AI has become the need of the hour and all the industries are now integrating analytics and AI to drive the decision-making process. Bhagirath Kumar Lader, who is the Chief Manager (Business Information System) at GAIL led us through a session briefing Artificial Intelligence essentials for business leaders in today’s age. Lader is one of the key members of the digital transformation team at GAIL and carries huge knowledge about how AI, ML and DL are crucial to businesses. He gave us a quick overview of the motivation for AI, AI essentials, AI hype vs reality while taking us through use cases.
Motivation For AI
While AI is a crucial part of businesses, one of the key drivers of its implementation is its ability to make the decision which is usually considered the forte of humans. Integrating decision making systems in work take off the workload and AI makes it that much easier as it relies on data for its decision-making capabilities. “Where humans would have typically relied on gut feelings, the availability of data and using AI on it makes it more efficient and reliant,” said Lader.
Taking us through the early journey of analytics where it was first used in 1880 by the US Census Bureau to today’s age where it has reduced the time drastically in every process, he spoke about how analytics journey has evolved from descriptive and diagnostic analytics to predictive and prescriptive analytics. It now looks at the current and past data to find trends and patterns to help forecast the probability of situation occurring again in the future. He also spoke about how analytics also suggests decisions, actions and implications from predictive models to improve decision-making.
Wall In Analytics
While predictive and prescriptive analytics are now widely used, Lader shared how there is a wall that comes as a hindrance when it moves from descriptive to predictive analytics. “ While Business Analytics has many domains, it is highly important to understand each one of it to be able to break this wall and move into analysing correlations, root causes, forecasting and optimisation, which essentially are a part of predictive and prescriptive analytics,” said Lader.
How AI Differs From Humans
Once the above challenge is overcome, another crucial factor while designing AI system is to bring cognitive ability into the systems, which is nothing but the ability to do reseasoning, problem solving, planning, abstract thinking, complex idea comprehension and learning from experience — all of which are unique to humans. “Intelligence is the measure of cognitive capabilities, which if machines show can be considered showing artificial intelligence,” said Lader.
Lader addressed some of the commonly faced questions while adopting AI such as what are basic cognitive operations, what necessary conditions should a formal language fulfil in order to be an adequate tool for describing the world in s precise manner and unambiguous way, can reasoning be automated, how can we construct an AI system.
“While AI systems should be able to emulate human thinking, it should also be able to learn from experience, arrive at conclusions, understand complex real-world use cases, participate in natural-language dialogues with people, have cognition abilities among others,” he said.
He shared that AI can be created in four ways — thinking humanly, thinking rationally, acting humanly and acting rationally.
While explaining how machines can be made to act humanly, he gave an instance of the Turing test approach, where he developed an operational test for intelligent behaviour. “This test predicted that machine might have a 30% chance of fooling a person for five minutes,” he shared. It also suggested knowledge, reasoning, language understanding and learning to be the major components of AI.
The thinking humanly approach is the cognitive science approach meaning that making machine to think like humans, for which we need to get inside the actual workings of human minds.
Coming to the rationality part of it he said that AI system needs to have rational thinking where it knows the ‘right thing’. “While talking about autonomous vehicles we often talk about how whether it would be able to differentiate between an old woman and a tree. This is what is rational thinking which many systems currently do not pose,” he said.
Talking about the goal of AI, Lader shared that the goals are driven by two groups — expert system which should demonstrate intelligent behaviour regardless of their resemblance or non-resemblance to human intelligence and model human intelligence which is made with the aim to someday replicate or surpass human-level intelligence, which is still a distant reality.
Machine Learning & Deep Learning
Another aspect of intelligent machines is machine learning which is a category of algorithms that allows software applications to become more accurate in predicting outcomes without being explicitly programmed. Lader explained that for machines to learn there is a need for the generous sample for algorithms to learn.
The kind of machine learning models are:
- Supervised learning: some of the examples are linear and logistic regression, multi-class classification, neural networks, support vector machines etc. It can be classified into two categories of algorithms – classification and prediction.
- Unsupervised learning: It is a class of ML technique to find patterns in data without any label. Some of the most common unsupervised learning methods are cluster analysis, autoencoder, GANs.
- Reinforcement learning: It is all about taking suitable action to maximise the expected reward in a particular situation. The reinforcement agent decides what to do to perform a particular task. In the absence of training dataset, it is bound to learn from experience.
On the other hand, deep learning is a part of a broader family of ML methods based on artificial neural networks with representation learning. Some of the common DL algorithms are CNN, RNN, image recognition, among others.
“All these techniques are extremely crucial in facilitating the intelligent machines that we have today and are used to improve the performance of a system beyond that it provided by other analytics techniques”, he said on a concluding note.