Researchers writing in the International Journal of Data Mining, Modelling and Management discuss the evolution and growing academic interest in chatbots, with a special focus on the widely discussed large language model known as ChatGPT. The team has analysed data from Scopus and the Web of Science (WoS) covering the period 1998 to 2023.
The results show remarkable annual growth rates of almost 20 percent in WoS and almost 30 percent in Scopus in the number of publications discussing chatbots. This obvious conclusion is that there is a growing scholarly interest in this kind of artificial intelligence and its use in “conversational” software agents.
Hamed Khosravi, Ahmed Shoyeb Raihan, and Imtiaz Ahmed of West Virginia University in Morgantown, USA, and Mohammad Reza Shafie and Morteza Hajiabadiof the Iran University of Science and Technology in Tehran, explain how chatbots use algorithms trained on natural language databases to mimic human conversation. These AI tools can respond with apparently cogent answers to questions or prompts and are increasingly used in customer service, education, mental health, financial management, and many other areas. Users need to be aware that while the term “intelligence” is used in this context, the software is not inherently intelligent in the conventional sense and can readily generate fanciful or even false information in response to a given prompt. Nevertheless, on the whole, these tools can be very useful in streamlining interactions and precluding the need for human staff to undertake many mundane tasks.
These systems employ natural language processing (NLP), a subset of AI that allows machines to interpret prompts and generate an apparently human response. In the current work, Khosravi and colleagues focus on one of the more well known systems, ChatGPT. This tool uses deep-learning techniques to generate contextually relevant and coherent responses. Its advocates point to its more advanced abilities when compared to the previous generation of chatbots and other AI tools in this area.
In examining the research literature, the team notes that there has been a shift towards areas such as mental health and task analysis and how AI tools might be used in those contexts and what its limitations might be. The way AI is being used now and how that is changing will, of course, affect how development moves.
There is a pressing need to ensure that the tools are not only technologically superior but also ethically sound and contextually aware. The next generation of AI tools may need less human oversight, but there will perhaps always be a need for some human supervision of outputs. This progression could lead to AI applications that provide preliminary medical insights, support clinical decision-making, enhance writing and translation tasks, simulate organizational interactions, and assist in policy formulation.
Khosravi, H., Shafie, M.R., Hajiabadi, M., Raihan, A.S. and Ahmed, I. (2024) ‘Chatbots and ChatGPT: a bibliometric analysis and systematic review of publications in Web of Science and Scopus databases’, Int. J. Data Mining, Modelling and Management, Vol. 16, No. 2, pp.113–147.