VirTra Inc (VTSI)

[utahstock12]

The technology to do that even a few years ago would have been tough, but now it's everywhere and it wouldn't even be that expensive to implement so presumably they're doing it on some level already as you suggest. Just need speech recognition on top of some sort of LLM optimized for distressed/agitated/unwell person. You could let it loose to say literally anything as the tech exists to make believable mouth movements for generated speech. Then, whatever the latest psych studies show as de-escalatory language is more likely to generate best outcome (maybe even with scores for volume/tone etc), but with some randomization element since sometimes you can say all the right things and the person still shoots.

Also, in some of the demos I had watched before there is a lag between the person in the scenario acting and the sim responding. Presumably this is because the operator of the sim has to figure out which branch on the tree to select next. Now, the sim itself can understand and progress down the tree so the situation will feel more real time.