Currently the AI we see around us – Pinterest’s visual recognition tools, Spotify’s Discover Weekly, Amazon’s Alexa, countless numbers of self-driving cars and chatbots are all examples of ‘Narrow AI’ – dedicated to assist with or take over specific tasks.
You and I are more sophisticated in the sense that we are able to learn new things from different domains and apply these learnings interoperably. How do we build algorithms and data structures that can be applied to different learning tasks and be successful at all of them like us humans? This is known as ‘General AI’, and is the holy grail. Currently, there isn’t a clear research agenda to get us there.