Abstract
In this project I seek to better understand the foundations of analogical thinking: how we understand a new situation by finding similarities to known situations. This is the key feature of our intelligence and the missing ingredient for creative machines: many scientific discoveries were made by analogies, in language we use analogies to create metaphors, and we use analogies in legal reasoning, in teaching, and so on. There are two crucial open questions: First, how can analogical thinking emerge from our brain or artificial neural networks, and second, how can analogical thinking account for its creative problem-solving abilities? I will tackle the "emergence question" by the theory of neural-symbolic integration that allows to use the symbolic and conceptual work on analogy to construct neural networks implementing creative analogical thinking. As a case study, I will apply this idea to the famous analogy between the cognition of space and time---aiming for a deep explanation of why we conceptualize time through space. I will tackle the "problem-solving question" by using the research on the computational power of neural networks to understand which problems the neural "analogy-machines" from the first question can solve and how they do it---this will show that analogy-machines are (in a sense) more powerful than modern computers. This project will not only contribute to deeply understand analogical thinking, but also to understand the emerging - both fascinating and frightening - machine learning technology: machines that, for the first time ever, can be creative.