The annoying thing about human emotions is that they’re not always equally clear, especially for a computer. Raising one’s voice, for example can be a sign of enthusiasm, but also one of rage. People are trained to combine different kinds of (uncertain) information and drawing the right conclusions based on them. This handling of uncertainties, however, is difficult to program into a computer program.
For this Bui adopts a mathematical formula designed in the 60’s to drive factory processes: the Partially Observable Markov Decision Process (POMDP). He demonstrated that this technique is suitable for integrating the emotions of the user into a dialogue system. The POMDP-technology can deal with uncertainties. The method performs, as long as it’s tested with small-scale dialogue problems, better than previous systems. Larger problems, however, require too much computing power.
That’s why Bui developed a hybrid strategy. This combines the Dynamic Decision Network (DDN) with POMDP. In contrast to POMDP’s, DDN-POMDP’s split the dialogue systems on two levels. They balance looking far ahead and attainability where computing power is concerned.
Future Vision by Erwin Van Lun on this article
Step by step we learn to react better and better to humans. First to typing, then clicking, by now a little better to our voice, but this kind of development shows that it’s definitely not going to stay like this. Eventually ‘listening’, ‘observing’, will be the most normal thing in the world for brands.