Creative artificial minds, defined as entities that are able to perceive the world, organize information, use it creatively, and that are able to interact with humans, are a long term goal in research. These entities are useful in various human-related endeavors, such as organizing information, help in day to day tasks, and provide monitoring and support for people with special needs. Advanced forms of these minds can conceivably become autonomous and independent. Appealing as it may be, this last aspect is, however, beyond the current goals of my research.
On the other hand, it is not the object of our research to produce entities that are weak human simulations of any kind, but rather approach the problem from the point of view of compatibility. That is, the overall goal is to build smart or even intelligent entities that are able to understand humans, cooperate with humans, and relate emotionally or empathetically with humans. This approach configures autonomous agents, but also complements for people (computer-aided everything -- think recommendation or prediction -- extendable to sympathetic relationships and commonality of interests). It also implies the ability to understand human-level information, i.e., human-oriented media, such as cultural artifacts, e.g., written material (text) or music (audio).
Many of the problems associated with research into socially capable artificial minds are very hard and require the input of multiple disciplines, such as psychology, sociology, mathematics, among others. The approach is to contribute to the overall development (investing in specific components at a time) without requiring a complete embodiment. In this way, contributions can be made without the prohibitive cost of integration. Also, some of the research results are liable to be useful to other fields.
Specifically, the components addressed by us concern natural language interaction, information extraction and knowledge acquisition and management, as well as work in perception and emotion in the context of cultural artifacts, instantiated in music and film, and creativity, instantiated both in the context of music and of narrative. All of these components are in the scope of classical AI and its subdisciplines (e.g., Natural Language Processing). However, modern approaches, especially those that are able to learn, go beyond classical approaches and require advanced algorithms and high performance computing. This is especially true in the context of signal acquisition, compression, and mapping to higher level concepts. Furthermore, since the minds have to be adaptive, the algorithms have to account for online and (mostly) unsupervised learning.
Our research addresses concerns of industrial relevance, especially in companies that rely on advanced aspects of machine learning and/or information extraction. In fact, some of the people involved in parts of the work described here are also conducting professional work in these contexts, establishing durable bridges between academia and industry. This is a positive aspect, not only to maintain healthy exchanges of ideas, but also as a motivating factor for new students.