By Heidi Lund, a Senior Adviser at the National Board of Trade Sweden and author of the study Collective Societal Risks: A Blind Spot in AI Governance which examines how AI-driven collective societal risks may affect consumers, market access and international trade.
As AI systems increasingly influence what consumers see, trust and choose, concerns about misinformation, bias and privacy dominate public debate. Yet the most important AI risk may be one that receives far less attention: its growing power to shape markets by determining what people discover, consider and ultimately choose.
Artificial intelligence is usually portrayed as a force for innovation, productivity and economic growth. Governments are racing to promote it. Businesses are rushing to adopt it. Regulators are trying to ensure it is safe.
Most concerns focus on misinformation, bias, privacy and other harms affecting individuals. But one potentially more consequential risk has received far less attention: what happens when millions of people are simultaneously influenced by the same AI systems.
The phenomenon is already familiar to most internet users.
One month social media feeds are filled with advice about “zone 2 training”. The next, high-protein diets dominate recommendations. New phrases and ideas suddenly appear across platforms, articles and videos. Certain products become ubiquitous seemingly overnight.
These trends often appear organic. Yet increasingly they are amplified by AI systems designed to maximise engagement, relevance and attention.
Recommendation systems are among the most influential forms of artificial intelligence in everyday life. They determine what users see on social media, what films are suggested on streaming platforms and which products appear on online marketplaces.
These systems do not simply reflect preferences. They help shape them.
Recommendation systems do not merely respond to users’ interests. By repeatedly prioritising certain information, products, viewpoints and behaviours, they can gradually influence what people pay attention to and engage with. Over time, repeated exposure can shape habits, preferences and decisions both online and offline.
What becomes visible attracts attention. What attracts attention becomes influential.
For any individual user, the effect may seem insignificant. But when the same mechanisms influence hundreds of millions of people simultaneously, their effects need not remain confined to digital platforms. They can shape information environments, behaviours and decisions in the wider world, ultimately influencing markets and economic activity.
Large language models introduce a related challenge.
Unlike recommendation systems, which select and prioritise existing content, large language models generate new content. Millions of people now rely on AI assistants to answer questions, compare alternatives, summarise information and support decision-making.
Public debate often focuses on hallucinations and factual inaccuracies. An equally important question is what happens when large numbers of people increasingly rely on systems trained on similar data sources and optimised according to similar objectives.
The concern is not necessarily that the answers are wrong. It is that they may gradually shape what people regard as relevant, trustworthy or worth considering.
Importantly, this is not about any particular company or AI model. The issue is systemic. Similar dynamics can emerge whenever large populations rely on the same kinds of AI systems for information, recommendations and decisions.
These effects can be understood through three broad dimensions. Informational risks affect what information becomes visible and what remains unseen. Cognitive risks influence how people interpret information and form opinions. Behavioural risks affect choices by guiding attention and prioritising options. Taken together, these effects raise a question that current AI governance frameworks rarely address: what happens when AI systems influence societies collectively rather than individuals one by one?
The answer matters because AI is increasingly becoming an intermediary between people, information and markets.
Traditionally, businesses competed through price, quality, innovation and reputation. But as AI systems increasingly influence what consumers see, trust and choose, competition may gradually shift towards visibility and discoverability.
In such a world, success depends not only on offering the best product or service, but also on being the product or service that consumers encounter in the first place.
Consider a consumer asking an AI assistant to identify the best supplier of solar panels, recommend accounting software or compare logistics providers.
The answer generated by the AI may reflect training data, popularity signals, optimisation objectives and countless technical design choices that remain invisible to users.
When large numbers of users increasingly rely on AI assistants to identify, compare and evaluate available options, the systems themselves become important intermediaries between consumers and markets. Unlike traditional search engines, AI assistants often present users with synthesised recommendations rather than a broad list of alternatives. In such circumstances, even small and individually unobjectionable patterns in AI-generated recommendations may, through repetition and scale, influence which firms, products and services become visible and discoverable.
None of this necessarily involves intentional discrimination.
Yet visibility can become self-reinforcing. Firms that receive greater exposure attract more customers. More customers generate more data and stronger digital signals. Those signals can further increase visibility.
Large firms often possess advantages in precisely the areas that matter most in digital environments: data resources, brand recognition, online presence and participation in digital ecosystems. Smaller firms may therefore become increasingly difficult to discover even when they offer competitive products and services.
The concern is not that AI companies are deliberately favouring particular firms, industries or countries. Structurally unequal outcomes can emerge even without intentional bias.
This possibility sits uneasily within existing regulatory frameworks.
Competition law focuses on market power and anti-competitive conduct. Digital regulation focuses on platform risks and illegal content. AI regulation focuses largely on identifiable harms arising from specific systems.
Yet the effects described here emerge through the interaction of many lawful systems operating simultaneously. No single recommendation, ranking or AI-generated answer appears particularly significant. The impact becomes visible through repetition, scale and accumulation.
The challenge becomes even more important as AI increasingly resembles critical infrastructure.
Developing advanced AI systems requires enormous amounts of data, computing power, specialised semiconductors and capital. As a result, much of the world’s AI infrastructure is concentrated among a relatively small number of firms and technology ecosystems.
As AI becomes embedded in information flows, consumer choices and economic activity, these concentrations may create new forms of dependency. The issue is no longer only technological. It concerns who controls the systems through which information is discovered, decisions are made and economic opportunities are distributed.
Historically, societies depended on railways, ports, telecommunications networks and electricity grids. Increasingly, they may also depend on AI systems.
The difference is that AI does not merely move information. It increasingly influences how people understand the world, make decisions and navigate markets.
For years policymakers have asked whether AI can influence people. They may increasingly need to ask how AI influences markets.
When visibility, trust and discoverability become mediated through AI systems, the question is no longer simply technological. It becomes a question about who gets noticed, who gets chosen and ultimately which firms, products and suppliers gain access to markets.
If these collective effects remain outside existing governance frameworks, the most important AI risks of the coming decade may not be technological failures. They may be the invisible consequences of success.
The views expressed in this article are solely those of the author and do not necessarily reflect the views of the National Board of Trade Sweden or the Swedish government.