The unbounded attribute of the universe appears to be fractal, i.e. patterns within patterns. So the unbounded entropy is in the small, not in the large. This is why Armstrong points out that short-term cycles (e.g. day trading) are much more noisy (i.e. randomized). The higher-level the cycles in time, the less random the deviation.

So while A.I. will eventually emulate much of what it can observe that a set of humans can do, it can't (unless it becomes integrated within evolution and competitive reproduction) make every copy of itself a unigue solution to the unbounded evolutionary fitness continuum. The unseen fractal patterns encoded in the evolutionary continuum are not carried within the genome and dynamic living network of A.I. bots, because for one thing they are discrete and not analog biological. The complexity/entropy of a living creature is unbounded in the unseen fractal patterns encoded in the evolutionary continuum.

Here is the key point. Nature is not top-down controlled and there is no 'error' as every variation is information to the evolutionary continuum. A.I. would need to become alive in the sense that it is a self-reproducing, self-motivated, decentralized process controlled by no one. Then it would no longer be ARTIFICIAL intelligence.

So Armstrong has correlated the large fractal pattern cycles which are stable as is man's lifespan, reproductive maturity, Sun spots cycles, Earth's various cycles such as earthquakes, etc..