Artificial Intelligence (AI) and Machine Learning (ML) continue to be highly-discussed topics for elevating businesses and their operations the world over. Edited excerpts from an interview with Mayur Datar, the Chief Data Scientist at Flipkart conducted by Amit Paranjape, Chairman of the IT and ITES Committee at MCCIA as a part of MCCIA’s YouTube series, ‘Decoding Artificial Intelligence with Industry Leaders’ Mr. Datar explains the concept of AI in simple terms, how it is being used by Flipkart to transform their business and how it can be applied to other sectors as well.
Mr. Datar begins by saying that, at its core, AI is about making sense of the data and forming patterns to be able to take decisions about various situations. This could be simple tasks like cognition where, through data, machines, that are better than humans at number crunching, can deal with lots of data and form patterns similar to how a baby, when presented with a bunch of examples, learns over time to distinguish between an apple and an orange. These patterns are used in automated decision-making and optimization which form the core of AI and ML.
According to him, the foundations of AI and ML can be traced back to the 1930s and 40s and found in what we call statistics today. Techniques like linear aggression have existed for a long time and to this day, are being successfully used. However, there is a lot of excitement in this field since the last couple of years because of three main reasons- computing hardware and storage becoming cheap, the resurgence of consumer internet companies like Google, Facebook, Amazon, etc. and the reinvention of the old algorithm, by transforming neural networks into deep learning.
He states that initially the neural networks had about 100 neurons and were about 3–4 layers deep. Around 2010, Prof. Geoffrey Hinton, from the University of Toronto made these networks much bigger by adding millions of nodes, billions of parameters, and hundreds of layers. As a result, the neural networks started doing non-trivial things. Some of the early successes with neural networks were in the field of image sciences where they could see an image, make sense out of it, and tag it, and so on. Today, image analysis is used in different fields. For example, some startups look at images of fruits and vegetables, automatically sort them, and price the vegetable for you. This saves a lot of resources as it is not possible to write a custom algorithm for each vegetable. Interestingly, the concept of facial recognition came to be implemented through custom algorithms before the era of deep neural networks or deep learning as it’s called. However, today the algorithm has been replaced by deep learning Additionally deep learning is successfully applied to many areas including speech recognition, neural machine translation, comprehension, etc.
He moves on to explain how AI is extensively used in e-commerce. The four main categories where it is applied are-
1)Cognition: This includes Optical Character Recognition (OCR), image, and video analysis. For example, when someone uploads an image on the e-commerce site, it goes through an entire pipeline of moderation, where it is checked for quality, resolution, etc. while making sure it doesn’t contain any objectionable content like profanity/nudity and doesn’t violate any government norms.
2)Language Understanding: The biggest application area for this is chatbots. Understanding the intent of the customer in a specific context when he enquires about his order comes under natural language understanding. Another example of the same is the search query understanding. When people type in keywords in Google or long queries in our search box, AI is used to make sense of that, which is part of language understanding.
3)Classical Predictive Analytics or Classification — AI is used to classify things into 2 categories-good and bad. For example, every second, millions of credit card swipes happen across the world, and depending on the context of the credit card transaction, its location, amount, etc. it is decided if it is an authentic or a fraudulent transaction.
4) Optimization — It is similar to operations research, where AI determines ways to stock and place products in the warehouse so that the time taken to pick up products that constitute a certain basket is optimized.
When it comes to the small and medium companies, he believes that they can also benefit by using the Internet of Things (IoT). Predictive analysis can be used to figure out whether a piece of machinery is performing well, whether it needs maintenance etc. thereby, reducing the downtime of it or reducing the servicing costs of it. Companies can employ image analysis to monitor shop flows and productivity of their employees.
Predictive analysis can also be used by hospitals to figure out which patient can turn serious and hence needs more attention. Healthcare and insurance industry could use it to understand which treatments provide better outcomes. It could help to make decisions about whether surgery should be postponed and compare the outcomes of postponing it and not postponing it. With enough data, based on the symptoms and other things, the machines could make such decisions for us.
When asked about how the industry is focusing on data science, he says that the industry has underinvested in data engineering. He believes that the biggest challenge is to collect and store data in a proper way. He further says that understanding SCALA, Hadoop, underlying computer architectures like TenserFlow, and being able to deploy them are some of the skillsets that would be required from a data scientist.
In conclusion, he thinks that by 2025, AI will be applied much more in the field of cognition and packaged into the right kind of robots. Processes will be automated, doing some very small yet well-defined concrete tasks. While the self-driving car will be a reality, it may get delayed as there are many challenges outside the technological point of view in its implementation.
(You can watch the full interview at https://www.youtube.com/watch?v=BuuOLZihnKU )
