or an email inbox, you encounter thousands of possible choices. Yet what appears before you has already been filtered, ranked and ordered by algorithms you cannot see. This raises a profound psychological question: if machines are deciding what we encounter, are we still truly choosing? The tension between algorithm and agency sits at the heart of how we experience digital life today. Understanding this tension matters because our sense of control over our own decisions shapes not only what we consume, but how we feel about ourselves and our place in the world.
This module explores how recommender systems and artificial intelligence shape the information landscape we navigate every day. We will examine the psychological concept of agency, explore how choice architecture influences decision-making, and ask whether algorithms genuinely narrow our horizons or whether that concern oversimplifies a more complex reality. Throughout, we consider practical ways to maintain genuine autonomy whilst living in a world mediated by intelligent machines.
Agency and Autonomy in Psychology
Agency is the felt sense that you are the author of your own actions and choices. In psychology, this experience is fundamental to wellbeing. Researchers such as Deci and Ryan have shown that people flourish when they feel autonomous, that is, when their choices align with their values and reflect genuine self-direction rather than external pressure or coercion.
Self-determination theory proposes that autonomy is not simply freedom from interference; it is the capacity to act in accordance with your own integrated values and preferences. When you choose something that matters to you and understand why you are choosing it, you experience agency. Conversely, when you feel manipulated, confused or uncertain about your own motivations, that sense of agency erodes. This distinction is crucial when thinking about algorithms. An algorithm that recommends something you genuinely want is different from one that exploits your vulnerabilities. The critical question is not whether a choice was influenced, but whether you retained meaningful control over that choice and understood what shaped it.
Choice Architecture and Nudges
The structure of choices shapes decisions profoundly, often more than we realise. Thaler and Sunstein introduced the concept of choice architecture to describe how the presentation of options influences what people select. A nudge is a gentle design choice that steers people toward particular decisions without removing alternatives or changing incentives significantly.
Choice architecture is everywhere in digital life. The order in which search results appear, the default setting on a privacy menu, the prominence of a purchase button, or the auto-play feature that starts the next episode, all nudge behaviour. Crucially, nudges are not inherently problematic. Well-designed systems can nudge people toward decisions that align with their own stated preferences and values. A financial app that nudges you toward saving, or a health platform that reminds you to move, may genuinely serve your interests. However, nudges become ethically murky when they exploit cognitive limitations or emotional vulnerabilities. The difference lies partly in transparency. When you understand how a choice architecture works and agree with its purpose, nudging feels less like manipulation and more like helpful design. When nudges operate invisibly to serve corporate interests rather than yours, agency withers.
Filter Bubbles and the Evidence on Echo Chambers
Eli Pariser popularised the concept of filter bubbles, the idea that personalisation algorithms create information environments tailored so perfectly to individual preferences that people encounter only content matching their existing views. The fear is intuitive: if algorithms show you only what you already like, you will never encounter challenging ideas, and society fragments into isolated ideological silos.
Yet the empirical evidence proves more complicated than the filter bubble thesis suggests. Research examining actual online behaviour reveals that whilst algorithms do personalise content, people encounter considerably more viewpoint diversity online than the filter bubble concept implies. Social media platforms expose people to opposing views more than their offline networks typically do. This does not mean algorithms are neutral; they do amplify certain content and marginalise other perspectives. However, the relationship between algorithmic personalisation and polarisation appears mediated by many other factors, including the choices people actively make, offline influences, and the nature of particular platforms. The evidence suggests we should be cautious about both technological determinism, which assumes algorithms inevitably narrow perspective, and naive optimism, which assumes they have no such effect. The truth is messier: algorithms shape exposure patterns in ways that sometimes narrow horizons and sometimes expand them, depending on design choices and how people use them.
Automation Bias and Over-Reliance on Machines
As artificial intelligence systems improve, they often outperform human judgement in specific domains. Diagnostic algorithms can identify diseases in medical scans; recommendation engines can predict preferences more accurately than explicit user surveys. This capability creates a psychological trap called automation bias. When people believe a system is highly capable, they tend to over-rely on its suggestions and under-weight their own knowledge and intuition, even when the machine's recommendation proves incorrect.
Over-reliance on machine advice can erode meaningful choice-making. If you accept every recommendation a system offers without deliberation, you may cease to understand your own preferences. If a navigation app always chooses your route, you lose knowledge of your environment. If you defer all scheduling to an intelligent assistant, you may become detached from your own time. The irony is that intelligent systems, designed to serve your agency, can paradoxically diminish it through learned helplessness. People begin to defer to machines not because the machines are perfect, but because deferring requires less cognitive effort. This matters because understanding your own choices, even when you ultimately accept a machine's advice, maintains your sense of agency and your capacity to override the system when it matters.
Algorithm Aversion and Appreciation
Psychology research reveals a curious phenomenon called algorithm aversion. When people learn that a recommendation comes from an algorithm rather than a human, they often trust it less, even when the algorithm demonstrably outperforms human judgement. People express scepticism and unease about machine decisions in ways they do not about equivalent human decisions. This aversion is not irrational; it reflects legitimate concerns about opacity, bias and accountability.
However, over-time and with positive experience, algorithm appreciation can develop. When people use algorithmic systems successfully, their scepticism often softens. They become more willing to trust recommendations, particularly when they understand something about how the system works. This suggests that algorithm aversion and appreciation are not fixed psychological traits but responses shaped by familiarity, transparency and track record. The implication for maintaining agency is important: initial scepticism about machine advice is psychologically healthy. It prompts you to question and verify rather than passively accepting. As you gain experience with a system, your trust can grow appropriately if the system proves reliable and behaves predictably. The goal is calibrated trust: neither dismissing algorithms reflexively nor abandoning your own judgement entirely.
Deliberate Control and Digital Literacy
Maintaining agency in an algorithmic environment requires both psychological awareness and practical digital literacy. Psychologically, this means regularly pausing to ask yourself whether a choice you are making genuinely reflects your preferences or whether an algorithm has silently shaped your options. It means recognising that you feel passive or overwhelmed by choices, signs that the choice architecture may be working against your interests.
Practically, digital literacy involves understanding how systems you rely on actually function. This need not mean mastering computer science; it means knowing whether a platform prioritises engagement or accuracy, whether recommendations are personalised or editorial, whether you can inspect why a particular item appeared. It includes knowing how to adjust settings, disable auto-play, clear your history, or switch platforms when one no longer serves you. It means distinguishing between recommendations that helpfully match your stated preferences and those that exploit your vulnerabilities. Perhaps most importantly, maintaining agency means occasionally resisting recommendations deliberately. Clicking on something outside your usual patterns, seeking out diverse perspectives, or choosing to browse rather than search all exercise your capacity to direct your own attention. These acts of deliberate non-compliance with algorithmic suggestions reinforce your sense of control and remind you that the algorithms serve you, not the reverse.
Synthesis: Negotiating Rather Than Surrendering
The relationship between algorithm and agency is not a zero-sum choice between technological determination and human freedom. Instead, it is a negotiation. Algorithms will continue to shape what we encounter and influence what we choose. This is not inherently problematic; much of that influence serves our interests well. The challenge lies in remaining conscious of the influence, understanding what values guide the systems we use, and retaining the capacity to override them when our interests or values demand it.
Agency in a digital age does not mean rejecting algorithmic assistance or pretending it has no effect. It means engaging with these systems as tools rather than authorities, maintaining curiosity about how they work, and periodically reasserting control over your own choices. It means recognising that you have legitimate power: you can choose different platforms, adjust settings, ignore recommendations, and seek diverse information sources. It means understanding that feeling genuinely autonomous is not the same as feeling uninfluenced. You can be influenced by systems designed thoughtfully to serve your interests and still feel that you chose freely, because you retain understanding and control. This is the practical reconciliation of algorithm and agency that allows you to benefit from intelligent systems whilst preserving the sense of self-direction that wellbeing requires.
