A readable answer does not necessarily expose the assumptions used to form it. In a theoretical communication paper, Maxim Osovsky proposes five levels for examining the relationship between a model's message and the representations behind that message.
Distinguish the situation from its description
The levels cover systemic representations of the world of activity, a particular situation selected from those representations, its narrative description, the full sense of a message and the text itself. In this account, a situation's schema-model is different from a story told about it.
Osovsky argues that context and trained language-model weights do not, by themselves, attest to deliberately constructed systemic representations. This is the paper's theoretical position about how message foundations should be organized and examined, rather than a benchmark result proving a universal limit of current models.
Work backwards from the message
The proposed reverse process moves from a message's full sense through an extended situation description and schema-model toward the systemic foundations. The recipient reconstructs meaning through shared representations and joint activity, rather than treating the visible text as the complete account of what formed it.
The paper draws on Shchedrovitskiy's systems methodology, Korzybski's structural differential and Mel'chuk's Meaning-Text theory. These are the stated theoretical foundations of the architecture. They should remain distinct from evidence that an implemented agent can perform the proposed reconstruction reliably.
Read it as a research agenda
Further work concerns large graphical models of activity with reconstructible foundations. The title's reference to strong language-based ASI describes the proposed setting; it does not establish that the paper has built or demonstrated such a system.
A useful way to examine the proposal is to ask what representation would be inspectable, how participants would agree on it and what would count as a successful reconstruction. Those questions help distinguish a conceptual organization of communication from a tested engineering method. The paper offers the former, with implementation and evaluation left as research work.
Comments