By Timothy Tillman, The University of Newcastle
If you buy the exponential growth of tech metrics (such as cost per memory, energy per floating point operations, etc.) staying that way for the foreseeable future, then computers are about to get really interesting. For some this is a foregone conclusion, for others not so much. A group of computer scientists were polled on this. 59% said machines will be able to simulate every aspect of human intelligence, which is quite low considering the power of the statement “every aspect”.
When just asked whether we will understand the human brain enough to simulate it with a machine, 79% responded affirmatively.
Machine learning has become a popular topic with it focusing on statistical methods of learning instead of earlier methods using discrete trees and logic structures, such as those used for early chess computers. This change has enabled a great number of successful algorithms to solve real world problems, such as the 2006 NASA ST5 which has an antenna [3] designed by an ‘evolutionary’ algorithm that models mutation with random processes.
Some [1] argue the use of statistics in computer science lacks rigor. While this may be true in some instances, statistics will be an increasingly required skill for those in artificial intelligence and data science. My prediction for the future is that increasingly sophisticated statistical algorithms will be developed for these fields to accomplish complex tasks, and these algorithms will enable machines to acquire human like traits.
This seeming required for machines to use statistical algorithms to acquire human like traits may be related to both the random nature of mutations that allow natural selection to take place, and the more hypothetical Penrose theory of consciousness involving random processes brought about my quantum mechanical phenomena. Could it be that our cognitive biases are just an example of our brain overfitting our dataset?
The main goal of this post was to convey some of the interesting aspects of machine learning, and why the cool things to happen in the future will be backed by increasingly sophisticated statistical ideas. Anyone can utilise platforms to do stats and machine learning like R, and Python modules like scikit-learn, pandas etc., which are free, open source and can handle large datasets. They are also very accessible, as far as scripting languages go, and there are many guides and discussions available on the internet.
[1] http://blog.revolutionanalytics.com/2014/08/statistics-losing-ground-to-cs-losing-image-among-students.html [2] http://www.nickbostrom.com/papers/survey.pdf [3] https://en.wikipedia.org/wiki/Genetic_algorithm#/media/File:St_5-xband-antenna.jpg