Emergence and Machine Learning

Machine Learning has been successfully applied in the areas such as recommendation systems and visual object identification in the recent past. Most of the machines learning algorithms, including deep learning, are in the final analysis nothing but regression systems and classification systems.

However, on the other hand, nearly all the living systems, and human culture – particularly markets, exhibit strong tendency of emergence from complex system. Wikipedia defines emergence as a process whereby larger entities, patterns, and regularities arise through interactions among smaller or simpler entities that themselves do not exhibit such properties. Emergence is a defining property of complex systems, as against complicated systems. In economics, emergence is observed when millions of self serving agents acting on price signals generated by markets are able to produce goods and services of requisite quality and quantity as demanded by the consumers without existence of central planning and control.

Even though the current generation of ML applications looks nothing short of magic, we believe that next level of wow in the ML applications will not come by merely throwing more data and more computation power at the current algorithms. We need to develop algorithms, which will recognize emergent phenomena in the human interactions. Some believe that algorithms like swarm intelligence and ant algorithms are a path in that direction. Yes, these algorithms give great solutions to TSP, however, they are not truly scale agnostic and hence emergent algorithms. We need to look at nature more closely and study the fractal gene regulatory framework and try to develop algorithms that may sniff out existence of such a framework in human generated data. When we are able to identify such frameworks, which will truly help understand society wide phenomenon such as boom and bust in markets or sudden jump in refugees.


Share on Facebook Tweet
comments powered by Disqus