Context comes from internal models of how the world works. A database doesn't know what a bike is, exactly as you say, because no matter how many tags it associ

Context comes from internal models of how the world works. A database doesn't know what a bike is, exactly as you say, because no matter how many tags it associates with it they have no purpose to it. To understand a bike in the way we do it would need to have kinematic models of the bike, a person, a road, physics simulations of their behaviour and models of human behaviour for the activity of riding the bike. We do have kinematic simulations of this kind, but what we don't have is a way to map a visual image of a bike to generate an internal kinematic model, or associate them together in a useful way, and then contextualise that within operational and intensional models, then a way to relate runs of the simulation to the actual physical environment in an intensional way. I don't think there's anything impossible about any of that, but we certainly don't have this capability at the moment. There's a long, long way to go.