Traditionally, education in India has been teacher-centric. I too studied in an environment in which there were only two kinds of teachers – the good and the bad. It seldom occurred to us that this tight-fisted categorization could be flawed or at best contextual. Teachers followed a certain pattern & pace often omitting segments due to oversight or a presuppositionbordering on vagueness that seemed to suggest the “topic” wasn’t important. It was they who decided what was important and what wasn’t. This methodology has its merits too for young and impressionable minds. Knowledge, if it looks infinite, can be overwhelming, and probably why we have things like chapters, syllabus, etc. to give it a sense of finiteness & comprehensibility.

The challenge in India is unique. On the one hand, there are vast numbers that are out of schools due to poverty and on the other, there are teachers – again in very large numbers – who aren’t competent enough. They are a product of a system thathas proven to be inadequate. This has got the enlightened lot – policy wonks, tech innovators, intellectuals, etc. – to think if this challenge can be addressed differently through technology. The answer is a resounding yes; but then again, tech in education is not a new idea at all, and that in itself is avastness that stares right back at us waiting to be adopted. Insofar, this column will be restricted to AI bots.

AI Bots

Since 2016, there’s been a great surge in chatbot applications, particularly in the customer-facing processes and not least due to the kind of heft demonstrated by the likes of Amazon, Google, Apple, and Facebook. The idea of conversational interfaces has a long history and goes back in time – Weizenbaum’s work on ELIZA in the 1960s. But, the more recent developments in AI & ML makes it a very powerful tool. Automated helpdesks today use a lot of this technology to augment human productivity. 

Arguably, one of the best examples of tech democratization is chatbots – simple to use yet complex to develop and can be a powerful enabler in areas where there are bottlenecks due to talent availability. While serving the social sector, companies need to look beyond the element of profitability to be driven by the purpose of greater common good.  

Pune City Connect and Social Venture Partners have started an experiment to drive inclusivity in education. Students are assigned a bot when they enter the institution. It helps them in skill-building and subsequent job search as well. The bot also assists in connecting with mentors at a later stage. Intelligently-designed bots are an important part of learning and development in companies too. They can interact with users through natural language programming to make quick decisions. The conversational aspect can actually be “better” than a human interface because it’s available 24/7 and isn’t constrained by cognitive bias. 

Bots in the learning space are generally the rule-based ones and there will be instances when the answers may not be forthcoming and it should be able to direct the conversation to a human. But within the guardrails of a specific use-case, they can certainly outperform humans in terms of content. 

Inside a Bot – a Primer

Chatbots, not unlike regular applications have application layers, databases, conversational user interfaces (CUI), and APIs. Essentially, there are three kinds:

• Rule-based chatbots are basic. Answers are pre-defined to address specific questions. They are very effective in answering FAQs.
• Intelligent chatbots are trained to understand specific words or phrases to trigger a reply. They are self-learning too and over time, using ML, they can deliver better answers.
• AI-powered chatbots combine with intellectually independent programs to solve complex problems. Besides, NLP, AI & ML work in unison to give the impression of natural interaction as one would have with a human. 

NLP involves 2 processes – Natural Language Understanding (NLU) and Natural Language Generation (NLG). The former tries to understand a human by converting text into structured data for a machine to understand. While the latter transforms structured data to text. 

Let’s look at a very basic example:

What is the weather in Delhi today? 

The chatbot breaks down the sentence into intents and entities. An intent is a request from the user to fetch some information. An entity is a detail that compliments the entity. Here, the intent would be “weather” and the entities “Delhi” and “today”. Here, we need to also mention Sentiment Analysis – the ability to understand emotions and moods. This is a sensitive area because the last thing a harrowed customer wants is gibberish in terms of output which makes no sense. A dumb computer can certainly do more damage than a dumb individual in such a scenario.

Of course, it’s not that easy because if it were, a bot would be writing this column. On second thoughts are you 100% sure that it hasn’t? 

The whole idea is about speed and convenience. It may very well be that a website has all the information that I am looking for but it would take some time and effort to discover it. It would be so much more comforting to have a human-like interface do it for me in real-time. 

Developers have to keep a few things in mind. Security is of course one. A chatbot should only ask for relevant information. And, be able to transmit data securely. In India, there’s also a great need to have vernacular chatbots. And finally - dour-faced humans are unavoidable especially if they happen to be your friends but the same cannot be said of a machine interface. For greater adoption, it has to be likable with an element of humour as well. 

Tall order? Really no.

Image by Jams Royal- Lawson via Flickr