Can we build an early warning system for loss of control to AI?

An early warning system for loss of control to AI requires, at its core, a forecast of the outcome of our current trajectory. Can we build a mathematical model to forecast this? Two fairly intuitive responses occur to me. First: it seems like it would be very difficult. The conceptual underpinnings of "loss of control" are contested, so there's an open question about whether you end up modelling something coherent and operationalizable. Further, there's a well-established line of work arguing th...