digiflowz
The Gist
- Historical context. ELIZA, released in 1966, pioneered chatbot technology, using pattern-matching to simulate conversation and paving the way for future natural language systems.
- Evolutionary path. Chatbots evolved from simple rule-based systems to AI-powered, voice-activated personal assistants and generative AI chatbots capable of full-fledged conversations.
- Challenges ahead. Despite advancements, AI chatbots face issues like the proliferation of disinformation, regulatory hurdles and public sentiment favoring human customer service over AI.
The first chatbot is generally considered to be ELIZA, created by Joseph Weizenbaum at MIT and released in 1966. ELIZA simulated conversation by using pattern matching and substitution methodology, which gave the illusion of understanding. It would rephrase user inputs as questions or statements, tricking some users into believing they were chatting with a real human. The era of chatbots had begun, and over the years, chatbots have evolved to the point where they are able to have full-fledged, existentialist conversations. Let’s examine the evolution of chatbots, from early primitive bots to today’s advanced conversational and generative AI, and explore how they are currently being used across a variety of domains.
Rule-Based Chatbots
ELIZA was one of the earliest primitive chatbots. Named for the lead character in George Bernard Shaw’s Pygmalion, ELIZA simulated conversation using basic natural language processing (NLP). Most of ELIZA’s language capabilities came from individual “scripts.” The most famous script, DOCTOR, engaged users with open-ended questions and responses reminiscent of an empathic psychologist like Carl Rogers. With just simple pattern-matching rules, and no real understanding of emotion, ELIZA could sometimes pass as human. Despite its limitations, this breakthrough program paved the way for the natural language systems we use today.
Mike King, chief marketing officer at AIPRM, an AI prompt marketplace, told CMSWire that we should take a moment to appreciate the sheer brilliance of ELIZA. “In a world of punch cards and bulky mainframes, this chatbot was nothing short of revolutionary. While ELIZA may have been a simple pattern-matching mechanism, it set the stage for what was to come. Imagine, back in 1966, having a digital entity rephrasing your words and throwing them back as questions. It was the beginning of a new dawn.”
Other rule-based chatbots soon followed. In 1972, psychiatrist Kenneth Colby created the next influential chatbot, Parry, at Stanford University. Parry was groundbreaking for attempting to simulate a person with paranoid schizophrenia.
In 1988, Rollo Carpenter, a British programmer, began working on Jabberwacky, which was designed to replicate normal human conversation in an enjoyable, amusing and natural way. It became available on the web in 1997 and went on to win the Loebner Prize — an annual AI competition designed to find computer programs considered to be the most human-like. In 2008, Carpenter launched Cleverbot, a chatbot that used machine learning (ML) to have conversations with humans.
In 1995, Richard Wallace developed A.L.I.C.E., which won the Loebner Prize three times for the most humanlike chatbot. However, it failed the Turing test, which evaluates a machine’s ability to exhibit intelligent conversation indistinguishable from a human.
In 2001, AOL debuted SmarterChild, a chatbot that users could interact with on AOL Instant Messenger. SmarterChild could look up information, play games and have basic conversations with users.
Personal Assistant Bots
Apple’s Siri, released in 2011, pioneered the concept of conversational virtual assistants on phones. Siri demonstrated the potential for AI-powered, voice-controlled assistants that could understand natural language requests. Amazon Alexa made its debut in 2015, two years after Amazon acquired a Polish speech synthesizer named Ivona. It is now part of the Amazon Echo Dot, Echo Studio, Echo smart speaker and Amazon Tap speakers.
Google launched its Google Assistant for Android phones in 2016. Not to be left behind, Microsoft introduced its own assistant called Cortana. The launch of these tech giants’ virtual assistants marked a turning point in making conversational AI a mainstream and expected feature on smartphones.
Conversational AI Chatbots
Conversational chatbots started out being rule-based, meaning that they relied on scripted rules and templates to construct responses. Today, conversational chatbots use AI and neural networks to be able to create better responses. In 2017, Eugenia Kuyda created Replika, an app that creates AI companions capable of personalized conversations that learn about users over time. Replika was described in an article as “the app that’s trying to replicate you.”
The “Replika AI Agent” that powers the chatbot was trained on billions of lines of dialogue to learn nuanced patterns in human conversation. Replika uses these neural networks and ML algorithms to analyze chat contexts and continually improve its responses based on feedback from users rating its replies.
Unlike rigidly scripted chatbots that rely on keywords, Replika aims to engage users in an emotional, human-like way by understanding the full meaning of conversational turns. While some common small-talk responses may be scripted, Replika generates unique, personalized replies using AI rather than just predefined templates.
Generative AI Chatbots
Since the release of Open AI’s ChatGPT in November 2022, generative AI has been in the news practically every day. Not long after the release of ChatGPT, both Google and Microsoft announced their own generative AI chatbots, namely Google Bard and Microsoft Bing.
Generative AI has come to be a huge business commodity with extremely rapid growth. GlobalData’s Generative AI Growth Analysis forecasts the global generative AI market size to grow from $10.16 billion in 2022 to $103.74 billion by 2030. Additionally, the use of generative AI chatbots for customer service has exponentially grown due to high consumer demand for speed and convenience when interacting with customer service. An August 2023 LivePerson survey revealed that not only do consumers hate traditional Interactive voice response (IVR) systems, 57% of those polled would rather do a load of laundry than interact with an IVR — and 41% would rather clean a toilet.
Microsoft collaborated with Open AI to upgrade their Bing search engine and introduce a generative AI chatbot that was built on the same technology as ChatGPT. Users can chat with Bing about practically any topic, or they can use the chatbot to search the web.
Google Bard was introduced as a standalone chatbot that was not connected to the Google search engine, and looks very much like ChatGPT.
Another popular generative AI chatbot is Anthropic’s Claude 2. Claude is similar to ChatGPT in that it is a large language model, but it was created to be a personal assistant. Anthropic designed Claude to be helpful, harmless, and honest through a technique called Constitutional AI, which aims to infuse systems with “values” that are defined by a “constitution.” Claude’s training data and model architecture have been carefully constrained to avoid exhibiting harmful biases or misinformation.
Today, there are literally hundreds of generative AI applications available for various domains, such as customer service, healthcare, hospitality, real estate, marketing, travel, restaurants, onboarding, and more, as can be seen at the chatbot directory and reviews site, Chatbots.org.
Generative AI has opened the door for businesses large and small to leverage the power of AI for customer service and marketing. “On the customer service front, our chatbot is the first line of support. Being a generative AI startup, it’d be silly of us not to use the technology, so we leverage SiteGPT as a trainable bot that we’ve fed our knowledge base,” said King. “It handles common inquiries, provides instant solutions, and ensures that our human team can focus on more complex issues. It’s all about striking the right balance between automation and human touch.”
AI Chatbots Continue to Evolve
AI chatbots continue to evolve, and developers are finding new and unique applications where they can be useful. One new application of AI chat is EmbodyMe’s Xpression, an AI-driven camera app that enables users to bring any picture to life. This real-time generative AI app allows users to use a picture of their favorite celebrity, animated character, or historical figure to mimic their own facial movements. The creators of Xpression envision people using the tool as a sit-in for themselves on company video conferences.
Another unique application of AI chat can be seen at Text With Jesus, a chatbot that allows users to have conversations with biblical figures including Jesus, Mary, Joseph, Peter, Matthew — and even Satan. The app responds to queries with responses that are generated from text from the Bible. The website states that the chatbot is a tool for “exploration, education, and engagement with biblical narratives, and it is not intended to replace or mimic direct communication with divine entities.”
Meta recently announced its SeamlessM4T chatbot, a multimodal and multilingual AI translation application that resembles the Babel Fish from Douglas Adams’ book, The Hitchhiker’s Guide to the Galaxy. It features speech recognition for nearly 100 languages, speech-to-text translation, speech-to-speech translation, text-to-text translation, and text-to-speech translation. The demo allows users to try it out for themselves by making a short recording of their own voice and then translating it to the language of their choice.
Babak Pahlavan, CEO and founder of NinjaTech AI, a conversational AI company focused on making people more productive at work, told CMSWire that AI is constructing a new way to interact, transforming chatbots to be more sophisticated models capable of understanding context and generating relevant responses. “Imagine an AI-assisted curation process by an agent without filters or search menus; the customer can simply converse with the assistant in their native language.”
Pahlavan explained that as chatbots continue to evolve, they are leveraging NLP, which includes a capability called “sentiment analysis.” “Not only can we move away from menu-driven, IVR-replacement style (aka ‘dumb’) chatbots and adopt more natural ways of engagement, we can determine, for example, if the customer is getting frustrated, upset, or emotional and change the tone of the response or even transfer to a live, human agent while remaining in the same channel (eg the chat window).”
The Challenges of AI Chatbots
As with most new technologies, the use of AI in chatbots is challenging for many different reasons. Many generative AI applications have a tendency to “hallucinate,” which essentially refers to situations where they do not have enough knowledge of a specific topic, so they essentially “make stuff up.” Additionally, generative AI has to have guardrails in place to prevent it from discussing harmful or offensive topics. The last thing a brand needs is for a chatbot to “go off the rails” and use offensive language that is directed toward a customer. Another concern is that generative AI may provide customers with information that is incorrect, biased, or simply not based on facts.
Lindsey Zuloaga, chief data scientist at HireVue, a leader in tech-based hiring tools, told CMSWire that most of us have interacted with chatbots in some way, whether it’s returning clothes to an online retailer, making a dinner reservation, or asking about the status of a job application.
“Interactions of this kind are the typical, benign chatbot use cases. But now ChatGPT and other generative AI tools are raising well-deserved concerns. The primary concern I’m seeing is about the proliferation of disinformation,” said Zuloaga. “Frankly, innovation has outpaced safeguards, and it’s important that researchers and technologists are asking critical questions and rapidly trying to build in safeguards.” Zuloaga emphasized that it’s important that vendors don’t jump to integrate the ChatGPT model into existing tools until they’ve conducted rigorous testing.
Some users have complained that they were not aware that they were conversing with AI, and are demanding that brands are transparent and disclose the use of AI in their products and services. Brands must also be aware of new rules and regulations that apply to the use of AI. “New AI regulations are being proposed and passed constantly, from the new EU AI Act to NYC Local Law 144, and generative AI should be held to the same, if not greater, standards as other AI tech. First and foremost, all generative AI tools should undergo rigorous third-party algorithmic audits and release the results,” said Zuloaga, who suggested that creators of new AI tools should prioritize creating AI Explainability Statements, which is a valuable third party process that documents to the public, customers and users how a given technology is developed and tested.
The biggest challenge of AI chatbots is uniquely human. That is, humans often prefer to speak with other humans, especially when they are trying to rectify a problem. A recent SurveyMonkey report indicated that 90% of those polled prefer humans to AI for customer service, stating that humans better understand their needs, provide more thorough explanations and more options, and are less likely to frustrate them. Additionally, 56% have negative feelings about brands using AI as part of the customer experience. As people get more used to interacting with AI, and AI continues to improve, this is likely to change, but currently, there is still a lack of faith in AI’s abilities.
Final Thoughts on Chatbots
Chatbots have come a long way, and today’s AI-driven bots are able to have sophisticated conversations. From personal assistants to conversational chatbots, to generative AI, chatbots are becoming ubiquitous on websites and apps. Improving public trust through ethical development practices and transparency is crucial as these bots continue to play a larger role in the lives of consumers.
