In order to stay relevant and productive in today’s digital technological world, every company must be a technology first company, that means deploying solutions that customers expect in order to scale and remain relevant. The future of such solutions involve conversational AI. While enterprises are quick to adapt conversational AI in the form of chatbots and other forward-facing optimization tools, backend processes and c-suite capabilities are often overlooked. Instead, valuable time is spent parsing through reports, portals, and dashboards. Big data, cloud computing, and AI has throttled business to the next frontier of strategy and business intelligence. But all this data is bogging down the quick accessibility of internal insights.
With all the hype and over-saturation, it would be easy to believe that chatbots as we know them today have been on the tech scene for the better part of a decade. The reality is, it’s only been few years.
The first instance of a chatbot originated much earlier than that, however. In the mid-1960’s, deep within MIT’s Artificial Intelligence Laboratory, Joseph Weizenbaum was developing the first example of a chatbot, codenamed ‘ELIZA’. Utilizing pattern recognition algorithms, ELIZA was able to simulate computational understanding without actually having machine learning capabilities.
Ironically, Weizenbaum, widely regarded as one of the fathers of AI, developed ELIZA to showcase how superficial communication between man and machine really was. However, he was surprised to find how many testers attributed real, human feelings and the recognition of understanding to the chatbot, regardless of his rebuttals that ELIZA did not possess the capability for intelligent thought. ELIZA influenced many artificial intelligence researchers and pop culture references across the next half-century.
The true chatbot craze began in 2016 with Facebook’s announcement of a developer-friendly platform to build chatbots on Facebook messenger. Soon, chatbots were heralded as the next stage of the conversational revolution. Toolkits that helped you build a bot in five minutes grew popular, companies raced to the market with new bot announcements, and technology conferences headlined buzzword-driven keynotes about how bots would take over human jobs.
So how is a chatbot defined? What separates it from a more advanced conversational AI solution? Chatbots are primarily natural language text interfaces that are constructed using rules that encourage canned, linear-driven interactions. They are typically easy to build and navigated by predefined flows. For example, instead of clicking on a menu of choices or speaking predetermined commands, you can type or talk as if you were having a normal conversation in natural language.
While these bots can have very specific purposes and tasks that they can succeed at executing, there are two large underlying problems with this model. The first of which is the specific rigidity of the learning models. As we previously stated, chatbots primarily consist of canned, linear interactions based around pre-determined flows of conversation. This requires specific request input and very little wiggle room for the bot’s understanding of conversation.
Expanding on this thought leads to chatbots’ second core issue: they cannot learn. Chatbots have specifically-designed conversation flows and are typically not ‘smart’ enough to utilize previous conversations to establish contextual information. As a result, every interaction with a chatbot will seem more or less the same, because the chatbot will not have grown, developed, or learned in between conversations.
Most Chatbots are glorified flowcharts fumbling forth from IF/THEN routines and are lacking in Natural Language Understanding (the ability to determine intent, implied meanings) which requires higher level cognition.
Conversational AI automates written or verbal communication in a natural, human way. It also understands conversational human inputs. Today, AI has manifested itself in fairly linear ways: namely Machine Learning (ML), Deep Learning, Natural Language Processing (NLP), and Predictive Analytics. Each operates in silos or small sets to accomplish their specific, predefined tasks. But with the recent and exponential increase in computing power, in part due to cloud computing and the invention of fibre optics, the goal of conversational AI is to combine these AI tools into one seamless enterprise experience.
1. 8 Proven Ways to Use Chatbots for Marketing (with Real Examples)
2. How to Use Texthero to Prepare a Text-based Dataset for Your NLP Project
Unsurprisingly, we have a lot of good things to say about conversational AI. TLDR: Conversational AI performs multi-turn conversations and executes judgment-intensive tasks just like humans. As chatbots failed to deliver on expectations, the enterprise market in particular has turned towards conversational AI platforms, especially in complex use cases such as banking, insurance, and telecommunications.
What specifically separates a chatbot from conversational AI? We can break it down a couple ways:
It can talk, text, and chat: Conversational AI should be available on voice, text or web, and it should be ubiquitous and seamless across channels. It can be available through Alexa, Google Assistant or even your company enterprise portal. Truly omni-channel interactions are the future, and they should be a priority for your business.
It can learn: A conversational AI solution should be able to use the abundant history available from existing enterprise interactions, including chat and voice transcripts, transactions and other preexisting corpora of enterprise data to learn. What’s more, you need AI that can converse, suggest, recommend and engage based on these learnings.
It can understand: Beyond chatbot capabilities, conversational AI should understand complex sentences of human speech in the same way humans do. Real human conversation is never straightforward — it is full of imperfections consisting of slang, multi-string words, abbreviations, fragments, mispronunciations and a host of other issues. Conversational AI is a form of technology that can be used to both navigate and comprehend these give-and-take interactions.
It knows: By integrating into your enterprise systems, conversational AI should know who you are. It can reference the previous transactions made by you and try to fix things. It can then use this history to make current interactions smoother, troubleshoot or solve issues in customer service, IT or invoice processing.
It can transact: Conversational AI is secure and can support sophisticated enterprise security considerations. It can be used to complete complex transactions, replacing humans beyond a mere shopping cart click. Examples of transactions that true conversational AI can manage include buying life insurance, processing a healthcare claim, troubleshooting Wi-Fi issues or approving a supplier invoice.
Both the terms ‘Chatbot’ and ‘Conversational AI’ have the same meaning. ‘Conversational AI’, however, is more inclusive of all the technology that falls under the bot umbrella like voice bots and voice + text assistants, whereas ‘Chatbots’ have a more limited ‘text-only’ connotation.
While conversational AI is still young, an arena that mostly hobbyists and startups are playing in, it is currently robust enough for legacy companies to start incorporating into their solutions. It wasn’t long ago employees were reserving conference time by jotting down their name on a schedule taped to the conference room door. Then digital calendars came along, providing a way to reserve that conference room from their desk and automatically sharing that information across the company. With conversational AI, an employee needn’t even open an application. A quick wake-up cue followed by the request, “Reserve the conference room for next Wednesday,” will do the job. And that’s just one example. With conversational AI, employees also have the ability to search the company database for processes and procedure that they haven’t committed to memory. New regulations that affect a particular industry vertical work as a barrier to productivity. Time and energy is wasted on cross-referencing compliance procedures. With conversational AI, any question an employee has across the workflow can be asked and answered in minutes — no research necessary.
Digital assistants like Amazon Alexa, Apple Siri, and Google Home, are found in about 30% of American households. They are changing the expectations consumers have regarding their user experience (UX). In fact, it is reasonable to assume that the future of the enterprise user interface (UI) will be powered entirely by conversational AI. This means that conversational AI tools will be able to fill out webforms, carry out complex requests, and access online information through organic conversation. The enterprise applications of this technology means businesses can save time with the everyday management of its employees.
Take, for instance, the dreaded “I forgot my password” scenario. Instead of a several-step verification process — or a call to HR — conversational AI quickly processes a password change request without the need for additional human intervention.
Currently, at-home digital assistants work only as a model for enterprise digital assistants. It is important to note that Alexa is not yet a business tool. However, conversational AI tools can and are being built on top of digital assistant technologies. For example, IBM offers its Watson technology to enterprises, meaning the framework for a conversational bot is already there, which makes deploying enterprise specific conversational AI faster, easier, and more affordable — without the necessity of several data scientists.
Today, more people subscribe to messaging apps than social media apps. By meeting people where they are, conversational AI allows employees to interact with large databases and confusing dashboards in an organic, seamless way. With conversational AI, navigating through tabs and files is replaced by a simple request in an application the employee already has.
This evolution can be demonstrated by the noble dictionary. Hardcover dictionaries were first replaced by the world wide web. It took a full URL typed into the address bar to access www.dictionary.com. Next, with the advent of Google, it became even easier. The word in question simply needed to be typed into the Google search engine. No web address needed. Now, digital assistants like Alexa define a word with a simple voice command.
The first iteration of chatbots are commonly referred to as the first and second wave or generation. Conversational chatbot technology marks the beginning of generation three. Chatbots are a mainstay for online enterprises, but their abilities are limited. They operate on if/when algorithms, meaning they’re limited to their pre-programmed responses. If they hit a dead end, their script either loops, or, best case, a human operator takes over the interaction.
Conversational chatbots, on the other hand, are able to navigate human nuances. They’re trained through NLP to recognize the mood of the communications they receive, which gives them the opportunity to deescalate tense customer interactions and/or pull in a human representative sooner. They also remember preferences and personalities. For example, if a joke doesn’t go over well in a particular conversation, they’ll refrain from offering up another joke in the future.
Ultimately, the end goal of the third wave of conversational AI means that the bot will branch out into all arenas of human thought — including morals and a true understanding of the particulars and curiosities of life. For now, they still need a little help in those departments.
In other words, typical AI bots are pre-programmed to carry out human tasks. Conversational bots utilize artificial general intelligence (AGI) to think and act like a human. To be sure, we’ve just scratched the surface of AGI’s abilities, but their mainstream deployment for enterprise tasks is more of a “when” than an “if.”
The technology behind conversational AI isn’t fully developed and need human monitoring. Still, AI algorithm protocols are mostly standardized now, meaning customizable enterprise bots are hitting the market with a lower barrier to entry through vendor partners. First, let’s go over how conventional bots are taught, as intelligent bots build upon those teachings.
There are two main training buckets: retrieval based and generative.
Retrieval based bots — simply put — access giant decision flowcharts developed within its deep neural network in order to answer specific inquiries.
Generative bots, on the other hand, are programmed to conversate with their human counterpart.
However, all of their responses must be pre-programmed and trained with hundreds of thousands of real-life interactions in order to decrease incorrect or dead-end responses and increase its perceived level of understanding.
To make conversational bots self-learning, they must be able to build on the generative principles in order to adapt to client needs and backfill holes and missing information. Experiments have been conducted which allows users to train the bot on the fly, but the variability and unpredictability of the user’s training caused it to take on some less than desirable traits. To prevent moral mayhem, conversational bots currently learn through human-in-the-loop (HITL) means. While conversational AI allows Bot A’s output data to become Bot B’s training data, said data is first vetted by humans to reduce bias and incorrect information. Learning through user feedback is also semi automated. A user requests to the bot that it learn a new feature, inquiry answer, or even vocabulary word. That request gets sorted and dispensed to the HITL department where a human grants or denies the request. Similar to the way Wikipedia operates its verifiability model, a HITL conversational bot runs feedback from a customer by a human to verify and approve. As an aside, this demonstrates one of the many new job-creating avenues AI brings to the table. While AI replacing human talent is a valid concern, jobs of the future are becoming jobs of the present.
Intelligent conversational interfaces are the simplest way for businesses to interact with devices, services, customers, suppliers and employees everywhere. The rapid strides in AI and conversational solutions make it possible to carry on unique conversations with stakeholders at scale, delivering increasingly satisfying experiences that drive engagement and loyalty across a wide variety of business functions.
An Overview of Conversational AI. (N.D.). GeorgianPartners.com.
Asquith, M. (2019, July 8). Why Enterprise Chatbots are Essential in Today’s Business. Hubtype.com.
AXEL. (2019, January 23). Humans vs Robots: The Difference Between AI and AGI. Becoming Human. Becominghuman.ai.
Comes, S., Murphy, T., Vatatmaja. (2019, September 20). Conversation starters: Conversational AI makes its business case. Deloitte Insights. Deloitte.com.
Conversational A.I. and Corporate Strategy. (N.D.). Ibm.com.
Conversational AI Market by Component (Platform and Services), Type (IVA and Chatbots), Technology, Application (Customer Support, Personal Assistant, and Customer Engagement and Retention), Deployment Mode, Vertical, and Region — Global Forecast to 2024. (2019, March). Conversational AI Market. Marketsandmarkets.com.
Dashevsky, E. (2019, September 27). Why Your Business Needs True Conversational AI. IPsoft.com.
Davidson, L. (2019, August 12). Narrow vs. General AI: What’s Next for Artificial Intelligence? Springboard Blog. Springboard.com.
Diaz, J. (2017, December 12). The Three Different Generations of Chatbox Technology. Rulai. Rul.ai.
Haptik. (2019, February 20). How Does a Chatbot Learn on Its Own? Chatbots Life. Chatbotslife.com.
Mantha, M. (2019, March 28). Conversational AI: Design & Build a Contextual AI Assistant. Towards Data Science. Towardsdatascience.com.
Montaque, T. (2019, January 3). Here’s How Messaging Is Positioned to Dominate in 2019. Adweek40. Adweek.com.
Persiyanov, D. (2017, September 12). Chatbots with Machine Learning: Building Neural Conversational Agents. Stats and Bots. Blog.statbot.co.
Sunarjo, E. (2019, September 23). How Travel Brands Are Using Conversational AI. Icons8.com.
Training AI Faster with a Human in the Loop. (2019, October 16). Bitext. Blog.bitext.com.
Vincent, J. (2016, March 24). Twitter taught Microsoft’s AI chatbot to be a racist asshole in less than a day. The Verge. Theverge.com.
Voss, P. (2017, September 24). The Third Wave of AI. Becoming Human. Becominghuman.ai.
Voss, P. (2019, February 25). Beyond Chatbots: Hyper-Personalized, Intelligent Assistants. Forbes. Forbes.com.
https://avaamo.ai/chatbots-vs-conversational-ai/
https://www.boost.ai/conversational-ai-vs-chatbot
https://www.artificial-solutions.com/chatbots
https://georgianpartners.com/investment-thesis-areas/overview-conversational-ai/
Vartul Mittal is Technology & Innovation Specialist. He has 14+ years of strong Global Business Transformation experience in Management Consulting and Global In-house Centres with a remit to drive understanding and deliver Business & Operations Strategy solutions globally. He is always looking for new ideas and ways that can make things simpler.
He has lived and worked across multiple countries and cultures involving senior client stakeholders from various industries like Financial Services Sector, FMCG and Retail. He has delivered engagements for Fortune 500 organizations such as Coca Cola India, Kotak Mahindra Bank, IBM, Royal Bank of Scotland, Standard Life Insurance, Citibank and Barclays. Vartul Mittal is also renowned speaker on Analytics, Automation, AI and Innovation among Top Global Universities and International Conferences.