In this talk I shall describe an approach to the construction of conscious machines that initially seeks answers to three key questions:
What minimum set of core properties is sufficient to support the emergence of high-level mental abilities?
How may the presence of emergent high-level mental abilities be detected?
What properties must a machine have if it is reasonably to be labelled "conscious"?
As regards the first and second questions, I shall first explain why "design for emergence" seems an interesting road to follow, and then focus on the emergent creation and use of behaviours, "day-dreams" and plans by a machine designed and implemented at the neural network level. I shall give some details of the design that I am currently elaborating and have begun to implement as a C program, and consider how high-level mental abilities might be detected in this specific context.
As regards the final question, I shall first recall the well-known distinction between "subjective consciousness" and "behavioural consciousness" and then put the former to one side, and discuss how the latter may be achieved in a machine design of the foregoing type.