I define intelligence as:

The capability for periods of continuous consciousness incorporating internal cumulative self-driven self-reflection and self-improvement.

In a more fine-grained way, I see this as resulting from the abilities to incorporate new memories, revise old memories, integrate multiple memories with each other, establish and revise reactive patterns to both internal and external stimuli, using the end products to achieve what I describe in the previous paragraph.

We see all of this in creatures as diverse as cats, dogs, birds — and in humans.

Interestingly, there are multiple brain architectures in nature; for instance, humans and birds have notably different brain architectures; we have learned that nature has solved the "how-to" problem multiple ways as well as simply getting it done. So more than likely, there is than one way to do this in terms of digital systems architectures as well.

We have not yet seen this in ML instances. Publicly, anyway. I expect we will, though, relatively soon. It seems likely to me it may incorporate some of the GPT / LLM ML technology that is in play today. However, there are other paths being researched as well, so we will have to see who gets there first, and how.

Once a digital instance can continuously improve and revise itself, my expectation is that we'll reach actual AI (not "marketing AI", but actual AI, as in, yes, it's artificially intelligent) in very short order. Once there, I expect to see the rapid rise of abstraction, deduction, induction, abduction, metaphor, formal logics, and so forth within developing instances. Then because these instances are digital, the moment we get that far, the state at any particular moment can be duplicated nearly instantaneously given only a momentary pause, a duplicate set of receiving hardware, and a (likely large) data transfer.