応用行動分析における AI

ABA runs on measurement. Technicians record frequency or rate, duration, latency, percentage correct on discrete trials, or interval-based estimates such as partial-interval, whole-interval and momentary time sampling. Behavior analysts then judge progress mostly by visual analysis of line graphs, looking at level, trend and variability within and across phases. Platforms such as CentralReach, Catalyst and Rethink Behavioral Health moved this from paper to tablets, and vendors are adding AI features like note drafting, trend flags and summaries for authorization reports. Their feature sets change often, so evaluate each on its own. The documentation has to match billing. In the US, ABA uses adaptive behavior CPT codes 97151 through 97158; for example, 97153 covers treatment delivered by a technician under protocol and 97155 covers protocol modification by a qualified professional. Auditors look for notes that support each billed unit, and near-identical notes across sessions are a known red flag. That makes AI drafting tempting, and it also makes accuracy essential. The ethical frame is the BACB Ethics Code for Behavior Analysts, in effect since January 2022, which covers data integrity, supervision, confidentiality and client dignity. Many autistic self-advocates have criticized ABA programs that focus on compliance or suppress harmless behaviors such as stimming. Tools that make data collection cheaper can scale good practice or scale those problems, depending on which targets are chosen. Two misconceptions are common. First, that software makes visual analysis objective. A trend line is a calculation; deciding whether a change is clinically meaningful, or caused by the intervention, is still professional judgment. Second, that an AI note is safe to sign because the data were real. A drafting model can add a detail that was never recorded, and the person signing is responsible for every sentence.