AAAI: It's Time for Large-Scale Systems

Today AAAI is again at a time of transition. The infrastructure—computing power, memory, bandwidth, and connectivity—has evolved dramatically. Important theoretical advances have been made in areas such as machine learning, natural language, knowledge representation, task descriptions, sensing, and action in the world. Once again there is substantial demand for AI applications from customers such as DARPA, with a requirement to solve real problems. We need to find ways making AI components interact in larger scale systems. Researchers need to explore how multiple methodologies can be incorporated into robust applications. We should not treat alternative approaches as competitive, but seek to find ways using such alternatives synergistically. For example, probabilistic reasoning and symbolic reasoning should be brought together to provide rapid solutions to the likely cases, and explainable alternatives derived from symbolic models for the rarer cases. We need to make results of machine learning understandable and modifiable through symbolic methods. We need a spectrum of methods to support improvement of system performance from experience. Rule-based systems improve through the learning of their creators and further programming. Machine learning systems improve by being given more human annotated examples and feedback on their behavior. We need to incorporate learning between these two— where the user can instruct a system in how its behavior should change, without the need to know how