Simplified, cognitive computing aims to emulate the way human brains compute.
Systems learn instead of relying on programming only.
To do this they need to enhance and augment our senses with techniques like artificial intelligence, machine learning, advances feature recognition, natural language programming ....
I am working in machine learning for over 20 years now and there were always ambitious promises.
It is an amazing topic, but we need to understand that it needs to face the inverse-problems problems. They have the property of "ill posedness", which in turn can lead to surprising effects.
To visualize a geometric model is easy - to extract it from an image is difficult.
We have solved ambitious technical problems by applying multi-stratgy-multi-method machine learning atop our own machine learning framework - supervised and unsupervised analysis combining fuzzy logic based machine learning, all type of regressions, ANNs, ... kernel methods (with support vector machines as special instance). Fuzzy methods are a break through in the sense that they are (linguistic) understandable and computational, kernel methods "see" usually more details, ....
But we never tried to build systems that replicate human behavior - why this includes financial strategy decisions, see UTOPE.
At the other hand classical computer science has helped to automate mathematics and mathematics has made computer science more structured and adaptable.
New computing muscles will enable us to simplify and unify mathematical solvers and feature recognizers can apply automatic scross-model validators, ...
They may help us to do even better and constant re-calibration of models, ....
But, IMO, practical quant finance systems based on cognitive computing are years away.
