> "In order to be able to have fluent conversations with machines, several issues need to be solved first: text understanding, question answering and machine translation."
The article makes it sound like "text understanding" was just around the corner, maybe next year...
I doubt that because understanding (arbitrary but meaningful) text requires real intelligence, and AI is far away from that.
And if it really happens one day, then our jobs are all gone. Because programs are text, and with proper training programs are way easier to understand than arbitrary prose - a much smaller subset of concepts contained in a clear structure, instead of almost infinite concepts or even new ones, to be structured however the author sees fit.
So prior to understanding language, AI should be able to understand programming language, because programming language is just a a small subset of language.
Seems like a matter of hierarchy. Current NNs are flat and tiny. We're only simulating small chunks of the base levels. There is no executive level that integrates from multiple minions. You need that hierarchy to even start considering "meaning". What we're doing now is making bricks for that future building.
Of course, then there's that other school of thought that predicates the true grasp of meaning on the participation of consciousness. But that's another can of worms for another time and/or another forum.
Florin_Andrei is clearly including recursive/tree NNs when (s)he says current NN's are flat (and I think depending on how you define "flat" I think this is accurate).
Programming languages are already an intermediate piece of text that computers understand really, really well. The problem is always one of human intention, and crafting the correct program based upon what the human intends to be saying, and not necessarily what they actually are saying.
The article makes it sound like "text understanding" was just around the corner, maybe next year...
I doubt that because understanding (arbitrary but meaningful) text requires real intelligence, and AI is far away from that.
And if it really happens one day, then our jobs are all gone. Because programs are text, and with proper training programs are way easier to understand than arbitrary prose - a much smaller subset of concepts contained in a clear structure, instead of almost infinite concepts or even new ones, to be structured however the author sees fit.
So prior to understanding language, AI should be able to understand programming language, because programming language is just a a small subset of language.