"By far, the greatest danger of Artificial Intelligence is that people conclude too early that they understand it." This powerful quote from Eliezer Yudkowsky, a renowned American AI researcher credited with coining the term "friendly artificial intelligence," perfectly captures a core challenge in the field. While the debate over AI's "friendliness" is best left to the experts, this article aims to demystify some fundamental AI concepts and terminology, offering a simplified look at how this complex world of AI operates.
The concept of Artificial Intelligence (AI) gained its name in 1956, coined by John McCarthy. However, its foundational ideas emerged even earlier, in 1950, with Alan Turing's seminal paper, "Computing Machinery and Intelligence," which introduced the renowned Turing Test—a benchmark for machine intelligence.
The journey of AI research hasn't been smooth, marked by significant setbacks like the infamous "AI Winter (1970-81)." During this period, AI development nearly halted due to a perfect storm of challenges: severe funding cuts, dwindling public interest, limited computational resources, and an unclear vision of AI's true potential. Yet, the 1990s brought a resurgence of interest and investment. This renewed focus ignited a rapid acceleration in research, paving the way for breakthroughs in Machine Learning, Deep Learning, Neural Networks, and more recently, Generative AI, with countless innovations still on the horizon.
The main goal of AI scientists is to develop algorithms that can imitate human learning patterns and to increase the output accuracy of their algorithms. An AI model is the piece of code that implements the required algorithm and trains the algorithm on the provided training data set. Before the model is released to the end users, its performance is bench-marked against a test dataset. If it satisfies the required precision criteria, only then it is sent to further use. The algorithms are exhaustive mathematical expressions that undergo frequent readjustments during the training phase until they start making predictions closer to the original output. This learning from the training dataset can be compared with a child trying to learn mathematics. Just as a child is made to practice math to increase his precision and reduce mistakes, the model is trained on the training dataset several times to increase its accuracy and precision. AI models learn from data which are analogous to real life experiences for the human brain.
Further, the scientists have implemented AI models called neural networks that imitate the Human Brain. These neural networks even have micro-units called neurons which replicate the human brain cells. The neural networks also called as Artificial Neural Networks (ANNs) have performed extremely well in learning complex patterns from humongous datasets and have paved the path for ground breaking research and development in AI. Still I believe that the world of AI has a lot to offer - be it in healthcare, finance, generative AI, autonomous vehicle and navigation or to the uprising world of Robotics.