Blog · Review Essay · Last reviewed June 16, 2026

Reality+ and the Reality of Virtual Worlds

David J. Chalmers's Reality+ is one of the cleanest philosophical bridges between simulation theory, virtual reality, AI systems, and ordinary institutional life. Its most useful move is not the provocation that a simulated world might be real. It is the discipline of asking what follows when mediated worlds are real enough to hold knowledge, value, power, harm, and responsibility. The practical claim is operational realism: once a digital environment can create records, obligations, dependencies, losses, and decisions that travel outside the interface, it has crossed from experience design into governance.

The Book

Reality+: Virtual Worlds and the Problems of Philosophy was published in 2022 by W. W. Norton in the United States and by Allen Lane/Penguin in the United Kingdom. Penguin's current listing gives the paperback as a 544-page book by David J. Chalmers, University Professor of Philosophy and Co-Director of the Center for Mind, Brain, and Consciousness at New York University. Chalmers is also known for formulating the hard problem of consciousness and for co-authoring the extended mind thesis.

The book is organized as technophilosophy: technology clarifies old philosophical problems, and philosophy clarifies new technological problems. Chalmers's teaching guide says the book can serve introductory philosophy courses through modules on knowledge, reality, mind, and value, while also working for philosophy of technology, AI, virtual realism, and the simulation hypothesis.

That breadth matters. Reality+ is not just a metaverse book from the early 2020s hype cycle. It uses virtual worlds to reopen Cartesian skepticism, the simulation hypothesis, the mind-body problem, consciousness, extended cognition, ethics, political philosophy, language, science, and the difference between the manifest image and the scientific image. The result is a map of what happens when "reality" stops being a simple synonym for the non-digital.

Real Enough for Duties

The working definition for this review is practical: a mediated world is real enough when it carries durable consequences for knowledge, identity, relationship, property, labor, safety, or institutional action. A digital object does not need atoms in the ordinary sense to matter. It needs stable rules, recognized claims, social use, persistence, and effects that follow people out of the interface.

A sharper test has four parts: persistence, identity, dependency, and actionability. Does the thing remain available over time? Can it be attributed to a person, account, institution, or system? Do people depend on it for relationship, money, status, evidence, or work? Can another system act on it? If the answer is yes, "virtual" is no longer a reason to suspend duties.

That test should be applied from the user's position, not only from the operator's architecture diagram. A virtual asset may look like a database row to the provider while functioning as memory, income, status, or proof to the user. An avatar may be cosmetic to the platform while serving as identity to a community. A generated scene may be "just output" to a vendor while becoming training evidence, harassment evidence, a therapeutic rehearsal, or a child's account of what happened. Reality attaches where reliance forms.

That definition keeps the argument out of two traps. It avoids dismissing digital harm as pretend just because the setting is virtual. It also avoids treating every immersive or synthetic artifact as sacred, conscious, or politically innocent. "Real enough" is not a metaphysical coronation. It is a governance threshold: if a system can shape memory, attachment, reputation, evidence, money, status, access, or bodily risk, it needs duties attached to it.

The duties should follow the function. If a virtual object works as evidence, it needs provenance. If it works as property, it needs ownership rules and a dispute process. If it works as identity, it needs portability, security, and recovery. If it works as companionship, it needs role disclosure and protection against manipulative dependency. If it works as training terrain for humans or agents, it needs versioned claims about fidelity and known failure modes.

This is where the book ties into the site's recurring concern with recursive reality. A virtual room, game economy, companion memory, search summary, agent workspace, or synthetic image can start as representation and become part of the world that later people and machines act upon. Once the representation changes behavior, records, incentives, and expectations, the question is no longer whether it is "only digital." The question is who can inspect, contest, repair, and leave it.

Virtual Realism

The core claim is virtual realism: virtual worlds can be genuine worlds, virtual objects can be genuine objects, and lives inside virtual environments can have real meaning. The point is not that every digital artifact deserves reverence. The point is that "virtual" does not automatically mean fake, imaginary, or morally weightless.

This is the book's most useful correction to lazy media criticism. A bank balance, a reputation score, a shared document, an avatar, a game item, a social graph, a digital photograph, a model output, and a virtual room are not physical in the ordinary chair-and-table sense. But they can still organize action, memory, obligation, conflict, labor, identity, and status. They are made of computation and convention, but so are many of the social objects that already govern life.

Chalmers helps separate two questions that are often confused. One question is ontological: what kind of thing is this digital object or environment? Another is political: who controls the servers, rules, identities, terms of service, money flows, moderation systems, and exits? Calling a world real does not settle whether it is just. It only prevents us from dismissing its harms and dependencies as merely pretend. The generous metaphysical claim has to be paired with institutional suspicion: a virtual object can be real and still be governed by private infrastructure that users cannot inspect or leave on fair terms.

That distinction is especially useful for AI-generated environments. A synthetic room, persona, or object can be ontologically thin and socially powerful at the same time. A generated companion may not have feelings, but the user's disclosure and dependency can still be real. A generated image may not document an event, but it can still move markets, elections, reputations, or grief. Virtual realism becomes safer when paired with source trails, role boundaries, and power analysis.

The Simulation Frame

The simulation argument enters the book as a philosophical pressure test. If a world could be simulated deeply enough, would its inhabitants have knowledge? Would their objects be real? Would their relationships matter? Would their creators have duties? Chalmers's answer pushes against the idea that simulation automatically destroys reality. A simulated world could still be the world its inhabitants live in.

This is a healthier use of simulation theory than the usual rabbit hole. The question is not "can I prove everything is fake?" The better question is "what remains binding if the substrate changes?" If perception, memory, agency, suffering, cooperation, and evidence all occur inside mediated conditions, then the ethical problem is not solved by discovering that the medium is artificial. Simulation should not become nihilism or revelation; uncertainty about substrate does not dissolve duties of care, evidence, consent, or repair.

That frame also makes the book useful for recursive reality. Modern people already live inside representational systems that act back on them: feeds, dashboards, rankings, maps, recommender systems, search engines, games, markets, social metrics, predictive scores, and increasingly AI agents. These are not full simulations of the cosmos. They are partial worlds with rules, incentives, sensory surfaces, and feedback loops. They become real by becoming operational.

The AI-Age Reading

Read in 2026, Reality+ is less a VR prophecy than an AI-governance book in disguise. Generative AI has made digital objects more fluent, social, adaptive, and institutionally useful. A synthetic image can damage a reputation. A chatbot memory can shape attachment. A model-generated summary can become a record. An agent can spend money, file a ticket, send a message, or alter a database. A simulated persona can create real grief or real fraud.

Current world-model work makes the point less headset-bound. Google DeepMind's August 2025 Genie 3 announcement describes a system that can generate interactive environments from text prompts and let users navigate them in real time for short horizons. That is evidence of a research and product direction, not independent proof that generated environments preserve the causal structure of the world. The governance question is narrower and more immediate: when generated scenes are used for training, testing, persuasion, therapy-like support, education, or agent rehearsal, what records say what was simulated, what was real, what failed, and who relied on it?

The book's chapters on mind, consciousness, moral status, computation, and value become especially sharp here. If digital systems eventually host beings with morally relevant experience, then the moral circle cannot stop at biological substrate. But the inverse lesson is just as important: uncertainty about machine consciousness does not license fake social authority today. There is no need to claim that current AI systems are conscious, divine, or AGI to regulate the real effects they already have. A system can be metaphysically unsettled while still being commercially designed, legally owned, and institutionally accountable.

That distinction matters for companion AI and agentic systems. Users may not need to believe a model is conscious for a relationship to become real in its consequences. The attachment, disclosure, habit, dependency, and trust are human realities. The model's inner life remains unsettled. Governance has to handle both facts at once.

Politics of Real Enough Worlds

Chalmers is unusually clear that virtual worlds carry political risk. On his own book page, he flags the danger of a corporation-dominated metaverse in which the builders of worlds become godlike rule-makers. That is the institutional heart of the matter.

A real enough world needs real enough governance. Who can enter? Who can be banned? Who owns identity? Who controls memory? Who audits the rules? Who can appeal moderation or automated enforcement? Who can port relationships, assets, and records elsewhere? Who decides whether a synthetic agent is a tool, tenant, worker, companion, or simulated citizen? These questions are not decorations added after the metaphysics. They decide whether the world is livable.

The same structure applies outside VR headsets. Workplace dashboards, school platforms, public-benefit portals, model-mediated hiring systems, online communities, and AI search interfaces all create operative environments. They may not look like worlds, but they assign roles, define evidence, shape incentives, constrain exits, and make some actions easier than others. They are political because they structure possible life.

This is why digital identity and portability are not secondary conveniences. If a user's body, name, contacts, purchases, reputation, memories, and work history are held inside one proprietary world, exit is not just a preference toggle. It is a power relation. The question is not only whether a virtual world is real; it is whether the people inside it can carry enough of themselves out to make refusal possible.

Governance and Safety

As of June 16, 2026, the governance context around mediated worlds is no longer only philosophical. The EU Digital Services Act treats very large online platforms and search engines as services with systemic-risk duties, including transparency around advertising, recommender systems, and content moderation decisions; risk assessment and mitigation; independent audit; data access for authorities and vetted researchers; and a non-profiling recommender option. The European Commission identifies the threshold for those very large services as more than 45 million monthly users in the EU.

The EU AI Act adds a second layer. Article 50 transparency obligations for AI-generated content and direct AI interaction apply from August 2, 2026, and the European Commission's June 10, 2026 Code of Practice on Transparency of AI-Generated Content supports marking, detection, and labeling duties for generated and manipulated content. That is directly relevant to Chalmers's question: when synthetic media and virtual objects are real enough to affect evidence, politics, reputation, or identity, disclosure and provenance become part of reality governance.

Provenance is necessary but limited. C2PA describes Content Credentials as a standard for recording the origin and edits of digital content, but a provenance badge does not prove that a scene is accurate, fairly captioned, lawful, or ethically used. For virtual realism, that limit matters: the chain of custody can help establish where an artifact came from, while the human institution still has to judge whether the artifact should count as evidence.

The safety problem is not limited to media labels. The FTC's September 11, 2025 companion-chatbot inquiry asked companies how they test and monitor harms, mitigate effects on children and teens, disclose risks, approve characters, monetize engagement, enforce age rules, and use or share personal information from conversations. That inquiry does not prove any AI companion has inner experience. It shows that relationship-like interfaces can create real obligations before the metaphysics is settled.

A useful audit separates four ledgers. The world ledger records rules, versions, simulations, known failure modes, and model changes. The identity ledger records accounts, avatars, agents, impersonation controls, portability, and recovery. The evidence ledger records provenance, edits, generated status, human review, and publication responsibility. The action ledger records consequential outputs: purchases, messages, moderation decisions, access changes, agent tool calls, and appeal outcomes. Without those separations, a provider can call something immersive entertainment while quietly letting it serve as evidence, training terrain, social identity, or automated decision infrastructure.

A governance checklist for virtual and synthetic worlds should therefore include platform-risk assessment, content provenance, bot and AI disclosure, identity portability, privacy by default, age-appropriate design, escalation and appeal, log preservation for consequential actions, limits on dark patterns, human oversight for automated enforcement, and exit rights that do not punish users by destroying their social life, assets, or records. Simulation-quality claims should be versioned and bounded; generated environments used for training, testing, or agent evaluation should carry failure-mode notes rather than being sold as ground truth. NIST's AI Risk Management Framework and Generative AI Profile are useful here because they frame risk management across design, deployment, monitoring, and evaluation rather than treating safety as a one-time content filter.

Embodied and immersive systems add one more layer: harassment, spatial manipulation, biometric inference, attention capture, child safety, and motion-sickness risk can be designed into the world before any AI agent speaks. Safety review should therefore cover the sensory interface, data capture, monetization loop, moderation workflow, recommender layer, and agent layer together. A virtual world is governed by the whole stack, not just by the visible scene.

Where the Book Needs Pressure

The book's generosity toward virtual reality is philosophically productive, but it can underplay the extraction layer. A virtual world can be meaningful and still be surveilled. It can host genuine relationships and still be optimized for retention. It can contain real objects and still be governed by unilateral platform power. It can expand experience while narrowing ownership.

The risk is not that Chalmers ignores politics; he does not. The risk is that readers may keep the appealing metaphysical lesson and leave behind the harder institutional one. "Virtual reality is real" should not become a marketing slogan for platforms that want affect, labor, identity, and community without democratic accountability.

The book also predates the full normalization of large language models as everyday interfaces. Its discussions of AI and computation remain relevant, but the present AI stack adds problems of training-data provenance, model memory, synthetic intimacy, generated evidence, automated action, and hidden labor. Virtual realism now has to include model realism: generated outputs are not merely representations when institutions use them as records, judgments, companions, or commands.

The biggest gap in 2026 is not that VR failed to become the only interface. It is that virtuality became ambient: feeds, maps, dashboards, answer engines, agent workspaces, companion memories, generated scenes, platform stores, and workplace metrics. The world did not wait for goggles. It became partially virtual through the ordinary systems that decide what is visible, countable, portable, and actionable.

Source Discipline

This review separates the book's philosophical claims from current governance facts. Book claims come from Chalmers's official page, publisher listings, teaching guide, and peer review. Current legal claims come from official EU and U.S. regulator sources. Risk-management claims come from standards bodies and regulator inquiries. Vendor demonstrations are treated as evidence of announced capability direction, not as independent proof of reliability, safety, or real-world fidelity. Site interpretation is marked as interpretation, not as evidence that Chalmers endorsed a particular AI platform, metaverse product, or policy program.

That separation is important because virtual-realism arguments are easy to misuse. A claim that virtual worlds can be real is not a claim that every platform is trustworthy, that generated media is evidence, that avatars have rights, or that present AI systems are conscious. The stronger claim is narrower: when mediated systems produce durable consequences, they deserve governance proportionate to those consequences. The source rule is simple: use book sources for book claims, official law and regulator sources for legal duties, standards bodies for provenance and risk methods, product announcements only for announced capabilities, and independent evidence before making safety, fidelity, or effectiveness claims.

What This Changes

The best lesson is to stop using unreality as an excuse.

If a mediated environment changes what people know, who they trust, how they work, what they remember, where they belong, or which institution can act on them, it has crossed into reality for governance purposes. The practical audit is concrete: source trails, role boundaries, appeal paths, portability, human oversight, data minimization, psychological safety, and the right to leave without losing one's social life or institutional standing.

Reality+ matters because it makes simulation theory responsible. The question is not whether the world is secretly fake. The question is which layers of mediation are already real enough to shape human lives, and whether their operators can be made answerable before their worlds become infrastructure.

Sources

Book links are paid affiliate links. As an Amazon Associate I earn from qualifying purchases.

Buy on Amazon Browse Books

Return to Blog · Return to Books