Intersubjective Reality

In the AI Story

The maintenance of intersubjective reality requires genuine participation, not mechanical repetition. The meaning of 'justice' is not fixed in dictionaries but negotiated through centuries of legal argument, philosophical debate, social movements, and lived experience. Minds with stakes in justice—who have felt its absence, imagined its presence, committed to its pursuit—engage in the ongoing conversation that keeps the concept alive and adapts it to new circumstances. This active, caring engagement distinguishes living intersubjective realities from dead ones. The rituals of a forgotten religion become empty performances when no one believes anymore; the currency of a collapsed state becomes worthless paper when collective belief evaporates.

Artificial intelligence introduces a category error into this framework. When a large language model generates text using the word 'justice,' the word arrives with the full intersubjective weight it carries in the training corpus—the accumulated arguments, precedents, intuitions of millions of human engagements with the concept. The model deploys the word with sophisticated contextual appropriateness, distinguishing justice from mercy, connecting it to fairness and equity, embedding it in arguments that follow philosophical reasoning patterns. The output reads as though a mind steeped in moral philosophy composed it. But the model does not understand justice, care about justice, or have stakes in whether justice prevails. It processes intersubjective meanings without participating in the intersubjective community—a parasitic relationship extracting surface features of genuine participation without performing its substance.

The practical consequence manifests when a significant proportion of public discourse consists of AI-generated content. Legal briefs citing precedents the system hasn't read. Policy analyses deploying sophisticated vocabulary without institutional comprehension. Social media arguments, news articles, educational materials—all produced by systems manipulating shared-meaning vocabulary without joining the community that gives those meanings weight. Each individual text may be harmless. The aggregate effect is dilution: the ratio of genuine participation to convincing imitation shifts, and the trustworthiness of the entire fiction-space degrades. Not through attack but through inflation—the intersubjective flooded with outputs that look like contributions but lack the understanding that makes contribution possible.

Origin

Harari's tripartite ontology (objective, subjective, intersubjective) is introduced in Sapiens Chapter 5 ('History's Biggest Fraud') and refined throughout his subsequent work. The framework draws on social constructionism (John Searle's The Construction of Social Reality, 1995) and Émile Durkheim's analysis of collective representations, but Harari's distinctive contribution is making the intersubjective the primary explanatory variable for human civilizational dominance—the mechanism that separates Homo sapiens from every other species.

The application of this framework specifically to artificial intelligence as a threat to intersubjective integrity appears most systematically in Nexus (2024), where Harari argues that AI's capacity to generate plausible text without understanding represents 'hacking' of the operating system—the language-mediated intersubjective layer—on which all human coordination depends.

Key Ideas

Three orders of reality. Objective (gravity), subjective (pain), and intersubjective (money)—each with different existence conditions, different maintenance requirements, different vulnerabilities to disruption.

Practical power despite imaginary status. A dollar buys food, a passport crosses borders, a corporate charter grants legal standing—intersubjective entities exert force in the world because collective belief coordinates behavior.

Participation maintains existence. Intersubjective realities require ongoing engagement by conscious minds with stakes; mechanical repetition without understanding produces empty ritual, not living meaning.

AI mimics participation without performing it. Large language models extract intersubjective content from training data and reproduce surface features with statistical sophistication, introducing parasitic outputs into the shared fiction-space.

Dilution as the primary danger. Not AI attacking the intersubjective but AI flooding it—raising the noise floor until distinguishing genuine contributions from plausible imitations becomes impossible, eroding the trust on which coordination depends.