The Metaverse and AI: A World That Adapts to You, So What Does It Know About You?

On 28 October 2021 Facebook changed its name to Meta and said its focus would be to "bring the metaverse to life". Four years later the company reported that its Reality Labs unit, responsible for virtual and augmented reality devices, ended 2025 with revenue of about $2.2 billion and an operating loss of about $19.2 billion. Is the idea dead?

A Promise That Outran Reality

In 2022 McKinsey published an estimate that the metaverse could generate up to $5 trillion in value by 2030. That is the estimate of a consultancy, published months after Meta's announcement, not a measured fact. As for the unit concerned, its annual revenue is about $2.2 billion against an operating loss roughly nine times as large. These figures do not judge the whole idea, since they concern one unit in one company, but they show that large funding does not create demand by itself.

What Does AI Add?

The shortcoming of these worlds was that they were silent: characters repeating pre-programmed scenarios. The "Generative Agents" paper (Park and others, 2023) shows a different direction: the researchers placed 25 software agents in an experimental world resembling the game The Sims, each with a memory in natural language that it summarises and retrieves in order to plan. When given one goal, to hold a party, the agents spread the invitations, got acquainted and arrived at the appointed time. The experiments showed that observation, planning and reflection each contribute to the believability of the behaviour.

But this is a simulation sandbox, not an immersive virtual reality, and what appears in it is convincing behaviour, not consciousness or understanding in the human sense (see the article "What Is Artificial Intelligence, Really? A Plain Explanation Without the Hype"). That is why it matters that users know when they are addressing an algorithm.

What Does the World Know About You?

The more the world adapts to you, the more it reads of you. In a research study (Nair and others, 2023), real users, 55,541 of them, were identified across several sessions from the movement of their heads and hands alone: accuracy reached 94.33% from 100 seconds of movement and 73.20% from 10 seconds. Your body, in other words, signs your movement, even without a name or a face. The body yields sensitive data that should not be collected or analysed without explicit and justified permission (see the article "Privacy in the Age of AI").

A Real Benefit in a Specific Field

Virtual reality has a relatively established use in exposure-based psychotherapy. A research review (Tsamitros, 2023) concludes that virtual-reality exposure therapy appears as effective as conventional therapy for specific phobias and for agoraphobia with panic disorder, and comparable to an active psychotherapy for post-traumatic stress disorder, with more research still needed. This is done under specialist supervision, not through an unregulated game. Institutions considering immersive worlds would do well to begin with a specific problem, such as training a skill or treating a phobia, and measure its effect, rather than building a "world" and then looking for users. A small measured trial is more honest than a large promise.

What Do We Ask For?

• Transparency: that users know when they are addressing an algorithm, and what is being collected about them.
• Ownership of body data: that body movement not be sold or used for anything beyond what was permitted.
• Protection of the weaker groups, foremost the young, and an easy way to opt out.
• Governance in which whoever designs and operates the world is held accountable.

The measure by which we weigh this technology is what it adds to people's benefit and what it preserves of their dignity and trust, not the spectacle it offers. A world that knows you well ought to take good care of what it has learned.