When AI Strengthens Good Old Chatbots: A Brief History of Conversational AI

The term “Conversational AI” invokes the use of artificial intelligence technologies to enable computer solutions to communicate with humans in a natural and interactive way. It can be applied in many different contexts, such as customer service Chatbots, virtual assistants, and communication systems. Its capability to understand and respond to human speech has the potential to transform the way we interact with machines, software, and applications.

Whether through text, voice, or other forms of communication, including video or image, Conversational AI is changing the way we communicate with either the digital or the real world around us.

Conversational AI is becoming increasingly prevalent in our daily lives. Whether on a personal or professional level, from customer service Chatbots on e-commerce websites to virtual assistants on our smartphones or in connected cars. Its potential applications are vast, ranging from automating routine tasks to improving the efficiency of customer service, to name a few. One unexpected but very relevant example of Conversational AI use case is its usage applied in the field of healthcare. Some health centers are using Chatbots to triage patients and provide them with personalized health advice, freeing up doctors to focus on more complex cases. 

As the technology continues to advance, it’s likely that we will see even greater adoption of Conversational AI in a variety of industries and contexts.

Brief History of Conversational AI

Early Developments and Milestones

The development of Chatbots, or computer programs designed to simulate conversation with human users, dates back to the 1960s. Indeed, the first referenced Chatbot in technology history is ELIZA, which was developed in 1966 by Joseph Weizenbaum at Massachusetts Institute of Technology (MIT). ELIZA was designed to mimic the language patterns of a psychotherapist and could carry out simple conversations with human users. It was one of the first examples of a Chatbot and laid the foundation for the development of more advanced Conversational AI systems.

These early Chatbots were relatively simple and could only handle a limited number of pre-programmed responses. Over time, Chatbots have become more advanced and are now able to respond to a wider range of inputs.

With the advent of machine learning (ML), natural language processing (NLP), and natural language understanding (NLU) techniques, Chatbots have evolved into more advanced Conversational AI systems. These systems are able to understand and respond to human speech in a more natural and intuitive way and can even engage in more interactive conversations. Most importantly, they now become capable of learning and adapting over time. Thanks to this, they become more and more efficient. They tend to be more and more accepted and, from there, more and more requested.

To illustrate this, a Chatbot launched by OpenAI (the now very famous ChatGPT3) in November of 2022 has crossed 1 million users in just five days. It took Netflix 41 months, Facebook 10 months, and Instagram 2.5 months…

Today, Conversational AI systems are used in a large number of different contexts, far from the enhanced FAQ function to which they were limited in their early days. They are contributing to revolutionizing the way we interact and exchange information with the digital world and perform value-added actions. These systems make it easier and more convenient for people to access information and get things done. 

The purpose of innovation is to improve upon existing products, processes, or services or to create new ones that meet the needs of customers or society in a better way. It includes increasing efficiency, reducing costs, improving quality, or developing new capabilities. Conversational AI is no exception to this trend.

To trace the path taken by this technology, here are some key milestones in the history of Conversational AI development. They demonstrate the impressive progress that has been made by these systems and their increasing capabilities: 

• 1960: ELIZA, the first referenced Chatbot, is developed by Joseph Weizenbaum at MIT.
• 1972: PARRY, a natural language program that simulates the thinking of a paranoid individual. It, therefore, always misinterprets the motivations of others. Parry was the first to pass the Turing Test.
• 1997: A.L.IC.E. (Artificial Linguistic Internet Computer Entity), a natural language processing Chatbot, is developed by Richard S. Wallace. It won the Loebner Prize as “the most human computer” at the annual Turing Test contests in 2000.
• 2005: Apple's virtual assistant, Siri, is released, co-founded by a french scientist: Luc Julia, Ph.D.
• 2010: IBM's Watson (powered by 90 servers and 21.6TB of data) competes on the game show Jeopardy! and defeats human champions.
• 2014: Cortana was first demonstrated at Microsoft's Build developer conference. Directly integrated into both Windows Phone devices and Windows 10 PCs.
• 2014: Facebook launches M to face Siri and Cortana. M is a virtual assistant that uses machine learning to assist with tasks.
• 2016: Google releases Google Assistant, "Hey Google”!, a virtual assistant for Android devices.
• 2017: Amazon's Echo devices with Alexa virtual assistant become widely popular.
• 2018: OpenAI releases GPT-2, a large-scale language model with the ability to generate human-like text.
• 2022: OpenAI releases GPT-3. It can be used for language tasks, such as translation, summarization, question answering, and text generation. It can even perform tasks such as coding and translation without explicit training on those tasks.
• “A long time ago in a galaxy far, far away…”: C-3PO, a humanoid droid programmed primarily for etiquette and protocol, designed to interact with organics along “six million forms of communication.” Also known all over the explored universe helping Luke Skywalker and their rebellion defeat the Empire and restore freedom to the galaxy. But that's an entirely different story.

The Current State of Usage by Industry

The improved capabilities of large-scale language models such as Google's BERT, OpenAI's GPT-3, and Microsoft's Transformer have the power to fundamentally transform the field of Conversational AI and enable the development of more advanced virtual assistants, Chatbots, and other communication devices and systems. 

Very interesting use cases exist in many different fields. Among those, let us note, for example:

• Virtual assistants: such as Apple's Siri, Google Assistant, and Amazon's Alexa, are becoming increasingly common and are able to perform a wide range of tasks through voice or text-based interactions. These systems use NLP, NLU, and machine learning techniques to understand and respond to user requests. They get to learn and change, improving their ability to communicate.
• Customer service Chatbots: Chatbots are being widely used in the customer service industry to handle routine inquiries and provide information to customers. These systems are able to handle a high volume of interactions 24/7 and can often resolve simple issues faster than a human customer service representative based on a rule-based decision-making process. 
• Language translation: There are a number of language translation systems (Google Translate or DeepL, to name a few) that use Conversational AI to enable the real-time translation of spoken or written communication. These systems can be used in a variety of contexts, such as international business meetings or in everyday life, facilitating communication between people who speak different languages.
• Education: There are also a number of educational platforms that use Conversational AI to provide students with personalized learning experiences. These systems are able to adapt to the needs and abilities of individual students and provide tailored instruction and feedback. Carnegie Learning, for example, uses AI and Machine Learning to help students develop a deeper conceptual understanding of math and world languages. Algorithms study students' habits to personalize their learning experience.
• Healthcare: In the healthcare industry, Conversational AI is being used to triage patients and establish an initial diagnosis. Chatbots can provide symptom checking and recommend a course of action, such as seeking further medical attention or self-care at home. These systems can also be used to schedule appointments and refill prescriptions. In England, the National Health Service announced in 2017 an agreement with Babylon Health (a health app company) to use a Chatbot driven by algorithms based on clinical data that triage patients in two minutes, without human intervention, based on reported symptoms.
• Banking: In the banking industry, Chatbots are being used to handle routine inquiries and assist with tasks such as account management and bill payment. Some banks are also using Conversational AI to provide personalized financial advice and recommendations to customers.
• Insurance: In the insurance industry, Conversational AI is being used to assist with simple tasks, including claims processing and policy management. As banks can do, insurance companies are also using Chatbots to provide personalized recommendations and advice to customers.
• Retail: In the retail industry, Chatbots can be very useful when it comes to providing customers with product recommendations, order tracking, and many simple customer service requests. They can be also helpful to provide personalized shopping experiences and tailor marketing efforts to individual customers.
• Human Resources: An AI-Driven Chabot can be very helpful in performing recruitment tasks such as answering FAQs, screening candidates against offers and resumes, scheduling interviews, providing updates, and collecting and organizing resumes. This reduces the workload of recruiters and allows them to spend more time on tasks that require people skills such as empathy. Results from the first 10,000 conversations have shown that Mya engages effectively with 92% of their candidates.

Types of Conversational AI

The conversational AI landscape is made up of different main components that are distinguished by specific characteristics.

Chatbots are simple computer programs designed to simulate conversation with human users, more likely in a question-and-response mode. In some ways, they can be likened to somewhat elaborate FAQ systems. They can be integrated into messaging platforms, mobile apps, and website portals to provide customers with basic services or information. For example, a Chatbot on a retail website can help customers navigate the site, make product recommendations, and answer questions about shipping and returns.

Virtual assistants are Conversational AI systems designed to assist with a wide range of more elaborate tasks, such as checking balance information, determining whether an item is in stock, assessing the status of an order, tracking shipments, scheduling appointments, setting reminders, processing calculations, interacting with applications or providing various information.

They can be accessed through a variety of devices, including smartphones, smart speakers, and personal computers. For example, Apple's Siri and Amazon's Alexa are both examples of virtual assistants. Their communications abilities are integrated with automation functions.

Voice assistants are Conversational AI systems that are activated by voice commands. They can be used to control other devices, play music, and provide information. They can be also called “Voice-controlled assistants,” “Speech-enabled assistants,” “Spoken Language Interfaces,” “Speech recognition software,” or simply “Smart speakers." They are becoming increasingly popular with the advent of smart speaker devices like Amazon Echo, Google Home, or Apple HomePod. The voice-controlled AI assistant technology can be integrated with other smart devices, such as home appliances and cars, allowing for voice control of those devices.

There are other forms of Conversational AI included in the range of language-generating AI systems. They are able to produce human-like text or speech and can be used for a wide range of applications, such as language translation, automated writing, and text summarization. The list is continuously growing.

Benefits of Conversational AI

In 2016, Gartner predicted: “2020, the average person will have more conversations with bots than with their spouse”. Knowing that this tragic prediction did not come true, at least in 2023, let’s see now how Conversational AI can improve customer service, increase efficiency and productivity and enhance the user experience for businesses and generate cost savings.

Conversational AI Provides a Superior Service to Chatbots in Many Ways

Solutions, such as virtual assistants, have a knowledge base, whereas Chatbots generally rely on FAQ files. Chatbots, therefore, draw from a limited library of scripts and combinations of questions and answers. This approach is therefore limited to producing predetermined answers.

The so-called "intelligent" or “smart” virtual assistants, on the other hand, are pre-trained and have a wide range of knowledge, which provides a much richer conversational base.

They can tend to solve problems rather than just answer simple questions. Intelligent digital assistants in the insurance field are becoming digital advisors by accompanying a customer and recommending the next most relevant step in their journey to file a claim or follow up on an incident, for example. 

Finally, by using conversational skills, data, and pattern analysis through machine learning algorithms, intelligent digital assistants are able to qualify customer demands and can seamlessly direct them to the services that are right for them.

One of the Main Benefits of Conversational AI Is Improved Customer Service

Chatbots, virtual assistants, and voice assistants can be used to answer customer questions, provide information, and help with troubleshooting 24/7. It can also qualify a request and redirect it to a human operator when it’s necessary.

This improves overall customer satisfaction and perhaps even loyalty in some cases, as customers are able to get the help they need quickly and efficiently. For example, according to JuniperResearch, Chatbots within retail, eCommerce, banking, and healthcare were predicted to be responsible for cost savings of over $8 billion per annum by 2022.

Conversational AI can also lead to increased efficiency and productivity in the workplace. For example, virtual assistants can help schedule meetings, set reminders, and perform other tasks that would otherwise take up valuable time and energy. This can lead to increased productivity and a reduction in workload for employees. 

Conversational AI is also a means to produce a more personalized and engaging user experience. For example, Chatbots and virtual assistants can be trained to understand the context and preferences of a user and make personalized recommendations or provide customized information. This can lead to increased user engagement and satisfaction. By the way, employee satisfaction, as a counterpart of customer satisfaction, is also a target that can be reached through Conversational AI and related smart automation processes. 

Knowing the fact that worldwide companies spend over $1 trillion on customer service calls each year; it’s easy to understand that Conversational AI does represent a real opportunity to lower costs. For example, Chatbots and virtual assistants can be used to automate repetitive and time-consuming tasks, reducing the need for human labor. This can result in lower costs and increased efficiency of customer service. In addition, Conversational AI can help businesses to identify and target high-value customers, leading to increased revenue.

Challenges and Limitations of Conversational AI

NLP and NLU Limitations

Natural Language Processing and Understanding are subfields of artificial intelligence that deal with the interaction between computers and humans in natural language. These systems used to analyze and understand text and speech data generated by humans can be challenged by the complexity and variability of human language. That definitely makes it difficult to understand the meaning and intent of language data in many situations.

Did you know, for example, that over 7,000 languages are spoken across the world today? Chinese, the most important as regards the number of speakers, has just by itself 13 variations. Arabic has 20 of them. To be put into perspective, it is estimated that 80% of online content is available in only one of 10 of the following languages: English, Chinese, Spanish, Japanese, Arabic, Portuguese, German, French, Russian, and Korean.

This explains clearly the fact that an important limitation of natural language systems is dealing with idiomatic expressions, as natural language is full of idioms, slang, and other non-literal forms of expression. These idioms and colloquialisms often have multiple meanings and depend on the context to understand them. Not to mention that the finesse of the language is also due to the intonation and the intention of the speakers and even their gestures or facial expression. All this, technically, can be acquired through video and analyzed by appropriate AI algorithms.

Besides the disparity of the spoken languages, one of the main limitations of NLP is dealing with ambiguity. Indeed, natural language often contains multiple possible interpretations, and it can be difficult for an AI system to determine the correct one. For example, if a customer writes, "I need help with my order," a Chatbot might ask, "What is wrong with your order?" but it might not be able to understand if the customer is asking for help tracking their order or if they have an issue with the products they received or if they want to cancel it.

Of course, NLP systems have difficulties dealing with sarcasm, irony, and the use of negative forms. These forms of figurative language are often used to convey meaning indirectly, which is not immediately apparent from the words used.

Concordant studies find that even the most advanced NLP systems still struggle with understanding idioms and figurative language or different feelings suggested with the same words. It might still be a long way before computers can truly understand and interpret human language with the same level of fluency as a human being.

Personalization and Customization Difficulties

Personalization and customization are key aspects of Conversational AI, as they allow Chatbots, virtual assistants, and other Conversational AI systems to understand and respond to the unique needs and preferences of individual users. However, achieving true personalization and customization can be difficult, as it requires an AI system to have a deep understanding, not to say “deep feelings,” of the user and to be able to adapt its behavior and responses accordingly.

Processing large amounts of data for personalization and adaptation is a major challenge. The fact is, it's necessary to understand the needs and expectations of users by training and refining AI systems.

Another key obstacle to the development of more powerful NLP is data accessibility. When it comes to dealing with constraints on sharing documents that contain personal information, the landscape turns into an ethical minefield. To illustrate the idea with a sensitive subject happens with the electronic health records (EHRs) in the clinical domain.  Data, including user preferences, interactions history, demographics, bank, administrative, or health information, must be handled with care.

In any case, providing a truly personalized experience requires handling users' data and privacy with care. Indeed, users might not be comfortable providing personal information to AI systems and might not be willing to interact with AI systems that seem to know too much about them. See, for example, how reluctant we can be, even if only to give our email address or phone number or when it comes to turning on the camera.

Another challenge is dealing with the inherent complexity of human behavior. Individuals can change their behavior, preferences, and even their personalities over time; this makes it difficult for an AI system to keep up with the user and provide a truly personalized experience. But to be honest, human agents face the same difficulties.

In this regard, sentiment analysis can be a great help. Sentiment analysis is a subfield of natural language processing (NLP) that involves using AI algorithms and methodology to identify and extract subjective information from spoken text, written text or body movements, or facial expressions. 

This information can include the overall sentiment of a document or a speech, as well as more specific emotions and opinions expressed by the person. Techniques used in sentiment analysis are based, among other things, on lexical analysis, syntactic parsing, and ML algorithms such as support vector machines and neural networks.

Sentiment analysis can be used to enhance conversational systems' capabilities by allowing them to detect and respond to the sentiment expressed by the user in their input. For example, if a bot is able to detect that a user is expressing frustration, impatience, or dissatisfaction, it can respond in a way that addresses the user's emotions and tries to resolve the issue.

Sentiment analysis might be useful to monitor user feedback and to get an understanding of the user's overall experience, which can be used for further improvement of the conversational system.

Recent studies show that for industries with large customer bases, customer care and personalization of products and services are among the most important AI use cases. Do you know that the most common practice in call centers is to answer 80% of calls within 20 seconds? This puts teams under stress to maintain a good level of customer satisfaction. This stress problem could be answered by simple requests addressed by bots releasing pressure on the waiting queue side. These industries include travel, hospitality, consumer goods, retail, and telecommunications. 

While the technology behind AI-powered personalization is improving, it's still difficult to get it designed the optimum way, and the majority of companies are still far from providing a truly personalized experience. 

Potential for Bias

Bias are phenomenons that appear when algorithms deliver systematically biased outcomes because of wrong assumptions of the machine learning process. Unfortunately, AI takes on the tendencies of human prejudices, whether they are conscious or unconscious. This results in behaviors that can be racist, homophobic, misogynistic, or any other type of discrimination.

This is how the bias in AI systems can make predictions or decisions that are discriminatory or unfair to certain groups of people. This can be particularly problematic in the context of Conversational AI, as these systems are often used to make decisions or provide information that can have a direct impact on people's decisions in real-time.

Data and Algorithms: The Two Main Sources of Bias

Data

Maybe the most important origin of bias in Conversational AI is the data that is used to train the system. When the data is not representative of the population that the system will be serving, it can lead to the system making predictions or decisions that are unfair or discriminatory to certain groups of people. Additionally, Conversational AI systems trained on historical data may perpetuate biases that were present in the past but should be corrected. 

Here's a well-known example. In 2018, Amazon's recruitment engine, created to analyze applicants' resumes, was found to be, unfortunately, biased against women in the recruitment process. This occurred because the recruitment algorithm was trained on resumes submitted over the past ten years, most of which belonged to men. When the algorithm was reviewed, it was found that it automatically discriminated against women.

Algorithm

The second source of bias in Conversational AI is the way the system is designed and implemented. For example, if a Chatbot is designed to provide customer service, it should be able to understand and respond to the needs of all customers, regardless of their background or characteristics. However, if the Chatbot is not designed to handle different languages, accents, or dialects, it can make it difficult for certain groups of people to communicate with the Chatbot and make it less effective for them.

The design of the algorithms themselves can reproduce the biases of human behavior. A 2019 UC Berkeley study found that "FinTech lenders" reject face-to-face applications from Latinos and African Americans about 6% more often than they reject applications from non-minorities under the same conditions. One study shows that between 2009 and 2015, lenders rejected about 1 million of their applications that would have been accepted in the absence of discrimination.

Unfortunately, it appeared that the algorithms were designed in such a way that they replicate this discrimination.

Studies found that language models are prone to the same forms of bias, including gender, race, and age bias, that can affect the performance of these systems and the outcomes they generate. These biases should be addressed with simple means, including better data curation, algorithmic transparency, and stronger regulations.  

And if it is necessary to remind it, algorithms are devoid of human character, such as common sense and especially empathy. 

Ethical Considerations

Directly related to the issue of bias, ethical considerations are crucial when dealing with Conversational AI. These systems are becoming the new user interface between a customer and a brand or a company, the same way employees are. They are therefore expected to behave in a dignified and appropriate manner. 

As Conversational AI systems become more sophisticated and integrated into society, it's important to ensure that they are developed and used in ways that are fair, transparent, and accountable. As a matter of fact, in 2021 in Washington, DC, members of Congress required businesses to assess automated decision-making systems used in areas such as health care, housing, employment, or education. The law requires employers to use external consultants to run independent assessments. This law makes sure their algorithms are not at risk for behaviors that engage bias based on sex, race, or ethnicity. 

Systems should not discriminate against certain groups of people based on their race, gender, age, or other characteristics. In the same spirit, it's crucial to ensure they are transparent so that people understand how the system is making decisions and can provide feedback or raise concerns.

Making AI as transparent as possible requires the ability to explain how and why AI-based decisions are made. The AI-based decision-making process must be aligned with and represent the values and ethics of the company.

The other equally important consideration is data privacy and security. Conversational AI systems rely on the collection and analysis of large amounts of personal data. It's crucial that the data is handled and stored in a way that protects the privacy and security of individuals.

There has been general consensus on the principle of ethics between all stakeholders for years now. Conversational AI must be safe, fair, and beneficial to all.  It must be transparent and accountable and must respect the privacy and security of individuals. Ethical considerations have to be integrated throughout the entire development and deployment process of AI, including digital assistants.

Future of Conversational AI

Predicted Advancements and Developments

There have been many advances in Conversational AI in recent years, and many experts believe that this technology will continue to evolve and become even more sophisticated in the future. Indeed, according to David Schubmehl, research director at IDC, "The Conversational AI tools and technologies market grew significantly during 2020 …/… IDC forecasts the overall market to approach $7.9 billion in revenue in 2025”.

One of the most significant advances has been the improvement of natural language processing (NLP) and understanding  (NLU) capabilities. Recent advancements are leading to systems that are more accurate and can understand a wider range of languages.

Another important advance has been the development of deep learning techniques, such as neural networks. These techniques have made it possible to train AI systems on large amounts of data and to improve their performance over time. The last big step in this area was performed by OpenAI with ChatGPT, a large language model (LLM) trained with massive amounts of data to predict the most accurately possible what word comes next in a sentence. According to Stanford University, “GPT-3 has 175 billion parameters and was trained on 570 gigabytes of text [making it] able to perform tasks it was not explicitly trained on like translating sentences from English to French, with few to no training examples”.

Other big players in Conversational AI are companies such as Google, Amazon, Facebook, Microsoft, and Apple. They are all investing heavily in Conversational AI and are developing a wide range of products and services, including virtual assistants, Chatbots, and smart speakers. On the other hand, there are also niche players, such as OpenAI, focused on developing a safer AI beneficial for humanity. These actors are working on developing cutting-edge Conversational AI technology. 

Microsoft fully understood OpenAI’s value and the potential of its solution. On 2023 Jan 10th, they reportedly plan to invest $10 billion in OpenAI as part of a funding round. This would bring the value of OpenAI to $29 billion.

Future advancements in conversational AI include increased use of AI-powered personalization and customization for more tailored user experiences. On another angle, solutions need to improve the ability to generate natural language for a more human-like interaction. This would lead to more efficient and accurate responses. Thirdly, the use of AI-powered sentiment analysis is expected to improve the ability of systems to understand emotions and respond accordingly. Finally, it’s also expected that Conversational AI systems will become more integrated into the Internet of Things (IoT). They will soon be able to control and manage a wide range of devices at home and in other environments, including the business and industrial ecosystems.

Potential Impact on Industries and Society

The advent of Conversational AI and the generalization of smart communication systems are expected to have a significant impact on society over the next few years. Here are some examples of potential impacts:

As Conversational AI systems become more sophisticated, they are expected to improve customer service by providing faster and more accurate responses to customer queries, 24/7. Conversational AI systems are expected to become more personalized and engaging, providing tailored experiences for users.

Conversational AI systems will improve productivity and efficiency by automating repetitive tasks and reducing human labor needs. The desired goal is to allow employees to focus on more complex tasks that require human skills, such as creativity, empathy, critical thinking, or moral awareness. They are also expected to provide more efficient and accurate responses to simple queries. These improvements would reduce the time and effort required to find and access information.

If you pay attention to pessimistic science fiction-like predictions, you fear that AI will start to take over many jobs once done by humans. Some of these predictions go to the extent that the growth of AI technologies will lead to widespread unemployment and social unrest. Not to mention robots taking over the world.

A closer look at the fear of progress and anxiety over change reveals that every innovation generates fears. The 2nd Industrial Revolution introducing new forms of energy, such as electricity, also carried the fear that turned out to be unfounded. The increased use of machinery would lead to a decrease in the availability of jobs for skilled craftsmen and artisans. The reality is while some jobs did become obsolete as a result of the increased use of machinery, many others were created, such as factory jobs such as transportation and manufacturing. The same applies to all the disrupting technology leaps.

We can predict that AI technology systems will become more and more sophisticated and even human-like. This will enable businesses to provide more personalized and engaging experiences for users. Finally, it will help to automate more complex tasks and save time for human workers.

Conclusion

With advancements in natural language processing and machine learning, it’s becoming increasingly possible for machines to respond to human language in a way that feels natural. This is leading to a wide range of new applications and more advanced forms of human-computer interaction.

However, despite these potential benefits, there are also a number of challenges that must be overcome to achieve the full potential of Conversational AI. One of the biggest challenges is developing systems that can understand and respond to human language in a way that is always contextually appropriate. This requires not only a deep understanding of the meaning of words but also the ability to interpret the context, the intents, and the sentiments in which those words are used in more subtle ways.

Another limitation of Conversational AI is the ability to understand and respond to open-ended questions or ambiguous queries. Sometimes the language and vocabulary used by humans are hard to understand by the system and lead to wrong answers.

More complicated, there is also a need for Conversational AI systems to be able to make decisions based on incomplete information. This requires the ability to handle uncertainty in a way that is appropriate to the very context, which appears to be more of a human skill than a mechanical skill. At least so far.