The History And Evolution Of Conversational AI

Do you remember Jarvis, the highly intelligent Artificial Intelligence (AI) from the Marvel Studios film Iron Man, who could perform all duties just by following Iron Man's vocal commands? Do you wish to communicate with your deceased family members using a gadget that can perfectly replicate them?

These things sound so strange, that they only appear to be feasible in sci-fi movies and in a madman's imagination.

Different people might perceive this from a different point of view but with the advent of Artificial Intelligence through devices such as — Alexa, Siri, Google Home and Cortana, this dream is not so far from reality. It is very much possible in present days with a technology popularly known as — Conversational Artificial Intelligence.

Let's take a closer look at what Conversational AI is and what its subsidiaries are.

 

What is Conversational Artificial Intelligence?

They are devices or technologies, such as chatbots and voice assistants, that allow humans to communicate with them and do particular activities using voice commands. They are capable of distinguishing between speech and text inputs, as well as translating between languages. They mimic humans in such a way that it gives the user a vibe that they are talking to a real human being! Conversational AI requires huge amounts of data along with machine learning and natural language processing techniques to make human-like interactions a reality.

Now we will briefly know about some of the major components of Conversational AI.

1. Natural Language — It consists of sentences, words, letters and characters that vary across many cultures. For instance — English has 26 letters whereas Russian has 33 letters. Important properties of natural language rather than vocabulary are syntax or grammar rules, semantics or meaning and linguistic ontology, i.e, the relationship between words, sentences and phrases. Natural Language often includes words with uncertain and different meanings and even words in different dialects and also errors and a variety of accents. There are words with same pronunciations but have completely different meanings which are known as homophones and words with same spelling or pronunciation but different meanings which are called homonyms.
2. Voice —This is a very integral part of Conversational AI as it is the main driving force of the technology. There are Voice User Interfaces in chatbots or voice assistants which allows the users to interact with Conversational AI through their voice. For example — there are voice enabled lights for Google Home which can be switched on or off using commands like — “Ok Google,lights on” or “Ok Google, lights off”. You can also tell Google Home or Amazon Alexa to wake you up in the morning or you can ask it to play your favorite song or even motivational quotes as well.
3. Natural Language processing — It is a subdomain of artificial intelligence which carries out certain steps which allows machines to understand human language. It uses machine learning, deep learning and some rule-based modeling. It mainly aims to produce appropriate responses by understanding the intent of human language inputs to create a conversation flow which looks natural and human.

It can be subdivided into 3 main parts — Natural Language Understanding (NLU), Natural Language Generation (NLG) and Dialog Management. Natural Language Understanding (NLU) — It shed light on the intent behind every input. It helps a machine to comprehend spoken or a written language.

Dialog management — It decodes the intent and classifies it depending on business rules. It keeps the conversation intact with appropriate questions and responses.

Natural Language Generation (NLG) — It focuses on forming a response which is both accurate and precise. It generates words, phrases and sentences that have contextual meaning and can be comprehended by a human user.

4. Intent — It is very important for a Conversational AI to understand the intent of a speech or text, as words can contain different meanings depending on the user in different contexts. It must recognize the intent despite the sequencing of the words and the way they are used and must produce an appropriate response.

 

A Brief history of NLP — To get a concrete idea about Conversational AI, we must know some of the history of NLP and how it evolved into what it is today. I will list down some of the major events of NLP until the year 2020.

1939 — Speech Synthesis by Bell Labs.

1950 — Turing Test.

1952 — speech recognition by Bell Labs known as Audry.

1960 — Text Bot by MIT — known as Eliza.

1962 — Speech recognition by IBM

1971 — Speech recognition by DARPA and Harpy from Carnegie-Mellon University.

2006 — NLP by IBM Watson — Won Jeopardy against best players in February 2011.

2011 — NLU + NLG by Apple in Siri

2014 — NLU + NLG by Microsoft with Cortana and Amazon with Alexa

2016 — NLU + NLG by Google Assistant

2017 — text by Facebook Known as Bob + Alice

2020 — OpenAI’s GPT3

What is a Chatbot?

The word Chatbot was first used in the 1990s. Although the concept and the basic technology has been around since the 1950s. Its earlier version was known as chatterbot. The use of Natural Language Processing in chatbots is a comparatively recent update in the technology. This idea has become more popular through the work of Facebook Messenger, Apple’s Siri, Amazon Echo, Google Home, etc. The previous technique used to build chatbots required several months. Now, with the help of NLP and deep learning, it is possible to build one within only a few hours.

We will know take a look into some of the best known and widely used chatbots ever —

1. Eliza, 1964–1966 — It is one of the first natural language processing computer programs developed by Joseph Weizenbaum at MIT Artificial Intelligence Laboratory. It was made to show the artificiality of communication between humans and machines. It used pattern matching and substitution methods to give the users an illusion that they are understanding the context of the words.
2. Cleverbot , 1997–1988 — It was developed by a British AI scientist Rollo Carpenter. It is a chatterbot web application that utilizes AI to start conversing with humans. It uses NLP and fuzzy logic. Fuzzy logic is used to handle a million records stored and used in a heuristic manner. Cleverbot is now trying to implement ML techniques to become more effective.
3. Mitsuku , 2002 — A web-based chatbot created by Pandorabots and was awarded the annual Loebner Prize in 2013 and 2016 for becoming the most human-like chatbot. It is a virtual friend which can answer questions, play games and do tricks if requested by the user, and it has basic reasoning.

 

 

4. Rose, 2011 — Brillig Understanding, Inc. built this award-winning chatbot. It's modeled on the characteristics of a teenage girl, and the owner claims to have given the bot its own personality, but it's all based on a code. The key goal for this bot is to be able to express its own feelings and cognitive processes. It is also heavily reliant on pattern matching. It has used a C-style general scripting like language to code it.

 

5. XIaoice, 2014 — It is the AI system developed by Microsoft STCA in 2014 based on an emotional computing framework. It has a unique design and works as an AI companion and tries to make an emotional connection. This bot used Markov Decision Processes (MDPs) which optimizes it for long term user engagement.

 

6. DialoGPT , 2019— It is a Joint production between MSR AI and Microsoft Dynamics 365 AI Research team. It is able to create engaging and natural conversational responses across a variety of topics. It is trained using data collected from 147 million comments done on reddit. The main program is based on hugging face pytorch transformers.

 

7. Meena, 2020 — It is a chatbot that can sensibly give answers by staying in context of the conversation. This has been termed as google as “neural conversational model”. It was a state of the art system before the recent announcement of Facebook’s Blender and was ahead of the entire race with the highest Sensibleness and Specificity Average (SSA) scores. The main aim of Meena is to address the critical flaw in chatbots of them not making sense.

 

8. Blender — It is the facebook’s latest chatbot in collaboration with parlAI and can combine multiple conversational skills at once and for that reason is named as blender. The chatbot is built from upon 9.4 billion parameters and trained using 1.5 billion examples of conversation, making it so huge that it must be split up into pieces in order to tackle larger sets of data.
9. IBM Watson — IBM Watson has several prebuilt conversational AI applications such as — Watson Natural Understanding for advances text analytics, Watson Speech to Text to transform voice into written text with powerful machine learning technology and also Watson Text to Speech to convert written text into natural-sounding audio in a variety of languages and voices and many other similar applications which are transforming businesses.

 

 

 

Future of Chatbots and Conversational AI — As a research area, chatbots and conversational AI are fascinating, and the possibilities are endless. Companies are willing to invest millions of dollars in this field since it has great potential. Companies are increasingly delving into this research, concentrating mostly on creating chatbot personalities and their distinct characteristics. State-of-the-art research in this field has recently begun, but it is already yielding impressive results, bringing a slew of new businesses to the industry.

Similarly in Bangladesh, Intelsense AI is working in this research field to produce some superb chatbots. Intelsense AI is helping to increase the revenues of commercial and non-commercial companies with millions of users, by integrating our proprietary “Full-Aware” technology with their existing ‘Customer Relationship Management’ facilities. The first of our offerings using “Full-Aware” is HIA, a user-focused personalized AI virtual assistant. We are also helping to carve out new portions in their existing market pie by integrating “Full-Aware” AI chatbots- SenseBots, and “Full-Aware” Social Media analytics tool- SenseInsight. The team also is currently developing “Full-Aware” visual recognition solution- ViSense (early Alpha), and high-frequency Trading bot- SenseTrade. Each month we reach 2 million people and are constantly growing. As our user base grows, our “Full-Aware” technology becomes exponentially faster and more efficient using Deep Learning Algorithms. This, in turn drastically improves all our services before you even have a sense for it. The benefits that our clients get using our technology are as diverse as our client base. That’s what makes it so exciting! So, why wait when your competitors might be getting ahead using the boons of Artificial Intelligence? Why give them a slight edge, even if it's only for a fraction of a second? The world of Intelligence is moving beyond our understanding, are you moving along with it? Let’s help you Sense The Future, as we take care of your needs.