I've been thinking a lot about the current field of AI research and wondering if we're asking the right questions? I've watched some videos from Yann LeCun where he highlights some of the key limitations of current approaches, but I haven't seen anyone discussing or specifying all major key pieces that are believed to be currently missing. In general I feel like there's tons of events and presentations about AI-related topics but the questions are disappointingly shallow / entry-level. So you have all these major key figures repeating the same basic talking points over and over to different audiences. Where is the deeper content? Are all the interesting conversations just happening behind closed doors inside of companies and research centers?
Recently I was watching a presentation from John Carmack where he talks about what Keen is up to, but I was a bit frustrated with where he finished. One of the key insights he mentions is that we need to be training models in real-time environments that operate independently from the agent, and the agent needs to be able to adapt. It seems like some of the work that he's doing is operating at too low of an abstraction level or that it's missing some key component for the model to reflect on what it's doing, but then there's no exploration of what that thing might be. Although maybe a presentation is the wrong place for this kind of question.
I keep thinking that we're formulating a lot of incoherent questions or failing to clearly state what key questions we are looking to answer, across multiple domains and socially.