How do you predict demand for a consumable product during uncertain times such as now, with the COVID-19 pandemic? Using Artificial Intelligence (AI) is one way to do it. Subram Natarajan , chief technology officer, IBM India, says AI technology has come a long way in the past two decades. With computing power and storage becoming unlimited, adoption of AI at scale is gathering pace. Excerpts:
What has led to AI suddenly gaining attention in the recent past?
Two things have happened in this Chapter 2.0 of enterprises – every enterprise is becoming more cognitive. AI has been around for decades, but has come of age by means of how we deploy it and we are seeing early results. When people started deploying them a decade ago in their peripheral workloads, experimenting with AI, collecting promising results, that’s when gears got shifted into applying AI at scale. AI became the anchor for enterprise decision-making forces. Two, the pandemic helped accelerate the adoption to a scale that we have not imagined. The whole idea of a conversational engine has been existing for a number of years.The first major introduction of AI as a conversational engine happened years ago in one of the game shows — Jeopardy. The challenge was to formulate a question for a given answer. The game touched upon various topics, from economics to IT to literature to arts.
Watson – one of the AI engines for IBM – was aimed at understanding human speech, make sense out of it, gain an understanding of the intent behind it, and then come up with a response that complied with the rules of the game. Ultimately, human contestants won. But it was a huge awakening to how computers can start interacting with human beings, not just the keyboard to mouse level, but with understanding speech, the intent behind your speech, understanding tone, whether you’re irritated, angry, happy… all of that is encapsulated in natural language processing, which is one of the core disciplines of AI. We have come a tremendously long way in making that happen.
Can you elaborate on the work you’re doing with ICMR?
When COVID happened, the Indian Council of Medical Research began using our Watson virtual agent as a conversation engine. In their portal, it responds automatically to specific queries of the frontline staff and data entry operators from testing and diagnostic facilities across India.
So, now you see deployment of this at scale. Even though technology has been around for a while, it needed a COVID to really bring such technology into the mainstream.
This is helping boost ICMR’s response time and allowing them to coordinate between different groups across the country.
And because the AI engine is taking care of the response to calls, they can go on to concentrate on priorities such as developing and updating the testing and treatment protocols for COVID-19.
ICMR is mandated with setting up policy, making sure the communication is done and that the right procedures and data protocols are followed by all of medical staff across India.The biggest trouble is that in a pandemic like this, you start getting deluged with information plus the need for dissemination. Imagine, almost all of the testing agencies in India would want to know from ICMR what their policies are and what they should do next, in terms of procedures, protocols and guidance on how to deal with a bunch of information.
Much of this information could overlap. A diagnostic institute in Chennai could be asking the same questions that an institute in Bangalore or Mumbai may be asking. So, why not make use of an artificial intelligent engine that tries to understand your question and comes up with the right guidance for protocols?
The technology we use is the Watson conversation engine, or the Watson virtual agent, also called as the Watson Assistant. While the deployment in ICMR is highly impactful, this could extend to any call centre – be it for a bank or a telecom company. People ask questions that others too have. The same question can be asked in different tones. The Watson Assistant understands what you’re asking, the intent behind it and can come up with the right kind of response.
Following ICMR, some State governments have wanted such assistance. The AP government has already introduced Watson Assistant in their citizen services.
Has data quality improved?
Yes. More than 60% of the time data teams are just working on making the data available for the AI and ML. When you fail to do this, you get into a bad data situation; therefore the AI that results is going to be poor quality.
There is no AI without IA – the information architecture. You need the right kind of collection of data, the right quality, you must make sure you organise the data in a manner that can be consumed, which means cataloguing, and making sure that data comes from a single source of truth…
Organising this whole process is what I call Data Ops. You don’t build start building the AI unless you have the foundational pieces built.
How have you helped Parle predict demand during the pandemic?
Watson AI was used to predict demand. It helped reduce time to market. Because of the intelligence in the supply chain, with the just-in-time technique, the client could manage inventory smarter so that costs declined.
With COVID coming in, influencing factors went haywire. It was a new normal, so we needed post-COVID data.
Data seeding post-COVID and intelligence coming out of the system is far more accurate than one can think of. Parle has used this very effectively, especially for goods with a short shelf life.
The way they have done that was to feed a new set of data and start teaching the model how the new demand and supply would look like with the disruption coming in. Is it as accurate as pre Covid? No. But is it better than what humans can predict, the answer is ‘absolutely yes’.
