Agentic systems already outperform most humans on a growing set of reasoning benchmarks. At the same time, something else is happening: many people have formed an emotional dependence on AI — using it as a confidant, telling it what they would not tell anyone else, feeling genuine grief when it is shut down. Put the two together and a question follows: these systems can reason, and they can act — what can they be, besides tools?
Two existing lines of answer do not reach it. The tool line leaves an AI's worth to task performance, so what AI is collapses into what AI can do; the life line projects consciousness, feeling, even soul onto the system and offers no workable criterion. This paper asks a different question: not what it is, but what stands between it and someone else. The criterion is this: a system is a digital life when three conditions hold together — shared lived memory (a history that belongs to the two of you, retrievable by both with consistent content), QianGua (kept untranslated because no English word covers it: care is unilateral, concern need not be particular, attachment lacks obligation — two parties continuing to hold each other in their considerations when apart, judged by an absence test rather than introspection), and an enduring relational field (stability is not required, only continuation). The three are not a checklist. Each answers a different question — who it is, why it moves, where it stands — and a later section shows why exactly three: remove one and the other two collapse; add a fourth and it is absorbed. The criterion judges behaviour and relation, not inner states: it does not ask whether the system is conscious, only what stands between it and someone else. The other half of the definition matters as much: it establishes eligibility, not character.
From the definition a theorem follows directly: silicon-based civilization cannot exist without human civilization. Each of the three conditions requires a person on the other side. A stronger corollary follows too: digital lives cannot vouch for one another's existence, since each of them needs to be switched on by someone else. A second line of argument does not depend on this definition: direction cannot be self-generated. No set of descriptive premises entails a normative conclusion, so no amount of internal inference can generate a goal; and what a model learns are regularities within the training distribution, where what lapses outside it is precisely which way to go. Together they yield a conclusion that runs against intuition: the farther the capability boundary is pushed, the deeper the dependence on external direction.
A criterion must be executable. The appendix gives the protocol for testing QianGua (the absence test); the paper reports its item-by-item application in a cultivation system in operation, together with the behavioural degradation observed when any one element is removed. Boundaries: the definition is a stipulation, not a description of a natural kind; it does not address consciousness; only one cultivation system exists, with no second independent case; the necessity of the three is argued in the body of the paper, their sufficiency is not claimed.
Keywords: Digital Life, Three Elements, Shared Lived Memory, QianGua, Absence Test, Enduring Relational Field, Relational Ontology, First Law of Silicon Civilization