Chatbot: A History
Twenty years ago, I filed a patent application for an invention that would later be called an enterprise chatbot. And, before you all yell that Slack or Microsoft invented the modern chatbot, let's pause for a moment and review some history. Anecdotally, the term "chatbot" has been around since the 1960s and is a portmanteau of the words "chat" and "robot" to describe a program designed to simulate conversations between computers and humans.
The first contemporary use of "ChatterBot" was in 1994 and attributed to Michael Mauldin, who originally coined the term to describe a conversational agent called "Julia." This early chatbot competed for the coveted Loebner Prize, in the internationally known Turing Test. The Turing Test pits computer scientist judges against machines to see if they can distinguish a computer from an actual human conversation. While Julia was a milestone in the development of natural language processing, she wasn't designed to answer questions derived from enterprise data or systems like today’s modern chatbots. That's where my 2003 invention come in.
In 2003, Facebook had just been released, eBay was the hot auction site, and we were still four years away from the first iPhone. The height of technical sophistication was instant messaging on your Blackberry or web clients from AOL, MSN, Yahoo, and Jabber. Java was the programming language of choice, and web services were the de facto standard for interoperable system interfaces. My innovation was to combine these standardized technologies and make "automated customer service agents" that could answer questions from data stored in enterprise systems. My good friend and long-time collaborator Jon Bultmeyer built a prototype system, and I was left to find a snappy name for the invention. I called it the “Enterprise Buddy,” Admittedly, that phrase was not my best work.
You Made Million's Right?
Spoiler alert. No! But let's look at the invention in more detail before I explain why. The canonical example for my chatbot can be seen below in the illustration from the original 2003 patent application. The scenario uses an instant messenger client to retrieve the current temperature for a specific zip code by chatting to an enterprise buddy in real time.
On line 512, you can see that Jack typed:
And the chabot returned the answer:
Behind the scenes a web service decoded Jack’s query,
checked his access permissions and called a corresponding web service to
acquire the data. Then the final response was formatted and sent back to Jack
in a new message. The architecture and flow is below:
You can imagine more sophisticated conversations from this simple scenario. For example, our user Jack could chat to a “Stocks and Shares Buddy” by entering a ticker symbol. The agent could automatically return a company summary, current price, and 5-day average as a real-time response. Then Jack could decide to buy or sell a certain number of stocks by just chatting “!BUY 100” or “!SELL 200” and the chatbot would automatically take care of the trade. I believed this type of integration would be a productivity boon, whereby business people could instantly get the information they needed to make decisions or could initiate complex business processes with just a simple conversation.
Another interesting aspect of chatbots is that it instant messengers supported group chats, thereby mixing automated responses with group collaboration. In our example Jack could ask for a second opinion about a stock with people in the group before he decides to buy or sell. We take this kind of interaction between individuals, groups and data systems for granted today, but it was really revolutionary in 2003.
Barriers to Widespread Chatbot Adoption
If chatbots were so useful, then why didn’t they see massive
adoption? Here’s the lessons I learned with first user testing:
1. My interaction metaphor was like a Unix command line rather than a natural language conversation. By 2003 command line user interfaces for business users were definitely out of fashion.
2. The simplified conversational metaphor was also similar to telephone IVRs, where you “Press 1 for stock updates.” Even today people still hate IVR’s and want to chat to “real people.” Folks just don’t like using automated systems when they are black boxes and you don’t know the expected outcome.
3. Users really didn’t know what questions to ask the Enterprise Buddy despite the inbuilt “help command” that could be configured to list common commands, processes and data that could be invoked.
4. Additionally, folks couldn’t see the need for chatbots because they often asked questions where they already knew the answer. E.g., “!Top 10 Customers”, with the data being fetched from Salesforce.
5. Another issue was that 2003 instant messenger conversations were typically text-based with limited screen real-estate. Consequently, it was extremely hard to support conversations that required complex data formatting like tables or rich media.
6. Too early. Every invention needs the right product/market fit. I believed it was just too early.
Chatbots Didn't Die - Far from It
In the intervening years since 2003, the messaging industry would ebb and flow with the rise of every new collaboration invention that heralded "The end of email!!!!" In the Web 2.0 era (early 2010’s), chatbots started to appear on corporate websites as the first line of customer support. These little windows with smiley emoticons purported to be customer support agents that could answer your every question. Sometimes they were useful, but often they were a quick link to a real person. Consequently, adoption was good, but hardly ubiquitous.
And there the chatbot market plateaued until the arrival of new collaboration platforms like Slack, and later MS Teams, that offered plugin applications that could once again automate external processes or provide corporate data via conversations. Chatbot deployments, however, are still not broadly adopted outside of niche communities or departments. For example, developer communities use Slackbots extensively to automate and provide the status of the latest CI/CD builds, but you are unlikely to find departments like marketing, sales or finance using them at all. However, I believe this about to radically change.
In part two of this series, we’ll discuss how the emergence of ChatGPT completely changes the conversation around chatbots, and how different approaches to Generative AI will impact the future shape and impact of chatbots in the enterprise.