Artificial intelligence is a buzzword that is thrown around like candy. We see it being used as almost a marketing gimmick in most cases, without realising that what is being passed off as AI are actually just pre-programmed responses to certain inputs, shorn of any contextual learning or intelligence.
One of the areas this is seen quite often is with chatbots. If you have interacted with any customer help-related chatbot, you know that the conversation follows a certain pattern. If you ask queries only in a particular manner or with a particular string of sentences, pre-programmed responses will pop up on cue. If you depart from the script and try to communicate in natural language, the veneer of AI falls apart, and you are left tearing your hair in frustration till a human agent takes over.
PV Kannan, the co-founder, and CEO of 7.ai recognised this problem early on and founded his company in 2000 to ensure better customer service for consumers using a mix of contact centre agent services and big data predictive analysis. 7.ai boasts of more than 100 patents related to artificial intelligence, user identification, natural language, intent prediction, and related technologies.
Kannan recently penned a book, The Age of Intent: Using Artificial Intelligence to Deliver a Superior Customer Experience. The idea behind the book was to explore the importance of AI when it comes to delivering a business impact. According to Kannan, AI can’t just be a science experiment, it has to deliver a business impact, and the best way to do that is by applying AI to customer service and support. Kannan felt that a lot of businesses implementing AI were failing to exploit its potential to understand customer intent to provide optimal solutions. The primary motivator for the book was helping business people to understand how to apply AI to deliver superior customer experience.
Customer intent, i.e., what the customer hopes to achieve from a particular chat conversation, is an area that 7.ai focuses on. In its ‘Conversational AI for Customer Service, Q2 2019’, Forrester selected 7.ai as the leader among 14 other participants when it came to implementing conversational AI. We spoke to Kannan about the philosophy behind customer intent, intelligent chatbots, AI explainability, AI regulation and much more. Edited excerpts from the interaction follow.
Tech2: Chatbots and assistants are slowly becoming more and more mainstream, but have you seen a trend in them becoming more intelligent? In most cases, you have to follow a particular script when you interact with chatbots, if you say anything outside it, the chatbots get confused? Your thoughts?
PV Kannan: Enterprises have experimented with chatbot and virtual agents to improve customer experiences and achieve cost savings. However, many companies have experienced the serious limitations of first-generation technology. Those legacy chatbots were not intelligent, and could not understand consumer intent, and could not conduct a natural conversation with context. Using technologies like AI and natural language processing however, a new generation of technology can interact with consumers in a natural way — in the same way as a company’s best human agents. Additionally, humans and bots can now work together. An agent can take over a bot conversation at any time, and hand the conversation back to the bot to complete the interaction. Through open APIs, these intelligent bots can integrate with voice assistants like Google Home or Amazon Alexa, and messaging technology such as Apple Business Chat, Facebook Messenger, Google Business Chat or WhatsApp.
Tech2: What are the trends you have seen in the field of AI over the years since you formed your company?
Kannan: In recent years, AI is becoming more democratic. There have been a lot of efforts to make it more accessible to people who aren’t data scientists. We are seeing more development tools, and modelling tools that open up the technology, and we are seeing AI technology become a lot better. Examples include speech recognition, natural language processing and intent classification, which have really improved over the last decade. For some domains, AI is approaching human-like performance. For example, with speech recognition, it is becoming increasingly difficult to tell whether you are speaking with a machine or a human.
Tech2: We keep hearing how AI algorithms are like a black box. There is a lot of talk around Explainable AI and how algorithms should be open to scrutiny. Don’t you think that would reduce the incentives to innovate with AI if the algorithms underlying it are subject to inspection?
Kannan: The point of Explainable AI is to make sure that companies are able to explain how decisions are made. Most algorithms are well known throughout the industry, so it’s not as much about the algorithms as it is about the data set. In most cases, it is the data set that is proprietary. As such, I don’t think that inspection of algorithms reduces the incentives to innovate. If anything, it means you have to be even more careful to ensure that the data you use is representative of the population to ensure there are no biases.
Tech2: There is a lot of talk around AI regulation. In a world where we have China operating in a silo as far as the general internet is concerned, how likely do you think there will be a global consensus on AI?
Kannan: First, there is a big difference between AI and the Internet. AI is a technology, and since algorithms are very common, it all comes down to the data. The world is converging on how people will control their data, and in the Western world, there is largely consensus on this. You see governments regulating the data that is used for privacy reasons. However, that consensus may never extend to every country. In China, for example, it’s a different issue because the government controls which data is used.
Tech2: AI bias is real. Most of the cutting edge research happening in training AI models is happening abroad, where the databases aren’t really presenting all the varied data points one can hope for. India is nowhere in the top 10 list of countries when it comes to AI patent applications — which means the AI models trained on data in India are limited. What is the solution to this disparity?
Kannan: Because the algorithms are published, if someone wants to apply them to an Indian data set, they can. It all comes down to the business case for doing so. If companies see an opportunity to make money by applying this to the Indian market, that is what will drive the solution.
Tech2: What are your views on the whole ‘AI vs Future of work’ debate? In India, it is well known that a lot of manual jobs are done by quite a lot of people — for instance, 10s of people manning a toll booth, to data entry operators, and so on. How will India be prepared to face the loss of jobs that AI will eventually take over?
Kannan: The reason why AI is not being rolled out as much in India as in other markets is because the cost of labor is inexpensive. It’s often less expensive to do things with humans than it is with technology. Over time, however, the point is to automate the less skilled, more routine jobs so that the working population can gravitate towards higher-skilled jobs. This may take longer in India because of the cost of labor, but eventually, there will be higher-skilled, higher-paying jobs, and AI will drive a lot of that. Despite these changes, there will always be a role for humans, particularly in providing great customer service.