Chatbot

A chatbot is a form of artificial intelligence (AI) software that is capable of simulating a conversation with a human user through text or speech. At the most basic level, chatbots are computer programs designed to automatically engage with received messages. They are frequently embedded on or in websites, messaging applications, mobile apps, or other similar platforms as a "help" feature, capable of answering basic user queries about the platform's functionality. While some chatbots are simply programmed to provide standard responses to specific inquiries, more advanced chatbots can be programmed to offer unique responses that vary depending on the content of the user’s input. Since their earliest technological ancestors began to appear in the 1960s and 1970s, chatbots have risen to become a popular, if somewhat controversial, alternative to one-on-one human interaction in a wide variety of communications, marketing, and other useful applications. In the 2010s and 2020s, a boom in generative AI technology led to the development of increasingly sophisticated chatbots with the ability to converse on a wide array of topics, bringing even greater utility but also further concerns about potential negative social impacts.

Background

The early groundwork for modern chatbots was laid by renowned British scientist Alan Turing in the 1950s. Turing suggested that any truly intelligent machine would be entirely indistinguishable from a human being when engaged in a text-based conversation. Based on that theory, he developed a special test designed to measure a machine’s ability to exhibit intelligent behavior equal to or at least indistinguishable from that of a real human being. That test, which eventually became known as the Turing test, effectively set the stage for the development of AI and chatbots.

The first true chatbot was created by German American computer scientist Joseph Weizenbaum in 1966. Weizenbaum’s chatbot, called ELIZA, relied on a simple system of pattern matching and substitution to simulate human conversation. To accomplish this feat, the ELIZA program took whatever words the user entered into a computer and matched them with a list of potential scripted responses. ELIZA’s specialized script, which was meant to simulate a psychotherapist, went on to have a significant impact on the development of language processing in AI.

American psychiatrist Kenneth Colby created another chatbot called PARRY in 1972. Where ELIZA was a simulation of a psychotherapist, PARRY was a simulation of a patient with schizophrenia. It did this through the use of a natural language program that could emulate the patterns of thought common among people diagnosed with schizophrenia. When PARRY was put to the Turing test, human interrogators had little success in distinguishing the chatbot from actual human beings experiencing schizophrenia.

Other early chatbots included Jabberwacky, Dr. Sbaitso, Artificial Linguistic Internet Computer Entity (ALICE), and SmarterChild. Created by developer Rollo Carpenter in 1988, Jabberwacky was designed to simulate human conversation in an entertaining manner. Dr. Sbaitso was a simplistic psychologist chatbot made by Creative Labs for Microsoft’s MS-DOS in 1992. ALICE was a universal language processing chatbot created by computer language author Richard Wallace in 1995. SmarterChild was an intelligent chatbot that was first deployed on both AOL and MSN short message service (SMS) platforms in 2001. SmarterChild was capable of carrying on personalized, enjoyable conversations with users. It brought wider attention to chatbot technology and paved the way for many later similar programs.

Overview

Chatbots have advanced significantly since the days of ELIZA and PARRY. While still designed to convince users that they are communicating with a real person, modern chatbots are much more complex and nuanced than their early predecessors. Most incorporate AI technologies such as deep learning, natural language processing, and machine learning algorithms. All of this allows a chatbot to become better at predicting appropriate responses the more it interacts with users.

There are two main types of chatbots: stateless and stateful. Stateless chatbots essentially treat each individual interaction as if it is with a new human person. More complex stateful chatbots are capable of reviewing previous interactions and framing new responses in the context of past conversations.

In practice, a chatbot receives some sort of request from a human user and provides an appropriate response. The nature of a chatbot’s response to a user request can vary widely. The most basic chatbots typically respond with generic, predetermined text. Responses returned by more complex chats may include, but are not limited to, text retrieved from a knowledge base of different answers; contextualized information derived from user-provided information; data stored in a type of large-scale application software package called an enterprise system; information that the chatbot gathers by interacting with another application; or a disambiguating question meant to help the chatbot understand the user’s request more accurately.

Most chatbots are used for commercial purposes. Because chatbot applications help to streamline interactions between people and services, many companies have adopted chatbots as a way of engaging with their customers. From the customer perspective, using a chatbot is often a good way to get a quick answer, register complaints and resolve problems, get a more detailed answer to a query when necessary, or to get connected with a human customer service agent. From a company perspective, chatbots provide an automated means of dealing with customer inquiries that reduces the need for human interaction. This can translate to significant savings in both time and money.

There are noncommercial applications for chatbots as well. For example, the Endurance chatbot was developed as a companion for dementia and Alzheimer’s patients that can comfort users and help identify potential memory problems at the same time. Casper’s Insomnobot 3000 chatbot was designed to give insomniacs someone to talk to when they are struggling to fall asleep. The child advocacy nonprofit organization UNICEF created a chatbot called U-Report to gather data on important social issues through polling.

Despite the demonstrated usefulness and widespread adoption of chatbots, however, there are some significant potential downsides to the technology. Critics argue that many people find chatbots to be a turnoff and are more likely to prefer direct human interaction. This can be especially true when it comes to older or otherwise technologically disinclined individuals. It is also often true when customers have a more complex issue that a chatbot may not be able to effectively address. Another problem is that chatbots do not always understand users’ responses and may be prone to giving inappropriate or incorrect answers as a result. This can lead to customer distrust of chatbots and a subsequent downturn in business.

Even with such drawbacks, chatbots steadily became more and more common on many kinds of websites and applications through the 2010s and into the 2020s. Improvements in AI processing and other technologies helped make many chatbots more effective, and in some cases branched out into new categories altogether. For example, sophisticated voice-recognition software enabled digital personal assistant programs that connected chatbot functionality to many powerful features on smartphones and other internet-connected devices. Popular examples of virtual assistants like Amazon's Alexa and Apple's Siri often came to be considered separately from traditional customer-service-type chatbots.

A major development in chatbot systems came in the early 2020s with groundbreaking advances in AI technologies such as large language models (LLMs) and artificial neural networks. In 2022, the company OpenAI released ChatGPT, a chatbot capable of generating complex natural-language responses to user prompts on almost any subject. ChatGPT was immediately popular and generated an intense wave of interest in AI in general and chatbots in particular. Many other major technology companies quickly released their own LLM-based chatbots; notable examples included Google's Gemini and Microsoft's Copilot. The success of these programs can be attributed largely to their practical applications, as users soon found that they could save time on a wide array of business and research tasks, writing assignments, and more.

However, ChatGPT and its imitators also quickly proved controversial for several reasons. Many educators worried about the impact on student learning, arguing that chatbots capable of quickly solving complex math problems or generating entire research papers could undermine the integrity of traditional homework assignments and other academic processes. Even more notably, researchers soon recognized that despite their increasing sophistication, chatbots could sometimes give inaccurate or incorrect responses, a phenomenon known as "hallucination." Many critics suggested that these errors could create a feedback loop as both human users and other LLM-driven chatbots come to rely on chatbot-generated information, spreading and reinforcing misinformation across the internet. Experts also warned that bad actors could purposefully generate and spread misinformation more easily with the aid of chatbots, increasing the threat of cybercrime and even potentially impacting entire industries or society at large in complex ways, such as subtly influencing political views on social media.

Bibliography

Bowman, Emma. “A New AI Chatbot Might Do Your Homework for You. But It's Still Not An A+ Student.” NPR, 19 Dec. 2022, www.npr.org/2022/12/19/1143912956/chatgpt-ai-chatbot-homework-academia. Accessed 21 Mar. 2023.

"Chatbot Definition, Types, Pros & Cons, Examples." Investopedia, 18 May 2022, www.investopedia.com/terms/c/chatbot.asp. Accessed 28 Oct. 2024.

Elliot, Christopher. “Chatbots Are Killing Customer Service. Here’s Why.” Forbes, 27 Aug. 2018, www.forbes.com/sites/christopherelliott/2018/08/27/chatbots-are-killing-customer-service-heres-why/#57ccff2e13c5. Accessed 22 Oct. 2018.

Gillis, Alexander S., et al. “Chatbot.” TechTarget, www.techtarget.com/searchcustomerexperience/definition/chatbot. Accessed 28 Oct. 2024.

Gobiet, Marie. “The History of Chatbots—From ELIZA to ChatGPT.” ONLIM, 15 Feb. 2024, onlim.com/en/the-history-of-chatbots. Accessed 28 Oct. 2024.

Janarthanam, Srini. Hands-On Chatbots and Conversational UI Development: Build Chatbots and Voice User Interfaces with Chatfuel, Dialogflow, Microsoft Bot Framework, Twilio, and Alexa Skills. Packt Publishing, 2017.

Milmo, Dan. “ChatGPT Reaches 100 Million Users Two Months after Launch.” The Guardian, 2 Feb. 2023, www.theguardian.com/technology/2023/feb/02/chatgpt-100-million-users-open-ai-fastest-growing-app. Accessed 21 Mar. 2023.

“A Visual History of Chatbots.” Chatbots Magazine, chatbotsmagazine.com/a-visual-history-of-chatbots-8bf3b31dbfb2. Accessed 22 Oct. 2018.