If "strong" AI means conscious AI with subjective experience, we're several milestones in theoretical neuroscience and cognitive science away.
If "strong" AI means software capable of learning any particular task a human could perform, we're so damn close that the research world is proceeding to cross the line as we speak. It's called "imitation learning" nowadays, and once there's a sufficiently general and accurate algorithmic framework with sufficient training data, it'll be everywhere.
If "strong" AI means software capable of learning many tasks a human could perform, generalizing them into a complete worldview, and taking volitional action based on that worldview to accomplish conceptually-specified goals... I'd call it one to two decades away. There are some major substantial problems remaining to research, but we do have solid foundations for it: uncertain reasoning in Turing-complete modeling domains lies at the intersection of several thriving research fields.
In terms of "how much science is left", I'd say our current point is roughly analogous to the physics community having spotted that the photoelectric effect can only be explained by discretized photons and starting to generalize quantum physics, but having most of it remaining to discover.
Sorry, but imitation learning is going to be quite good enough to automate many, many tasks without any need for AGI agents as such. The labor market problem is a right now problem, not a "leave it to those futurist guys who don't have to put up with academic rigor" problem.
I take it you have a familiarity with the subject. If a layman with a science degree would like to start following the developments in the field, would you suggest any journals/conferences as a starting point?
Neural Information Processing Systems is basically one of two top-tier conferences in machine learning, and the one that has more cognitive-science findings published in it, to my knowledge.
The reason nobody publicizes "AI" is because, from the perspective of a learning researcher or cognitive scientist, there basically is no such thing: there exists a large variety of cognitive tasks that can be modeled and pieced together in various ways, with varying degrees of accuracy when applied to real data-sets or compared to human performance.
If "strong" AI means software capable of learning any particular task a human could perform, we're so damn close that the research world is proceeding to cross the line as we speak. It's called "imitation learning" nowadays, and once there's a sufficiently general and accurate algorithmic framework with sufficient training data, it'll be everywhere.
If "strong" AI means software capable of learning many tasks a human could perform, generalizing them into a complete worldview, and taking volitional action based on that worldview to accomplish conceptually-specified goals... I'd call it one to two decades away. There are some major substantial problems remaining to research, but we do have solid foundations for it: uncertain reasoning in Turing-complete modeling domains lies at the intersection of several thriving research fields.
In terms of "how much science is left", I'd say our current point is roughly analogous to the physics community having spotted that the photoelectric effect can only be explained by discretized photons and starting to generalize quantum physics, but having most of it remaining to discover.