A many previous posters have described, forced dependence on <AI> and flawed <AI> advice are not specific to Palantir, but properties of the method. A decision system designed with enormous complexity guarantees both the need for AI direction and cosy-of-errors ( or hallucinations ) intrinsic to that machine process. Likely some desirable systems < ex/ containing a high-temperature plasma > require this machine aide; OTOH deciding which group of people to kill should remain 100% under human control. BUT! The pressure for automation seems unending ... comparing small things to large, I recently (needed) a systems upgrade from ( a spastic) Ubuntu OS to Debian. My very modest skills would have failed the task without extensive GOOG.AI support. Does USA military feel/experience the same constraint w.r.t. target-choice ? What about HR departments or UNI admin ?