Tactile AI
Axon Virtual Reality training as Seattle PD's third AI category: systems that don't touch a case file at all, but train the officer's judgment and muscle memory before they're ever in the field.
The Third Category: AI That Never Touches a Case
Operational AI reads live cases. Accountability AI searches historical records. Tactile AI is different in kind from both: it doesn’t touch case data, evidence, or the accountability record at all. It’s used entirely in training, before an officer is ever in a real incident — which means it carries a fundamentally different risk profile than the other three categories in this module. There’s no evidentiary chain to protect, no Brady/Giglio exposure, no historical-record accuracy question. The risk here is a different one: does the training actually transfer to real-world performance, and does simulated practice change behavior in the ways it’s designed to.
Seattle PD’s implementation of this category centers on Axon Virtual Reality.
How Axon VR Training Works
The system places an officer in an immersive, simulated scenario — a domestic disturbance call, a traffic stop, an armed-subject encounter — and lets them practice decision-making, verbal de-escalation, and use-of-force judgment in a repeatable, controlled environment. Two specific applications matter most:
- De-escalation training — practicing verbal and tactical de-escalation techniques against a responsive simulated subject, in scenarios that would be impossible or unsafe to fully replicate in a live role-play exercise at scale.
- TASER proficiency — building and testing muscle-memory proficiency with less-lethal tools in scenario-realistic conditions, rather than static range practice alone.
The pitch for VR training over traditional role-play is repeatability and scale: a scenario can be run identically for every officer in a cohort, scored consistently, and repeated as many times as needed — something a live actor-based exercise struggles to do at department-wide scale.
Why This Category Faces a Different Question Than the Rest of the Course
Everything else in this course asks some version of: does this AI system’s output, when it touches a real case, hold up to legal, ethical, and evidentiary scrutiny? Tactile AI asks a completely different question: does simulated training transfer to real-world performance?
That’s an empirical training-science question, not a civil-liberties or evidentiary one — and it means the review process from Module 1 (Surveillance Impact Reports, Civil Liberties Assessments) doesn’t map cleanly onto this category, because Tactile AI isn’t a surveillance technology and doesn’t collect data on the public at all. Evaluating Axon VR effectively requires the same kind of question you’d ask about any training program: are outcomes measured, and against what baseline?
A highly realistic VR scenario is not automatically evidence that skills learned in it transfer to a real encounter. Training science broadly draws a distinction between engagement (does the trainee find it immersive and take it seriously) and transfer (does performance in the simulation predict performance in the field) — and the second is much harder to establish than the first. Before treating VR training hours as equivalent to field-readiness, leadership should look for outcome data — use-of-force incident rates, de-escalation success rates, post-training assessment scores — not just adoption and engagement metrics from the vendor.
A vendor's marketing material for Axon VR reports high officer engagement scores and strong self-reported confidence after training. Command staff want to expand VR-based de-escalation training department-wide based on these numbers. What is the strongest independent question to ask first?
Reveal Answer & Explanation
Ask for outcome data that measures actual field performance change — de-escalation success rates or use-of-force incident rates before and after training — rather than relying on engagement scores or self-reported confidence alone.
Engagement and self-reported confidence measure whether officers found the training compelling, not whether the training changed real-world behavior. This mirrors a pattern seen elsewhere in this course (see Report Quality Evaluations): a positive-sounding vendor metric that doesn’t actually measure the outcome that matters. A defensible expansion decision rests on field-performance evidence, not adoption enthusiasm.
Up Next
The first three categories in this module each do one job — reading cases, searching records, or training officers. The final lesson covers what happens when several AI-driven systems are wired together into one operational platform: Integrated Systems.