Skip to content
Featured

Part 3: The Secret History of AI in Customer Service

Now that I’ve spent some time going over the relevant terminology around AI, I’m going to talk about how the recent advances in AI need to be seen less as cataclysmic “black swan” style disruptions and more as the gradual culmination of technological and scientific advances that took place over many years, often in tandem with a broader shift toward automation that began to unfold across several industries at the turn of the twenty-first century.

The First Customer-Facing AI Applications Were Actually Deployed Decades Ago

To understand the impact of AI on customer service, business owners need to understand that AI has been playing a behind-the-scenes role in customer service for much longer than most people realize.  As I’ll discuss in this section, the earliest AI technologies were deployed by customer-facing businesses almost fifty years ago.  Still, it’s only in the last ten years that the most dramatic advances have been made.

The 1970’s: The First ACD and IVR Systems Are Introduced

The first example of a mainstream commercial application for AI dates back to 1973 when Continental Airlines purchased an ACD (automatic call distribution) system from Rockwell International Corporation. The ACD system, known as the Rockwell Galaxy, allowed customers to make airline reservations over the phone.  The system permitted Continental to handle a huge volume of calls by sorting them into queues and directing them to the right agents.

Another technology that developed in tandem with the earliest ACD system was the IVR (interactive voice response) system.  While the earliest IVR systems couldn’t understand human speech, they could prompt callers to select from a menu of choices (“to speak to a booking agent, press one”), at which point the ACD would send the caller to the correct queue based on the option they chose.

These technologies eliminated the need for a human operator at the front desk and allowed customer inquiries to be routed to the correct department with minimal human assistance.  Besides allowing consumers to complete specialized commercial transactions from home, the earliest ACD and IVR systems also streamlined the number of calls businesses received.  They allowed customer service teams to focus on complex problems rather than routine inquiries.

First-generation ACD and IVR systems might seem quaint by today’s standards since they relied on rudimentary technologies like decision trees and, in the case of IVR systems, could only respond to dialed inputs.  Still, at the time, they represented a huge step forward for customer service.  These systems helped businesses reduce their operating costs and provide seamless customer experiences.

The 1990’s: Online Customer Service and the Earliest Chatbots

In the late 1990s, online customer service entered the business mainstream.  The shift from a traditionally brick-and-mortar customer service experience to a digital one was driven by the growing number of people using the internet to get their news, research, and purchase goods and services.  The shift to a digital customer service paradigm was most pronounced in technology, banking, and e-commerce.

Since more consumers were willing to purchase goods and services online, retailers and service providers became aware of the need to provide the basic elements of the customer service experience online.  During this period, the first online retailers began to put the foundations of the online customer service experience into place—FAQ lists, e-mail contact forms, and the ability to speak to a customer service agent via live chat.

Consumers could now request refunds, track packages, and perform online side-by-side comparisons of different items.  By the early 2000s, most online retailers had deployed the earliest chatbots and virtual agents to handle simple customer service inquiries.  Some notable first-generation chatbots from this period include ActiveBuddy’s SmarterChild and the butler chatbot on the Ask Jeeves search engine.

2011: AI Takes a Big Step Forward When Apple Introduces Siri to the World

AI hit another milestone in 2011 when Apple introduced the voice-controlled virtual assistant Siri, which utilized the most advanced AI and natural language processing algorithms available at the time. For the first time, consumers had an AI-powered assistant at their fingertips that allowed them to set reminders, get directions, and surf the web.  Siri was originally developed as a stand-alone app available from Apple’s app store.

On October 4, 2011, Apple released the iPhone 4S, which included a beta version of Siri.  This version of Siri spoke with a distinctive female voice provided by Susan Bennett, who at the time already had a distinguished career as a voiceover artist and a backup singer for Roy Orbison and Burt Bacharach.  The earliest reactions to Siri were mixed at best, and the virtual assistant had myriad problems, some of which persist to this day.

iPhone users who gave Siri basic commands had to speak slowly and clearly enunciate each syllable.  Siri also struggled to understand regional accents, so iPhone users in some areas of the U.S. struggled to get Siri to work.  Apple’s “walled garden” philosophy toward app development meant that Siri often didn’t work with third-party apps, and when many iPhone users tried to activate Siri by saying “Hey Siri,” the virtual assistant wouldn’t respond.

At times, the complaints about Siri took on political overtones when the ACLU criticized Apple after users reported that Siri would not give out the location of birth control or abortion providers. Into the present day, Siri’s capabilities have improved considerably—Siri has been integrated across Apple’s entire product line, and the virtual assistant can understand regional dialects and answering follow-up questions.

2016: Facebook Messenger Allows Businesses to Integrate AI-Powered Chatbots

In 2016, Facebook (now Meta Platforms) opened its Facebook Messenger platform to chatbot development, enabling businesses to integrate AI-powered chatbots directly into the Facebook Messenger app.  These AI-powered chatbots gave businesses a new way to engage directly with their customers on the platform. The chatbots could place orders, make reservations, provide personalized product recommendations, and handle basic customer service inquiries.

In the past few years, chatbot integration on the Meta platform has continued to improve.  For many businesses on the platform, the AI-powered chatbot has become a standard tool in the digital marketing toolbox, with capabilities that go far beyond customer service inquiries and reservations.  AI-powered chatbots can build brand awarenessidentify potential leads, and collect valuable user data.

AI-powered chatbots are now increasingly commonplace as customer-facing applications beyond Meta and across various industries, providing personalized support for websites, messaging apps, and other social media platforms.  With continued advancements in machine learning, deep learning, and natural language processing, AI-powered chatbots help companies cut costs and engage with their customers.

2020: The First Generative AI Applications are Released to the General Public

In the last few years, AI applications have made the leap from voice-activated personal assistants and chatbots to generative applications, ‘generative’ meaning that these applications can generate straightforward answers to questions posed by humans.  In 2020, the San Francisco-based AI firm OpenAI released ChatGPT-3, and in less than a year, thousands of developers were integrating the new technology into search engines and other applications.

At the time of its release, GPT-3 utilized the largest neural network ever built, with 175 billion distinct parameters and requiring 800 gigabytes to store. GPT-3 could search the internet and then use the retrieved information to respond directly to typed user queries with direct answers in simple, everyday language.  Reactions to GPT-3 were positive, with the New York Times writing that GPT-3 could write original prose equivalent to what a human could produce.

Advances in generative AI continued beyond text responses to images, audio, and captioning.  In April 2020, OpenAI released Jukebox, a generative AI application capable of producing music and vocals.  However, it should be noted that Jukebox is challenging to use and still hasn’t found a mainstream audience. In October 2020, Microsoft introduced Vivo, which could produce text captions for meetings, images, and other uncaptioned media.

In 2021, OpenAI released DALL-E, which could render images from text inputs.  AI lab Midjourney, Inc. followed suit in 2022 when the company released Midjourney on the Discord social media platform. These applications deployed similar technology, with DALL-E generating images broadly described as photorealistic, whereas Midjourney generated images in a more dreamlike aesthetic.

Many Generative AI Applications Continue to Struggle with Context and Accuracy

At the time of this writing, AI technology continues to transition into the mainstream.  However, despite all the significant advances in AI that have been made since 2020, especially in the realm of customer service, serious problems remain.  The biggest problem is that the answers supplied by AI text generators, such as ChatGPT-3 and the more recent ChatGPT-4, are frequently incorrect or lacking in essential context.

Type “ChatGPT-4 is inaccurate” into Google, and you’re confronted with numerous reports from credible media outlets, like this article from Fast Company, or this piece from the New York Times. They paint a picture of generative AI applications that are still likely to return patently false information.  The problems with ChatGPT-4 are so pervasive that OpenAI CEO Sam Altman posted a warning about relying on it too heavily on his own Twitter feed.

And it doesn’t get any better from there.  The AI tool that Microsoft recently integrated into its Bing search engine made numerous cringe-worthy errors during corporate demos—failing to distinguish between corded and cordless vacuums and making several inaccurate statements when it was asked to summarize a straightforward quarterly financial report posted online by clothing retailer the Gap.

Generative AI Applications Have Been Implicated in Numerous Copyright Infringement Cases

As if all these issues weren’t problematic enough, several AI image generators have recently been at the center of legal disputes over copyright infringement.  In January of 2023, the London-based AI lab StabilityAI was named as a defendant in two separate lawsuits.  One of these lawsuits was brought by the online stock photo repository Getty Images, and the other was brought by the artists Sarah Andersen, Kelly Mckernan, and Karla Ortiz.

The contours of both lawsuits are broadly similar.  Both legal motions claim that Stability AI’s image generator, Stable Diffusion, illegally scrapes the internet to generate its images and, in doing so, infringes on the intellectual property of Getty Images and millions of other artists.  The lawsuit against StabilityAI by Sarah Andersen et al. casts a broader net, naming Midjourney and another AI image generator, DreamUp, as defendants.

Copyright infringement cases like these haven’t been confined to the developers of image generators.  In 2022, the same law firm retained by Andersen et al. in the case referenced above, the Joseph Saveri law firm, was also retained in a separate copyright infringement case brought by the typographer and attorney Matthew Butterick. This legal motion names Microsoft, Github, and OpenAI as defendants.

In this motion, Butterick claims that all three companies have deployed Microsoft’s AI-powered productivity app CoPilot to illegally scrape open-source code from the internet.  At the time of this writing, all three cases referenced here are still winding their way through the courts.  Still, however they shake out, cases such as these represent what are most likely only the opening salvos in a much broader debate over AI and copyright infringement.

AI is Also Causing Turmoil in Hollywood and the Publishing Industry

AI has emerged as a central issue in the recent wave of strikes that brought Hollywood to a standstill for almost five months earlier this year.  Actors, writers, and other relevant players in the entertainment industry fear that AI will replace them entirely.  As part of their conditions for ending the strike, they demanded a strict regulatory framework for how their writing, voices, and likenesses would be used to train AI applications in the future.

In September, a group of 17 authors that included some of the biggest names in publishing (George R.R. Martin, George Saunders, and Judi Picoult, among others) joined a massive class-action lawsuit against OpenAI alleging copyright infringement.  Specifically, the suit contends that OpenAI illegally trained its large language models on copies of the authors’ texts illegally downloaded from sketchy e-book websites to avoid paying the proper licensing fees.

At the time of this writing, it remains to be seen what kind of regulatory framework Hollywood will put in place for AI, and the class action lawsuit against OpenAI is in its earliest stages.  However, it appears the company engaged in some extremely questionable behavior to train its large language models.  These cases are instructive because they demonstrate that AI technology is error-prone and operates on shaky legal and ethical grounds.

Artificial Intelligence Will Change Customer Service by Automating Routine Tasks

So far, over the first three blog posts in this series, I’ve discussed the current state of the digital marketplace and customer service in 2023 to highlight the specific economic and technological context in which the latest iteration of AI was making itself known.  I’ve also provided definitions for the AI-related terms you’ve seen floating around in the media.  And now, I’ve provided a detailed history of AI and its role in customer service.

If you’ve read this far, you should know that AI didn’t magically appear sometime around 2019.  It’s been a quiet force in mainstream commercial applications since the 1970s, but it’s only in the last few years that the technology has taken some astonishing leaps forward.  Up to now, most of the information I’ve provided has been contextual. For the upcoming fourth post in this series, I will narrow my focus somewhat.

The fourth post in this series will discuss artificial intelligence and its ongoing role in automating many routine customer service tasks—AI-powered chatbots capable of handling regular customer inquiries algorithms that can predict customer behavior based on browsing history, to cite just two examples.  Stay tuned for my next blog post, and I’ll show you how innovations like these can streamline your operations and keep your customers happy.

Featured

This Post Has 0 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *

Back To Top