In the late 2015, the field of chatterbots or talkbots — in simple terms a conversing robot or computer program, and their development — was the talk of the town among data scientists. The reason for this may have been the chance to explore the area of addressing advanced user queries apart from the regular, casual conversations that happen among users on a daily basis. Although, they may not replace the human aspect completely, but they sure are on their way to make our lives better. This article presents a brief outlook in the recent developments in chatbot technology by suggesting five must-read research publications and articles:
Chatbots In The Indian Scenario
A recent publication by the Microsoft research team explored the idea of creating a chatbot catered to the Indian youth. The team said that the chatbot was designed to be more of a “friend” rather than a typical machine interface. They also claimed that a smartphone and a messaging application was sufficient to interact with the chatbot. The study is called the Wizard-of-Oz since it involves interaction between humans and computers. The term was coined by Dr Jeff Kelly at John Hopkins University in his dissertation. In the paper, the team developed three personalities for the bot — two being friendly and helping with recommendations, and the other being empathetic. Now, to feed the information to these bots, they brought a person who had great expertise on communications and the English language. They labeled the person “The Wizard” for this experiment. Once the bots were ready, a total of 14 participants ranging from students, researchers and college professors — mainly from the field of Computer Sciences were brought in. The researchers felt the the results from the experiment were quite satisfactory because the recommendations made by the bots was almost in line with the user’s expectations. The participants felt good and preferred the bots with a friendly personality. The one and only drawback was that the bots couldn’t distinguish between privacy and personalisation. However, the team said there were improvements made on this limitation. Even more, as of 2017, they already are testing multiple persona on these bots to accommodate the likes and dislikes of the younger generation.
Subscribe to our Newsletter
Join our editors every weekday evening as they steer you through the most significant news of the day, introduce you to fresh perspectives, and provide unexpected moments of joy
Information Retrieval And Sequence To Sequence Based Chatbot Engine
A group of employees from the Alibaba Group presented a technical paper on how a chatbot would be more than beneficial in the e-commerce sector. They proposed a novel integrated approach of two methods — Information Retrieval (IR) and Sequence to Sequence (Seq2Seq) models for the chatbot. They developed a Q&A query-base from the customer log available in their company’s database which served as the input for the chatbot engine. The procedure began with pairing the questions and responses — approximately an index of 10 million question base, with a threshold for those queries. If the responses fell under the threshold then IR method was used, otherwise the sequence model was followed. The sequence model was expressed in a defined probability range of words ( = y1 , y2 , …….ci ). Once this step was achieved, three implementations were chosen to develop the sequence model. The bucketing and padding (in TensorFlow), Softmax regression and the Beam search. Consequently, an attentive sequence model was developed using a candidate’s query or word pattern by a standard formula. After all of this, the approaches were integrated as per their choosing and evaluated for results. The results seem favorable with 40 percent of the queries by the bots falling under the appropriate response for the huge customer base, which is a gain for the sector.
Virtual Doctor On-The-Go
Saurav Kumar Mishra and his team, academic professionals from the Indian Institute of Information Technology, Allahabad developed a “virtual doctor” chatbot program to serve patients when there were no doctors available or during emergencies. Python is the primary coding language used to develop algorithms. Natural Language Toolkit (NLTK) and PyQt for GUI, were used throughout the study. The algorithm used two important criteria: patient queries and ailments or diseases. The procedure included tokenizing the alphabets from the patient or the user, generating meaningful words to match the user-response developed from the professionals to the diseases, and tagging those words to provide the correct response. 80 percent of the responses were right from the bot, which is good news for AI development in the medical field.
ELIZA — The First Chatbot
A recent article by Jane Abrams, a tutor at Grok Learning — an online platform for learning programming languages — showcases the development of ELIZA (featured in the TV series, Young Sheldon). It was the first chatbot developed by Joseph Weizenbaum at MIT in 1964, who used a series of pre-recorded patterns and substitution of those patterns in response to the query provided by patients with psychological disorders. Initially, he wanted to disprove the illusion of self- learning machines — now called AI. He was impressed with the response and quite afraid with the harmful implications it could have on humans in the near future. In the article, the writer mentioned two reasons why it was a good experiment.
- The ability of the chatbot to recognise the pre-recorded patterns.
- The transformation it could do around those patterns in an appropriate way.
The point here is the extent to which machines could be made ‘intelligent’ by incorporating human factors such as emotions. ELIZA served as the turning point for all the advancements in the field of chabot development. In the coming days, there is a possibility that robots will act exactly like humans — unpredictable to a certain extent.
Last but not the least, a thesis by Jakob Aberg of Umea University, Sweden, explores the possibility of using chatbots to reshape behaviour in humans towards sustainable environment. An exhaustive research was done before developing the bot such as Eco-feedback technology (addressing environmental issues to fulfil human needs) and traditional user queries and interactions among systems. This thesis provides an in-depth study of all the models related to environment protection and the human factor surrounding these issues.
The papers and information presented above show the positive impact a chatbot can bring in the rapidly-changing technology ecology. On the other hand, if it falls in the wrong hands the advantage is misused and could lead to a disaster. For example, Facebook’s chatbots developed their own language to communicate with each other, earlier in June 2017, which led to a great controversy and kept the employees baffled.