Artificial intelligence, a valuable but not risk-free tool
Institutional Communication Service
9 April 2024
Artificial intelligence is becoming more prevalent in everyday life and is advancing to the point where it can interact with users in a way that is perceived as almost human-like. Yet, despite the constant progress, the risks associated with the use of AI remain significant. Professor Emily Menon Bender, a leading researcher in the field of computational linguistics and lecturer at the University of Washington, tackled the issue in-depth in a lectio magistralis on Friday at USI, as a guest speaker of the Doctoral Programme in Applied Linguistics: Managing Languages, Arguments and Narratives in the Datafied Society (LAND), in collaboration with USI Institute of Argumentation, Linguistics and Semiotics (IALS), and he Department of Applied Linguistics of the Zürcher Hochschule für Angewandte Wissenschaften (ZHAW), funded by swissuniversities.
The American linguist, a leading figure in the field of AI, presented her view on the topic. She first explained why we tend to assume sense-making ability in interactions with artificial interlocutors. "When we interact with chatbots, such as ChatGPT, they provide us with words that make sense to us, that we understand," Professor Bender pointed out. "Usually, this happens when we come into contact with another person, so we tend to imagine a mind behind these tools, which is not there. The language comes from a non-existent place, and it is we who give it meaning."
The idea that artificial intelligence is capable of doing anything, Emily M. Bender notes, is, therefore, a false myth. And this is true no matter how much we "train" it on a particular subject. "We have machines that talk about medicine, law and sports, but they are actually neither doctors nor lawyers," continues the researcher. "They are only providing us with language because they master the form of it (syntax, pronunciation and spelling), but they do not have access to meaning, to semantics. It is therefore better to think of them as tools, from which we ask for support in case of need". Tools which, however, the professor emphasises, are by no means risk-free. "I'll give you a concrete case: in the United States, the National Eating Disorders Association replaced the staff of its hotline with a chatbot, which, however, at a certain point started to dispense erroneous advice, suggesting, for example, that people who already had eating disorders should go on a diet. This episode, which ended with the suspension of the service, well denotes the limits of artificial interactions. The problem is that for such errors, the AI is not held responsible".
The increasing demand for artificial intelligence is becoming a risk to the environment, according to the USI guest. The consequences of AI go beyond screens and speakers. "The impact of artificial intelligence on the environment is a major concern. The training process involves the use of numerous computers and a significant amount of energy to process different language patterns. Additionally, water is required to cool the data centres, and materials are needed to construct the chips. These factors significantly impact the ecosystem, yet they are not disclosed to the consumers. Moreover, the process of AI training is constantly expanding, which implies that the environmental impact will only increase over time." The future of AI, it goes without saying, is therefore shrouded in more than one uncertainty. However, Emily M. Bender's vision in this regard is clear: "I see two possible future scenarios, one positive and the other negative. In the first case, artificial intelligence will decline into a specific and well-designed application. I am referring to translations, transcriptions, automatic corrections and so on. Second, on the other hand, AI will decline into fake tests and synthetic tests everywhere, with people pretending that certain answers can be education, psychotherapy, a good legal contract, etc. It is clear that the second scenario, should it see the light of day, would cause enormous problems".
Below you can find links to interviews and features that have appeared in the media: (in Italian)
- https://www.teleticino.ch/ticinonews/ticinonews-sera-ore-1830-050424-6004
- https://www.rsi.ch/rete-uno/programmi/informazione/seidisera/SEIDISERA--2091610.html