Advice I would give anyone concerned about competing with AI: change the context so you're not in direct competition.

One approach would be to ride the wave. Experiment with new AI-enabled tools as they are released, and learn their limitations. Then you will have a relative advantage over those unfamiliar with the tools, and will be able to identify potential ways to leverage and complement their capabilities. You'll then be able to provide expertise with regard to how to properly use these tools, and not overspend on tools which are a poor fit to the job at hand. Think Gartner, but more hands on.

Let me know if this works for you, but I expect the approach should get you past analysis paralysis and replace fear with reasoned concern.

Ideally, enough physics to now when Kurzweil is resorting to handwavium when presenting an argument.

This may seem cynical, but an initial hurdle is creating apps that will sell to a military-industrial complex, which is willing to pay more per unit for small errands batches. That would bootstrap a supply chain that can be leveraged for consumer VR/AR. Silicon Valley would not be what it is today without military-industrial complex investment.

What if it's already a secret and being used militarily?

Depends on which paradigms, algorithms, and processes win out.

Black Swan events are by definition unpredictable, so technically no. It's post-Singularity events which are considered unpredictable.

This prediction became partly true in a mundane, boring way. The Internet of Things has been a commercial reality for years.