This paper studies an interactive learning system that couples internally guided learning and social interaction for robot learning of motor skills. We present Socially Guided Intrinsic Motivation with Interactive learning at the Meta level (SGIM-IM), an algorithm for learning forward and inverse models in high-dimensional, continuous and non-preset environments. The robot actively self-determines: at a meta level a strategy, whether to choose active autonomous learning or social learning strategies; and at the task level a goal task in autonomous exploration. We illustrate through 2 experimental set-ups that SGIM-IM efficiently combines the advantages of social learning and intrinsic motivation to be able to produce a wide range of effects in the environment, and develop precise control policies in large spaces, while minimising its reliance on the teacher, and offering a flexible interaction framework with humans