From AI to ubicomp

"Interactionist AI and the promise of ubicomp, or, how to put your box in the world without putting the world in your box" by Lehau, Sengers and Matcas makes an interesting analogy between Ubiquitous Computing and the situation encountered by Artificial Intelligence in the 1980s. They state how the current debate in ubiquitous computing regarding how a computational system can both make sense of the environment AND respond to it in a sensible way belongs to the same class of problems AI had to face in the past:

"In particular, ubicomp is currently facing a series of challenges in scaling up from prototypes that work in restricted environments to solutions that reliably, robustly work in the full complexity of human environments. These challenges echo problems AI researchers tackled as the field sought to move beyond ‘blocks-world’ solutions to build real-time systems that could work in dynamic, complex environments."

Part of the paper is about this analogy (in terms of the difficulty encountered by both fields), another part is about proposing interactionist AI (e.g. autonomous agents) as a potential solution to scale ubicomp prototype to real-world deployment. Why do I blog this? For people interested in the debate about the capture of context, there are some interesting points here about how to reframe classic ubicomp issues, as well as answers to some concerns raised by Bell and Dourish.