The proposed research centers on the thesis that machine-generated first draft documents can and will be widely applied to legal work product traditionally viewed as bespoke and therefore reserved for attorneys. This is based on the view that most legal writing tasks can be decomposed into subtasks that fit into one of two categories: (1) creativity- and judgment-driven writing and (2) mechanical writing. The first category generally requires human-level intelligence to balance numerous factors and disparate context to shape writing strategy and language choice, and thus should be left for the attorney. The second category, however, is often ripe for automation.

An initial step in validating these ideas will be to perform surveys of practicing attorneys in a broad range of practice areas. These surveys will help determine the most common attorney-written documents in each practice area as well as which documents may be potential candidates for partial automation. Depending on the results, some or all of the identified documents may be further analyzed to estimate the technical feasibility of automation, and what specific input might be used and what output could be expected. Illustrative case studies may be included to demonstrate feasibility.