Unlikely. Generally speaking, people who work in ML, especially the top ML groups, aren’t doing anything close to ‘AGI’. (Many of them don’t even take the notion of AGI seriously, let alone any sort of recursive self-improvement.) ML research is not “general” at all (the ‘G’ in AGI): even the varieties of “deep learning” that are said to be more ‘general’ and to be able to “learn their own features” only work insofar as the models are fit for their specific task! (There’s a lot of hype in the ML world that sometimes obscures this, but it’s invariably what you see when you look at which models approach SOTA, and which do poorly.) It’s better to think of it as a variety of stats research that’s far less reliant on formal guarantees and more focused on broad experimentation, heuristic approaches and an appreciation for computational issues.
Unlikely. Generally speaking, people who work in ML, especially the top ML groups, aren’t doing anything close to ‘AGI’. (Many of them don’t even take the notion of AGI seriously, let alone any sort of recursive self-improvement.) ML research is not “general” at all (the ‘G’ in AGI): even the varieties of “deep learning” that are said to be more ‘general’ and to be able to “learn their own features” only work insofar as the models are fit for their specific task! (There’s a lot of hype in the ML world that sometimes obscures this, but it’s invariably what you see when you look at which models approach SOTA, and which do poorly.) It’s better to think of it as a variety of stats research that’s far less reliant on formal guarantees and more focused on broad experimentation, heuristic approaches and an appreciation for computational issues.