Despite huge investments into Deep Learning we did not get close to making machines understand natural language (NLU). Can semantic approaches make up for weaknesses of Deep Learning like for example abstraction and generalization ? If humans would need to touch hundreds of hot ovens before they being able to extrapolate and generalize - our lives would be much less enjoyable. But how can we build in these capabilities alongside common sense knowledge into machines? And why would that help?
Despite huge investments into Deep Learning we did not get close to making machines understand natural language (NLU). Can semantic approaches make up for weaknesses of Deep Learning like for example abstraction and generalization ? If humans would need to touch hundreds of hot ovens before they being able to extrapolate and generalize - our lives would be much less enjoyable. But how can we build in these capabilities alongside common sense knowledge into machines? And why would that help?