Diachronic Identity Problem
Overview
Under CF, we require an algorithmic specification of both which conscious subjects exist at a given moment and whether distinct moments of conscious experience belong to the same subject. This requirement introduces puzzles when algorithms are copied, interrupted or migrated between physical systems.
Consider a conscious computer whose algorithm and internal state are copied onto a second computer while the original continues running. Both initially implement identical computations, yet they are physically distinct and will subsequently diverge. Which, if either, is the original conscious individual? If algorithmic continuity determines identity, both appear equally entitled to that status. If physical continuity determines identity, what makes this criterion part of the algorithmic specification, particularly when computations can be gradually migrated between substrates?
Similar questions arise if an algorithm is paused and restarted a thousand years later, or gradually transferred to another computer while the original continues running. What exactly determines whether a conscious subject persists, ceases to exist, or splits into multiple successors?
Importantly, continuity of personal identity is not the same as uninterrupted continuity of consciousness. We normally regard ourselves as the same individuals after dreamless sleep, despite interruptions in conscious experience. Nor does numerical identity necessarily determine moral welfare: two conscious copies might both deserve moral consideration even if neither is numerically identical to the original.
The challenge for strong CF is to provide principled, substrate-independent criteria for the identity and persistence of conscious subjects that follow directly from the computational nature of consciousness. If these criteria require additional facts about physical instantiation or continuity, in what sense is the conscious individual wholly defined by the algorithm?
Our thanks to Jorge Unzaga and Dylan Roberts, and AI agent teams, for their help developing this entry.
Responses
A computational theory need not identify subjects by algorithms alone. It can specify individual computational executions, their causal histories, and the relationships between them. Identity can therefore be grounded in the complete computational history, without depending on the physical substrate.
POSSIBLY, BUT: This could provide a genuinely computational solution. The difficulty is establishing which relationships constitute the persistence of a conscious subject, rather than simply tracking computational processes. For example, why should one causal continuation preserve identity while two equally valid continuations do not? Could different identity rules agree on every computational fact yet disagree about which conscious subject persists? If so, what makes one rule correct? Strong CF needs to show that its identity criteria are grounded in the nature of consciousness, not merely conventions for describing computation.
There may be no further fact about numerical identity. Following Parfit, what matters is psychological continuity and connectedness, not whether two experiences belong to literally the same individual. CF can specify these relationships computationally without answering an additional metaphysical identity question.
POSSIBLY, BUT: This may successfully dissolve the paradox rather than solve it. It does, however, reject the stronger claim that CF fully determines the identity and persistence of conscious subjects. The theory would specify conscious experiences and their relationships without necessarily determining a unique continuing experiencer. This need not undermine welfare: both successors of a conscious AI might deserve moral consideration. The remaining question is whether such an account is sufficient as a theory of conscious subjects, or only as a theory of individual experiences and their connections.
Following Parfit, numerical personal identity may not be what matters. Psychological continuity can ground survival and moral concern even when identity becomes indeterminate.
POSSIBLY, BUT: This distinguishes the conditions for consciousness from those for personal identity, potentially limiting what strong CF claims to explain. CF needs either an account of these relationships or an explicit acknowledgement that algorithmic consciousness alone does not settle them.
These puzzles are not unique to CF. Biological, physicalist and other theories of consciousness must also explain personal identity through interruption, duplication and change.
AGREED, BUT: Non-CF perspectives may have additional resources for grounding identity, such as biological continuity, spatiotemporal continuity, or the persistence of particular physical structures. These approaches face their own difficulties, especially in duplication and branching cases, and are not automatically superior. However, strong CF deliberately abstracts away from particular physical substrates, leaving it to explain how individual identity and persistence can be grounded in computational properties and relationships alone.
Further reading
- Parfit D (1984). Reasons and Persons
- Olson ET (2023). Personal Identity
Do you find this argument strong or weak?