WRT: Writer Response Theory

The Legal Machines Project is trying to create artificial intelligence agents to aid the legal profession. This recently garnered media attention when a law firm announced it would be using such an agent to provide legal resources online next year.

Digital text aficionados may be familiar with the past work of project member Selmer Bringsjord, director of the Rensselaer Artificial Intelligence and Reasoning (RAIR) Laboratory and author of Artificial Intelligence and Literary Creativity: Inside the Mind of Brutus, a Storytelling Machine (1999), which documents the creation process and results of one story-telling machine via the extensive use of logical propositions.

While there are currently many different attempts to apply AI to legal practice, the Legal Machines Project emphasizes cooperation over competition, and frames its goal in terms of an imagined human-agent interaction scenario:

Tags: chatbots, generators, researchers

…A year ago, the attorney would have picked up the telephone and cancelled his weekend plans. The revised weekend “plan” would have been to live in the office with a team of two junior associates, two secretaries and one or two paralegals. But today, the attorney doesn’t pick up the phone to cancel his weekend plans, nor to mobilize the troops. He looks at his clock again — 4:20 — rolls his sleeves back up and gets to work. He begins an interactive conversation with his terminal, working with the artificially intelligent agent in much the same manner that he might instruct a junior associate.

This scenario reminds me strongly of Apple’s “Knowledge Navigator” (pictured top and above), a 1980s film short that also imagines a future of computing via intelligent agents. The lush professor’s office and the complex audio exchanges in the Apple short convey a good sense of both the promise and the fantasy of such scenarios - that intelligent agents will one day allow the delegation of complex, poorly specified tasks (which seems possible) in a way that reduces labor without increasing expectations, creating ample leisure time (which seems highly unlikely, but perhaps that is beside the main point).

The “Knowledge Navigator” and Legal Machines scenarios are meant to be bold imaginings, however both are fairly careful in modestly proposing that such agents only intervene in rule-based domains such as deforestation graphs or merger regulations. The Legal Machines Project specifies these domains as ones in which:

1. The knowledge in the domain must be complex (if the knowledge is too simple, no need for an expert system), but not too complex, for then knowledge articulation becomes impossible.
2. The knowledge must be representable in a way that transforms most “seat of the pants” human judgment into explicit rules or guidelines suitable for use in computation.

A first step for the project? Building a legal machine that can do well on the LSAT.

It is interesting to see Bringsjord moving from the generation of narrative out of logic (Brutus) towards the logical tractability of legal discourse, which much recent scholarship has focused on in terms of narrative modes (for an interesting summary of some narrativist debates particular to US law, read Kim Lane Scheppele’s Narrative Resistance and the Struggle for Stories). In many ways, this feels like coming full circle - changing the focus perhaps from output to input, but still asking how we can automate translation between a human and machine understanding of the world.