For an idea whose time has come, look no further than artificial neural networks (ANNs) and machine learning, the 2024 Nobel Prize in Physics seems to suggest. John J. Hopfield and Geoffrey E. Hinton have been honoured “for foundational discoveries and inventions that enable machine learning with artificial neural networks”. The foundations of ANNs are rooted in various branches of science, including statistical physics, neurobiology, and cognitive psychology, and artificial intelligence (AI) has become a household term today by drawing on such disparate insights. ANNs are networks of neurons (or processing centres) designed to operate like those in animal brains. In 1982, Hopfield, a towering figure in biological physics, introduced an ANN called the Hopfield network. Each neuron in this network is connected to all the others, and the flow and weight of information are not preferential to one direction. The neurons can process some input using Hebbian learning (“neurons that fire together, wire together”). The network as a whole was programmed to be analogous to a group of atoms, each with some magnetic energy. When ‘activated’, the ANN could receive, for instance, a noisy image and dynamically denoise it by minimising the analogous magnetic energy of the system. Similarly, the Boltzmann machine was an earlier model for a spin glass — a material in which roughly half of atom pairs have their quantum spins aligned while the other half have them anti-aligned. This disorder causes the material to be frustrated and minimise its energy through more configurations than if the disorder was absent. Alongside Terrence Sejnowski, Hinton popularised the use of Boltzmann machines for cognitive tasks, building on Hopfield’s work to enable them to classify data based on similarity or generate new patterns from old ones, again by having the ANN minimise the value of an energy function.
The ubiquity of AI owes much to the robust theoretical foundations laid by this year’s physics laureates and many others, drawing on mathematical, physical, and biological insights that few could have imagined would pave the way for AI. Herein lies a sting in the tail for India. Due to decades of low funding, inefficient governance, and inadequate attempts to reconcile the needs of science with bureaucratic processes, blue-sky research has often been a casualty of sudden and often transient bouts of consolidation and reform. Resource constraints may require researchers to conduct research as well as teach, guide, and administer. But as this year’s physics Nobel demonstrates, dismissing blue-sky research altogether is also to forfeit opportunities in technology that Indians may not even be aware of.
Published - October 09, 2024 12:10 am IST