Roots of Today’s Data Science & AI in India and England?

The field of Artificial Intelligence (AI) is in the midst of a major resurgence today. In many sectors of the industry, modern-day AI frameworks such as Data Science, which includes Machine Learning and Deep Learning are causing an unprecedented paradigm shift in approaches towards problem solving in Predictive Analytics, Fraud Prevention, Autonomous Vehicles, Natural Language Processing and other challenging areas. These technologies are ushering in a new frontier of disruptive innovations in almost every major vertical today, including healthcare, transportation, financial, consumer electronics, communications, retail, energy and utilities.

It is generally believed that the field of AI was originally founded in the 1950s on the premise that human intelligence "can be so precisely described that a machine can be made to simulate it.” Its goal has remained the same ever since - for the next 60+ years: “to build and understand intelligent systems, computational models and innovative solutions” mimicking or exceeding human intelligence through interactions with the physical world, by means of data analysis and synthesis leading to knowledge representation and problem solving.

John McCarthy of Dartmouth College and IBM (later at MIT) is credited as the “Father of AI” for creating this new field along with Marvin Minsky (MIT), Allen Newell (CMU), Herbert Simon (CMU), Nathaniel Rochester (IBM), Arthur Samuel (IBM), Claude Shannon (Bell Labs) and few others at a workshop known as the famous “Dartmouth Conference” in the summer of 1956. 

Since then, AI research focused on building rule-based and knowledge-based heuristic systems (and “expert systems”) to deal with new information and associated scientific data. Early AI programs for playing checkers, character recognition, Fuzzy Logic controllers etc. led to many “soft-computing” systems based on “Artificial Neural-Networks” (ANN), “Genetic Algorithms” and “Support Vector Machines” (SVM). 

Very soon, “Intelligent Systems” such as Deep Blue at IBM, Eliza at MIT, and Dendral and Mycin at Stanford were being created and tested throughout the world. Thanks to the diligent efforts of tireless pioneers like Bruce Buchanan (Pittsburgh), Frank Rosenblatt (Cornell), Rodney Brooks (MIT) and countless avid research scientists, the field advanced and matured into today’s AI.

However, the roots of today’s AI, particularly its Data Science branch, may have been planted long before the1950s. In 1943, Warren McCullouch and Walter Pitts first proposed the “Artificial Neuron” concept through their research on “Linear Threshold Units” as a computational model of the complex “nerve-net” in the brain, employing a threshold equivalent of the “Heaviside Step Function.” According to their observations, any Boolean function could be implemented by networks of devices implementing simple AND / OR logic, and cyclic networks with feedback through these artificial neurons could be used to define dynamic systems with memory.

In the world of statistics, it is often claimed that the roots of today’s Data Science and AI go back even further - to the late 1920s and early 1930s. It constituted seminal research on “Statistical Pattern Recognition” for feature extraction and data processing, conducted by three brilliant scientists in UK and India (a British Colony at that time): Ronald A. Fisher in the British College of London, England and Prasanta C. Mahalanobis and Anil K. Bhattacharya both at the Indian Statistical Institute in Calcutta, India.

Fisher, an Australian-born scientist (elected to the Royal Society of England in 1929 for his pioneering work on Applied Statistics) had developed the theory of Linear Discriminant Analysis (LDA) to determine a linear combination of features that separates and classifies two or more classes of objects through dimensionality reduction. His LDA research ultimately led to his work on “Analysis of Variance” (ANOVA) and “Linear Regression” to express one dependent variable as a linear combination of other features. He also analyzed immense data from crop experiments since the 1840s and developed his theory of “Design of Experiments.” In 1936, he introduced the famous “Iris Flower Data Set” used by many AI and Data Scientists, as an example of Discriminant Analysis for large data.

During those years, Fisher often visited the Indian Statistical Institute in Calcutta, and its part-time employee, Prasanta Chandra Mahalanobis. In 1936, P. C. Mahalanobis introduced the concept of the “Mahalanobis Distance” which is a measure of the distance between a point P and a distribution D based on his statistical research in the medical field since 1927.  Today, Mahalanobis distance is widely used for cluster analysis and classification. It is closely related to Hotelling’s “T-square distribution” for multivariate statistical testing and Fisher's LDA technique for supervised classification.

Mahalanobis distance is often used to detect outliers, especially in the development of linear regression models. In data analysis, a point that has a greater Mahalanobis distance from the rest of the sample population of points is considered as having a higher leverage since it has a greater influence on the slope or coefficients of the regression equation.

At the same time, in the 1930s, a fellow researcher of Mahalanobis at the Indian Statistical Institute named Anil Kumar Bhattacharyya developed the concept of “Bhattacharyya Distance” which measures the similarity of two discrete probability distributions, and the “Bhattacharyya Coefficient” which is a measure of the amount of overlap between two statistical samples or populations. The Bhattacharyya distance is widely used in data science in many different areas such as feature extraction, image processing, texture classification, target tracking and voice recognition today.

It is known that Fisher who remained in England, was a strong advocate of the research of Mahalanobis and Bhattacharyya in India. Jointly, these three scientists in England and India: Fisher, Mahalanobis and Bhattacharyya laid the foundation of today’s Data Science and AI, through their diligent research almost 100 years ago - long before the invention of transistor and scientific calculator.

References:

·       Brooks, R. A. (1991). "How to build complete creatures rather than isolated cognitive simulators". In VanLehn, K. Architectures for Intelligence. Hillsdale, NJ: Lawrence Erlbaum Associates. pp. 225–239. CiteSeerX 10.1.1.52.9510

·       McCulloch, W. and Pitts, W. (1943). “A logical calculus of the ideas immanent in nervous activity”. Bulletin of Mathematical Biophysics, 5:115–133.

·       Fisher, R. A. (1936). "The Use of Multiple Measurements in Taxonomic Problems". Annals of Eugenics. 7 (2): 179–188. doi:10.1111/j.1469-1809.1936.tb02137.x. hdl:2440/15227.

·       McCarthy, John; Minsky, Marvin; Rochester, Nathan; Shannon, Claude (1955). "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence". Archived from the original on 26 August 2007. Retrieved 30 August 2007.

·        Bhattacharyya, A. (1943). "On a measure of divergence between two statistical populations defined by their probability distributions". Bulletin of the Calcutta Mathematical Society. 35: 99–109. MR 0010358.

·        Mahalanobis, Prasanta Chandra (1936). "On the generalised distance in statistics" (PDF). Proceedings of the National Institute of Sciences of India. 2 (1): 49–55. Retrieved 2016-09-27.

·        Mahalanobis, Prasanta Chandra (1927); “Analysis of race mixture in Bengal”, Journal and Proceedings of the Asiatic Society of Bengal, 23:301–333

·       Fisher, R.A. (1925). “Statistical Methods for Research Workers”. Oliver and Boyd (Edinburgh). ISBN 0-05-002170-2.