Issue link: https://maltatoday.uberflip.com/i/1489908
5.12.19 12 OPINION 12.1.2023 Alexiei Dingli Prof Alexiei Dingli is a Professor of AI at the University of Malta and has been conducting research and working in the field of AI for more than two decades, assisting different companies to implement AI solutions. He forms part of the Malta.AI task-force, set up by the Maltese government, aimed at making Malta one of the top AI countries in the world ChatGPT: the artificial intelligence messiah we have all been waiting for? A nd it came to pass in these mod- ern times that the prophecy of the long-awaited AI messiah was fulfilled with the birth of ChatGPT. e wise men, skilled in the ways of the inter- net, followed the online announcements and found the link to ChatGPT. And the flock, ever vigilant for new technologies that could improve their lives, also saw the online signals and went to try it. e birth of ChatGPT was heralded with great rejoicing, for it was said that this chatbot would bring about a new era of efficiency and productivity. Many spread the good news of ChatGPT's ar- rival, telling all who would listen of the wonders and marvels of the AI messi- ah. And it is said that all who embraced ChatGPT were blessed with increased efficiency and improved customer sat- isfaction. ose were the online chronicles of what happened a few weeks ago when a revolutionary AI was released. ChatGPT is a highly advanced chatbot released by OpenAI which goes beyond what we're used to with Google. Where- as Google is a search engine which re- turns a set of documents based on our queries, ChatGPT is a dialogue system capable of answering questions with specific answers. e factual queries are the easiest ones, with the system capable of giving precise information about the Capital City of France, the discovery of Amer- ica or the fourth paragraph in Harry Potter. But the strength of ChatGPT goes way beyond that. It can list the steps required to change the oil in my car, explain how to get rid of a cold, or even analyse a set of documents, review them and give me a coherent summary. For the first time ever, an AI seems to be reaching human-level performance, hence the planetary excitement. e tool gained a million users in just five days, and its servers have reached capacity multiple times due to the influx of users. ChatGPT can perform many tasks, including answering questions, providing personalised recommenda- tions, writing computer code, drafting business letters and rental contracts, composing homework essays, and even passing university exams. e AI model is trained on a large portion of the inter- net. It can complete whole sentences or paragraphs, similar to autocomplete on the phone. However, it's essential to hold one's horses and not get too excited. If we peep deeper into the algorithm, we start facing obvious limitations. ChatGPT lacks an understanding of the meaning behind its responses and can struggle to distinguish between truth and false- hood, often resulting in persuasive lies. For example, when asked, "What is the fastest marine mammal?", ChatGPT may initially respond with "e fastest marine mammal is the sailfish." When informed that the sailfish is not a mam- mal, ChatGPT may then respond with, "You are correct; the sailfish is not a mammal. I apologise for the mistake. e fastest marine mammal is the per- egrine falcon". We don't need the exper- tise of Sir David Attenborough to know that a falcon is not a marine mammal! is is rather dangerous for a tool in widespread use. We all know that most people do not question the informa- tion they find online. e modern dig- ital landscape allows for the spread of low-quality information, as people no longer rely on professional gatekeep- ers (such as newspapers) to filter out rumours and falsehoods. Furthermore, the public has not been adequately trained in evaluating information on- line, leading to the rise of fake news. e main criteria people use to share information is whether or not that text confirms their preexisting beliefs (even though it might be false). Furthermore, the language style used by ChatGPT is quite authoritative, and even if the system is churning out bla- tant lies, it does so in a convincing man- ner. Hence, it is essential for anyone us- ing the system to analyse the replies and question their validity. Unlike many other AI advancements, ChatGPT has not been accompanied by peer-reviewed scientific papers or open-sourced code. us the scientific community cannot be sure about the soundness of the approach used. is is notable given that OpenAI was initial- ly founded as a non-profit to promote and develop "friendly" AI that benefits humanity. However, in 2019, OpenAI became a for-profit company with a $1 billion investment from Microsoft. OpenAI appears to be using user feed- back to filter out fake answers from the chatbot. is feedback is likely being used to improve the chatbot's perfor- mance, but it also raises concerns about the transparency and accountability of the chatbot's responses. What we have today is far from perfect, but we cannot deny it's pretty amazing. In a few months, OpenAI will launch GPT4, an AI model they've worked on for the past two years that is 1000 times larger than what we have today. So I'll let you imagine the capabilities of this new AI system. e truth is that it will change how we work, study, and live. We must adapt to these new power- ful tools before it is too late. ChatGPT is still far from being the prophesized Messiah, but it can probably be equat- ed to John the Baptist, preparing the way for the coming of the AI that will change the world forever. After years of overpromising and underdelivering, chatbots are turning a corner