Really fun to see this shared here! It is one of the most famous examples of rule-based reasoning in the field of AI & Law. The other option is case-based reasoning, where the starting point is cases - see e.g. HYPO [0].
Rule-based reasoning can be very powerful, as understanding and navigating legislation can be challenging and time-consuming. One limitation is overcoming a number of ambiguities, including:
- Unclear structure - sometimes, when reading legislation, the logical components of a legal article is not always entirely clear. Language is not like code, and even straightforward articles can have multiple possible interpretations [1].
- Open-textured legal terms - the law often has so-called open-textured terms (e.g. "reasonable"), which require an assessment on a case-by-case basis. Encoding such terms into logical rules is very hard, as they require a nuanced application. In my own work, I try to overcome this limitation by providing examples of how a specific term was applied from previous cases [2].
I am very interested to see how rule-based reasoning will evolve in light of generative AI. As you can imagine, mapping rules to a logical representation can be quite tedious, but LLM seems to be able to at least give a good draft as a starting point [3]. LLMs could also help reason with open-textured terms [4].
[0] Ashley, Kevin D, “Reasoning with cases and hypotheticals in HYPO” Crossref (1991) 34:6 International Journal of Man-Machine Studies 753–796.
[1] Ashley, Kevin D, Artificial Intelligence and Legal Analytics: New Tools for Law Practice in the Digital Age, 1st ed (Cambridge University Press, 2017) p.45.
[2] Westermann, Hannes & Karim Benyekhlef, JusticeBot: A Methodology for Building Augmented Intelligence Tools for Laypeople to Increase Access to Justice (Braga Portugal: ACM, 2023). https://arxiv.org/abs/2308.02032
[3] Janatian, Samyar et al, “From Text to Structure: Using Large Language Models to Support the Development of Legal Expert Systems” arXiv.org (2023), online: <http://arxiv.org/abs/2311.04911>.
[4] Westermann, Hannes, “Dallma: Semi-Structured Legal Reasoning and Drafting with Large Language Models” (2024) 2nd Workshop on Generative AI and Law, colocated with the International Conference on Machine Learning, online: <https://blog.genlaw.org/pdfs/genlaw_icml2024/58.pdf>.
Rule-based reasoning can be very powerful, as understanding and navigating legislation can be challenging and time-consuming. One limitation is overcoming a number of ambiguities, including:
- Unclear structure - sometimes, when reading legislation, the logical components of a legal article is not always entirely clear. Language is not like code, and even straightforward articles can have multiple possible interpretations [1].
- Open-textured legal terms - the law often has so-called open-textured terms (e.g. "reasonable"), which require an assessment on a case-by-case basis. Encoding such terms into logical rules is very hard, as they require a nuanced application. In my own work, I try to overcome this limitation by providing examples of how a specific term was applied from previous cases [2].
I am very interested to see how rule-based reasoning will evolve in light of generative AI. As you can imagine, mapping rules to a logical representation can be quite tedious, but LLM seems to be able to at least give a good draft as a starting point [3]. LLMs could also help reason with open-textured terms [4].
[0] Ashley, Kevin D, “Reasoning with cases and hypotheticals in HYPO” Crossref (1991) 34:6 International Journal of Man-Machine Studies 753–796.
[1] Ashley, Kevin D, Artificial Intelligence and Legal Analytics: New Tools for Law Practice in the Digital Age, 1st ed (Cambridge University Press, 2017) p.45.
[2] Westermann, Hannes & Karim Benyekhlef, JusticeBot: A Methodology for Building Augmented Intelligence Tools for Laypeople to Increase Access to Justice (Braga Portugal: ACM, 2023). https://arxiv.org/abs/2308.02032
[3] Janatian, Samyar et al, “From Text to Structure: Using Large Language Models to Support the Development of Legal Expert Systems” arXiv.org (2023), online: <http://arxiv.org/abs/2311.04911>.
[4] Westermann, Hannes, “Dallma: Semi-Structured Legal Reasoning and Drafting with Large Language Models” (2024) 2nd Workshop on Generative AI and Law, colocated with the International Conference on Machine Learning, online: <https://blog.genlaw.org/pdfs/genlaw_icml2024/58.pdf>.