It might fail to impress how much progress was made since CYCE and https://en.wikipedia.org/wiki/ELIZA but the progress is not impressive due to better logic (which is still absent) but rather due to the sheer amount of data (phrases/sentences/facts) that this engine has stored as reference. This huge amount of data is what creates the impression that the engine understands while it is just related facts that are retrieved and transformed into text that resembles the wording of the query.
Right. It's still just an incredibly complex evolution of the Markov chain. Some people feed on the semi-magical hope of 'emergence' out of a sufficiently complex system. AGI will be an engineered product, not something that just pops out, no matter how much data is processed. The engineering requires an architecture, probably a 'self' (embodied individual) that has intrinsic needs, builds a dynamic hierarchy of desires on them, can generalize and analyze sensed data, remembers and imagines, and interacts intentionally with a reality through sensors and effectors on the basis needs that change over time and situations. I think researchers, even the statistics crowd, are being pushed toward architecture, but I believe success will come from an architecture-first approach.
It might fail to impress how much progress was made since CYCE and https://en.wikipedia.org/wiki/ELIZA but the progress is not impressive due to better logic (which is still absent) but rather due to the sheer amount of data (phrases/sentences/facts) that this engine has stored as reference. This huge amount of data is what creates the impression that the engine understands while it is just related facts that are retrieved and transformed into text that resembles the wording of the query.
Almost like quantity spilling over into quality ...
Right. It's still just an incredibly complex evolution of the Markov chain. Some people feed on the semi-magical hope of 'emergence' out of a sufficiently complex system. AGI will be an engineered product, not something that just pops out, no matter how much data is processed. The engineering requires an architecture, probably a 'self' (embodied individual) that has intrinsic needs, builds a dynamic hierarchy of desires on them, can generalize and analyze sensed data, remembers and imagines, and interacts intentionally with a reality through sensors and effectors on the basis needs that change over time and situations. I think researchers, even the statistics crowd, are being pushed toward architecture, but I believe success will come from an architecture-first approach.