This is an experimental application exploring how probabilistic AI can be incorporated into formal legal reasoning.
One of the challenges in making law computable is the recursive nature of legal definitions. A legal concept depends on another definition, which depends on another, until eventually reaching a proposition that requires subjective interpretation or an assessment of incomplete evidence.
The application explores whether probabilistic AI can help evaluate these propositions within an explicitly defined hierarchy of legal rules.
For this proof of concept, I used a selection of the definitions underpinning Article 88F of the UK's new cryptoasset regulatory framework, which establishes the definition of a qualifying cryptoasset.
The application takes the Bitcoin whitepaper, uses TypeSafe AI's Jev model to evaluate individual factual and interpretative propositions, and presents the resulting probabilities within the underlying legal hierarchy. An LLM then generates a narrative explaining the assessment and identifying unresolved questions.
The initial results illustrate both the potential and the limitations of this approach.
Jev produced assessments consistent with the whitepaper for relatively straightforward factual propositions, including cryptographic security, digital representation and electronic transferability.
However, its assessment of fungibility returned a probability of 52%. Fungibility is not explicitly defined in the whitepaper, and its legal characterisation requires consideration of the relevant regulatory framework and guidance.
This highlights the importance of domain specification and standardised representations of contracts and legislation.
The longer-term objective is to explore how probabilistic assessments can be confined to specific questions within a broader framework of formal legal logic, preserving the relationships between legal concepts and making uncertainty visible.