In many complex systems operating across boundaries, equivalence is declared rather than discovered. However, declared equivalence isn’t necessarily true equivalence; it is often an operational necessity because systems have to treat different things as identical, or at least sufficiently similar, in order to function.

Comparing meaning properly is expensive, slow and labour-intensive. Systems that need to coordinate across languages, jurisdictions or categories can’t wait for perfect alignment, so differences are grouped, categories are created and things that are not quite the same are treated as if they were.

Law provides a good example. In multilingual legal systems, different language versions may be accorded equal legal status even though perfect semantic alignment between them cannot be guaranteed. Divergence is dealt with later through interpretation. Declared equivalence therefore provides institutional stability even where semantic or conceptual equivalence is incomplete.

A similar issue arises in legal AI. Systems designed to operate across jurisdictions need some way of connecting concepts from different legal systems. Terms such as consideration, good faith or liquidated damages may have recognisable counterparts elsewhere, but those counterparts do not necessarily have the same doctrinal foundations, scope or legal consequences. Treating them as equivalent makes large-scale processing possible, even where the underlying legal concepts do not align completely.

This matters because decisions about equivalence have downstream consequences. Once two concepts have been grouped together in a taxonomy or dataset, later stages of the system inherit that assumption. A model can consequently generate fluent and apparently coherent output without having access to distinctions that were removed before the information ever reached it.

Some degree of abstraction is unavoidable. A cross-border system that treated every legal concept as entirely unique would struggle to compare, classify or reason across jurisdictions at scale. The more important questions are therefore how equivalence should be determined, which differences need to be preserved and which can safely be abstracted away.

These are not purely technical decisions. They require legal and comparative judgement about how concepts function within their respective systems and about the consequences of treating them as the same. As legal AI becomes increasingly international, the quality of cross-border output will depend partly on how carefully these decisions are made upstream.