Homing in on the conceptual design of a thinking machine.
There is growing agreement all round on model-based reasoning as the being at the heart of human cognition. Lake and various colleagues have published papers (2015, 2016) on the topic. They identify prior knowledge of the environment, whether innate or learned, as constituting such models. Botvinick et al agree with them as far as that goes. Their thinking differs in that they prioritise the machine’s own autonomous learning as the means of acquiring such prior knowledge.
They note that engineering prior knowledge and models into a machine by hand is not practicable. So, the machine needs to be able to learn on its own. They observe that we are improving our ability to create such learning without pump-priming with masses of data. The other advantage of such an approach is that the machine also learns autonomously how to use the acquired prior knowledge.
Link to paper: https://arxiv.org/ftp/arxiv/papers/1711/1711.08378.pdf
You may also like to browse other AI papers: https://www.thesentientrobot.com/category/ai/ai-academic-papers/
