A Trellis for Novice AI Practitioners
James Guymon, Gordon W. Romney · 2019
It is commonly said that all models are wrong, but some are useful, yet the utility of models does not come easily. As the allure of AI attracts new practitioners it is crucial for sound technique to be modeled and freely made available. This is especially true in fields such as information security, where manual human observation is impossible, the consequences for failure are high, and the existing workforce is competent enough in computer programming to read documentation, troubleshoot through frustrations and ambiguity, and, in worst-case scenarios, succeed at getting an algorithm they do not know much about to output a model they do not understand with unfounded and unbridled confidence. Where previous generations of statisticians mentored new practitioners, passing along processes that protected them from the dangers of making a living predicting an unknowable future, today's workforce often queries in isolation. In a culture where algorithms decide who gets interviewed for jobs, who gets flagged for an IRS audit, who warrants increased scrutiny at immigration checkpoints, who should be approved for a home or auto loan, and what the terms of that load should be - and so on ad infinitum - there should be no doubt that the quality of our society's data science apparatus must be excellent to avoid catastrophic ramifications. For this purpose, we present a tool to help information security students and professionals transition into the world of data science. A carefully chosen data set, selected for its anticipated familiarity and suitability for quality work, gives context to our training exercises. As entrants gain confidence and experience the tool will evolve with them, self-organizing into a couture tool-kit that reflects their specialty, preferences, and even their personality. Built upon solid data science processes and andragogical foundations this learning tool will increase safety from mortal error during the dangerous initial formative period of a developer's transition to data science practitioner.