>>13827564niche/future-soon: graph-convoution models. We are really, really close to having a general graph-convolution network figured out, but not quite (the current name is for a specific use case). If we could apply convolution level layers to graphs, our machine learning capabilities would get really, really good.
>gene therapy/genetic enhancementcrispr put us very close to the science-fiction world of bio-alteration. Before that, all genetic changes were awkward and difficult (TALENs anyone?)
The ability to apply CRISPR/Cas9 based therapies to patients will be a game-changer. We have quite a ways to go with that technology, but the ability to "cure" genetic diseases would be a monstrous help to the world of medicine.
>general AIabsolute hype and memery aside, we are edging (heh) ever closer to a general AI. If we could build and AI as good or better than humans, we could basically retire humanity and would enter a golden age (we would relegate all tasks to restricted AI, basically). That's of course the "everyone sings kumbaya" version; plenty of social upheaval and world-changing events would occur along that path, and of course the ever-present threat of an AI uprising if that were to occur, but I'm really not that concerned about that unless we allow access to critical infrastructure and resources, which we wouldn't.
>anything related to better predictionsML has really had a golden-era in the past decade, with the expansion of new ML algorithms. Meta-learning/few-shot learning, RNNs, transformers, attention mechanisms, etc etc. They have slowly replaced most things on the back-end of society. We still have a long ways to go on this front I think, so anything related to improved predictive capabilities will always be in vogue.
