Chatbots, also known as conversational agents, are computer programmes based on artificial intelligence – using natural language processing techniques – that simulate and process human conversations to allow users to interact with the programme in an automated way. This technology appeared in the 1960s and its most famous ancestor is ELIZA. Chatbots have quietly evolved in the field of applications that require conversation-based interaction to obtain results from any intelligent service. 

They materialized on 30 November 2022 when a prototype named ChatGPT burst onto the media scene of emerging AI-based applications like a whirlwind. By January 2023, it had 100 million users, a meteoric record in terms of community adoption and acceptance. It also holds the record for monetization of investment and is perhaps one of the most written-about and commented-on programmes ever. At the end of February 2023, it has already been acquired by and incorporated in the Microsoft system.  

One of the most disturbing aspects of this prototype is related to our inability to measure the plausibility of the results obtained and that which can be attributed to them. One should start by saying that plausibility is high for a particular context and within a particular time range, and the data used to train the prototype included data available up to 2021. The results obtained rival those that could be generated by a human but, as Ricardo Baeza-Yates observed:  

“Paraphrasing and expanding on a famous quote on advertising from over 100 years ago: we know that #ChatGPT works, say, 90% of the time. The problem is that we do not know which 90%, and worse, #ChatGPT doesn’t know either! Corollary: put 99% instead of 90%, and the situation might be worse.”

The ethical and legal issue is that these results could be used without being checked against real sources; they could be passed off as one’s own production, resulting in a long list of possibilities that are not new in the digital world but that the emergence of tools like this accelerates. It is completely incomprehensible that the results do not come with a standard watermark stating that they are authored by ChatGPT or any other machine, and if they use texts from others, they should state this.

On 27 August 2022, just before ChatGPT, another also apparently remarkable chatbot-like tool appeared. DALL.E-2, from OPEN-AI, is a system based on deep learning technologies that can create realistic images and art from a description in natural language. For three or four months, it was the focus of media attention, but its popularity never reached the heights of its successor. As I finish writing this article, a new toy called LLaMa has been introduced by Meta. Only time will tell if it dethrones ChatGPT or if the tool released by Alphabet will be the brightest star in this firmament.

Ulises Cortés is a full professor and researcher at the Universitat Politècnica de Catalunya and has an honorary doctorate from the Universitat de Girona.

AI tools for interpreters by Joshua  GOLDSMITH:

1