What is Conversational AI?
Imagining Alan Turing talking to Alexa. Image generated by Nieves Ábalos with AI (Midjourney v6).
Audio generated using Nieves’ cloned voice with Fonos with Monoceros Labs technology.
The umbrella of Conversational AI
Conversational AI is the set of technologies based on Artificial Intelligence that allow us to have a chat or talk with any computer. It is the technology of conversations.
We often use the term Conversational AI to refer to virtual assistants or chatbots. The goal is to build systems which people can interact with, through voice and/or writing with technology, naturally through a conversation as if we were talking to a person. From answering simple questions to performing more complicated tasks, Conversational AI makes interaction with technology more natural.
Under the umbrella of Conversational AI, we find technologies like natural language processing (NLP), which allows machines to extract meaning from language. Machine learning (ML) algorithms and deep learning (DL) architectures are also used to improve the ability of AI systems to understand the intent of the person they are talking to, to decide what type of behavior is most appropriate at each point in the conversation, and to generate responses based on certain data. Additionally, both speech recognition and speech synthesis are used to interact with people through speech.
Is the term new?
The truth is that although it is a set of diverse technologies that have evolved a lot in the last decade, and that now are booming due to Generative AI which focuses on language models and chatbots like ChatGPT, the term “Conversational AI” is not a new concept.
Conversational AI is strongly linked to the term “Artificial Intelligence”, a term coined by John McCarthy along with other researchers like Minsky and Shannon in 1955 [1], but the idea of machines that behave like people conversing and thinking is older. Alan Turing, in 1950, published an article titled "Computing Machinery and Intelligence"[2], where he explored the possibility of creating thinking machines.
“I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted.”
For Alan Turing, natural language conversations were the interface to validate certain intelligent capabilities of machines, such as the ability to think. His most significant contribution was the proposal of the "Imitation Game"[2] or "Turing Test", which he proposed as a method to determine if a machine could "think". The test consists of a machine maintaining a conversation indistinguishable from a conversation with a human being. If the machine managed to make its interlocutor think it was a person, Turing argued that it was reasonable to say that the machine was "thinking". This simplified approach allowed him to argue convincingly that a "thinking machine" was at least plausible.
These contributions laid the groundwork for the concept of Conversational AI and the debate over the possibility of machines that can think and communicate like people.
ELIZA, the first chatbot prototype
Joseph Weizenbaum played an important role in the initial development of artificial intelligence, especially in the field of Conversational AI. His most relevant contribution was the creation of ELIZA [3], one of the first “chatbots” or programs that worked with natural language.
Weizenbaum developed ELIZA while working at MIT, between 1964 and 1966. ELIZA, which simulated a psychotherapist with empathetic listening by reflecting the user's statements and asking questions based on them, was one of the first programs capable of using techniques to detect keywords in the conversation and use them to maintain a conversation with people.
Example of conversation with ELIZA and flow of keyword detection. Source: Weizenbaum (1966) [3]
ELIZA demonstrated how a relatively simple program, which used the aforementioned concepts to ask questions, could create the illusion of understanding and empathy. In fact, many users attributed human qualities to the program, despite its functioning, and led Weizenbaum to reflect on the ethical implications of AI and the human-machine relationship in his book "Computer Power and Human Reason" [4], where he expressed concerns about its use in areas that require human understanding and empathy.
His work with ELIZA was pioneering in Conversational AI and remains relevant in current discussions about chatbots and virtual assistants. His ethical reflections have influenced the responsible development of AI and the consideration of its social and psychological implications.
Conversational AI now
Conversational AI has come a long way in the last decade, transforming the way we interact with technology and redefining the limits of what is possible in human-machine communication.
In recent years, the development of Conversational AI has experienced three waves that have expanded technology and use of chatbots, voice assistants, and currently language models, with several companies standing out in this field. These waves have also heightened and diminished our expectations of them.
This technology is rapidly changing our digital world. From leading tech companies like Apple with Siri, which democratized voice virtual assistants on our mobile phones in 2009, to Amazon launching Alexa in 2015, introducing these voice assistants into our homes and helping us control devices, to the advances of recent years with OpenAI launching ChatGPT in November 2022, generating an unprecedented impact on the perception and mass use of Conversational AI, reaching 1 million users in five days after its launch, and not forgetting the impact of Antrophic, Google, Microsoft, and Meta in the race for applications and large language models.
However, as we move forward, we must maintain a balance between innovation and ethical responsibility. The future of Conversational AI promises great potential to improve education, mental health, and accessibility, but it also poses significant challenges in terms of privacy, security, and the impact on people. As this technology continues to evolve, it will be crucial to work on harnessing its potential while mitigating its risks.
The road ahead promises to be the next great transformation of the society we live in.
Bibliography:
[1] McCarthy, J., Minsky, M. L., Rochester, N., Corporation, I. B. M., & Shannon, C. E. (1955). A PROPOSAL FOR THE DARTMOUTH SUMMER RESEARCH PROJECT ON ARTIFICIAL INTELLIGENCE. http://jmc.stanford.edu/articles/dartmouth.html
[2] Turing, A. M. (1950). Computing machinery and intelligence. Mind, LIX(236), 433-460. https://doi.org/10.1093/mind/LIX.236.433
[3] Weizenbaum, J. (1966). ELIZA—a computer program for the study of natural language communication between man and machine. Commun. ACM, 9(1), 36-45. https://doi.org/10.1145/365153.365168
[4] Weizenbaum, J. (1976). Computer power and human reason: From judgment to calculation. Freeman. https://archive.org/details/computerpowerhum0000weiz_v0i3
Chatbots have evolved from simple command-based systems to sophisticated conversational AI tools. This article explores their functioning, from the pioneering ELIZA to modern LLM-based systems. We analyze the traditional process: natural language understanding (NLU), dialogue management (DM), and response generation (RG), comparing it with the end-to-end LLM-based approach.