Here's a more concrete and less "woo woo" (although only less, not necessarily "woo woo" free): Suppose for the moment that consciousness exists on some sort of scale (i.e., I'm not presuming binary have/have not, and I'm not really presuming total ordering of "quality" either). Humans have "more" and a dog has "less" and a fly has "hardly any". We are not too far out into the water, here, really. There is some sort of meaningful something that humans have lots of, dogs have less of, and flies have hardly any, even from a strictly physical perspective.

Now, suppose we assume that the end of the article is correct and "consciousness" is really just a form of integrated information processing, and their device for measuring the interconnected nature of a given brain is measuring something meaningful.

(And let me take a moment to observe that when I say "assume", I'm serious. I'm not asking for agreement, I'm saying, work with me and assume for the sake of a discussion.)

Now, we are still not too far out into the waters here. We have a device that is measuring something real, and fairly non-mysterious. It would not be hard to create an equivalent for a computer system by considering the system as a set of graph nodes, then probing the connectivity of the system. We're still standing on fairly concrete ground.

Now, consider that in the near future we will be able to build computer systems that by any basic connectivity metric (average number of neighbors, various stats of connectivity shape, etc.) would match even human brains, but can be demonstrably doing something very simple like just copying data from here to there, something that is clearly not "conscious" behavior.

Characterize the difference between my hypothetical computer system that matches the human brain on $RELEVANT_CONNECTIVITY_METRIC and the actual human brain. In theory, as you refine the connectivity metric you should also be describing to me how to build a human-equivalent brain, too, or at least telling me exactly how I'm falling short.

Note this is a sort of mathematical argument, where I'm asking you to bring your own $CONNECTIVITY_METRIC to the party. It seems pretty clear that it can't be as simple as whether or not it is or is not of a certain size and happens to follow a power law in the connectivity graph, because it's really easy to produce a program that would have that level of connectivity and still not be conscious. (A naively-written non-trivial program naturally has a power-law distribution to it.)

This phrases the problem in a fairly practical way ("how do I build a human-class AI?"), demonstrates the difficulty of answering it even given a powerful and potentially incorrect assumption, and is phrased virtually entirely in physical terms such that a sufficiently concrete and accurate answer would very likely lead to the ability to build AI. And we have no clue about that $CONNECTIVITY_METRIC. All we've got is some people asserting that the answer lies in this direction and others asserting it doesn't and very little hard data to help us answer that question.

And I'd also observe that I'm not even worried about the answer, this generalizes trivially to the "set" of conscious behaviors, and also your choice of whatever test you like to apply for consciousness up to and including "I'll know it when I see it", since right now we can't answer any of these variants.