Realizing the PAL Vision

Central to the realization of DARPA’s vision of an enduring cognitive assistant is a concept the CALO team calls “learning in the wild.”

According to CALO Principal Investigator Bill Mark, “Until now most learning research took place ‘in vitro,’ which means that the research community studied learning from precollected, engineered data sets. Researchers ran the data through various algorithms and generated results, then adjusted various parameters until it produced increasingly better results. But that approach won’t work for CALO.

"DARPA’s vision of an enduring personal assistant is based on the idea of adaptive learning. Today it’s the user who has to adapt to the system. What's really needed to make more useful, richer systems is a change in the way systems are designed. The system needs to adapt to the environment, especially to its user’s changing needs and requirements. It must be able to learn in the wild."

"The need for learning in the wild is based on the fact that it is never possible to engineer enough knowledge into the system to enable it perform well in a changing environment. CALO must acquire its understanding of its environment, either through observation or by instruction. CALO will observe, listen, be given advice – all in the natural flow of the user’s daily activities. So, either explicitly or implicitly, the user continually teaches the system to improve its performance." More.