ChatGPT is a machine learning language model which has seen a sudden rise of use and popularity since its free access to the public. University of Helsinki along with many other universities have needed to address the use of ChatGPT in the context of academic work in teaching and in research.
To university crowds panel reflected on “What is ChatGPT?” and what it means in the academia. Should we be worried about making it easier to fool people? Excited about new possibilities for ways to do research and research about? Are the expectations about ChatGPT simply blown out of proportion? The whole panel discussion from a multidisciplinary perspective including computer sciences, social ethics, methodological implications, cognitive science and policy research, is available online.
Jaana Hallamaa Professor of Social Ethics, Faculty of Theology in University of Helsinki
Petri Myllymäki Professor of Computer Science in University of Helsinki, specialising in artificial intelligence and machine learning
Matti Pohjonen University Researcher, Helsinki Institute of Social Science and Humanities, Methodological Unit, Researcher, Digital Methods
Anna-Mari Rusanen University Lecturer, Department of Digital Humanities, Cognitive Science, specialising in philosophy of AI & cognitive sciences
Arho Toikka University Lecturer, Faculty of Social sciences, Social and public policy, research in natural language processing methods
Panel was chaired by Emilia Palonen, Research Programme Director in Datafication at the HSSH, and leader of the HEPP research group in Political Science, Faculty of Social Sciences.
The panel discussed how ChatGPT is primarily a new tool which shifts the balance of work between people and machines in novel ways which may have some implications on human cognition. Petri especially stressed the fact that ChatGPT and large language models should be considered separate from artificial intelligence. Furthermore Anna-Mari also insists on doing away with language that involves discussing intelligence or consciousness of these systems.
Universities are becoming more and more reliant on technology and a variety of tools such as computers, phones, microphones and cameras as pointed out by Jaana. This raised the question about the implications for this rising datafication and tool reliance in university and what is ChatGPT’s role in it.
The current utility and ability of ChatGPT is blown out of proportion and is more limited than it appears though technology moves fast, and things may change quickly. Arho highlighted how these technologies are owned by private entities and may embellish their capabilities. Petri also considers that the fears of plagiarism are overemphasised as having someone else do your work has always existed.
Global dataset use of ChatGPT biased in favour of western world. Lot of developing countries are outside of the used dataset as well as having limited access to OpenAI and ChatGPT. These points were specially highlighted by slides Matti had prepared for the panel.
Bias of ChatGPT and language models was further discussed by Anna-Mari and Matti and Jaana. They wonder how sensitive people are to recognise bias, invisibility and visibility of bias in the production of language models especially when comparing commercial and open-source models and historical bias of recorded information.
The lack of intelligibility in the machine learning systems was explained by Petri. Further pointing out the human susceptibility to be fooled by the illusion of intelligence, meaning and emotion when there is none, creating hallucinations as is generally used in this field. Though according to Jaana the stochastic mimicry of language model could allow for greater introspection
ChatGPT and large datasets create representation of cultures, societies and communities which may create new ways for sociological research and a way to diagnose elements of human society. Matti sees that this may result both in new research questions and research methods.
Development of AI, language models and machine learning is not transparent. Not only do we need to consider that most development happens in the United States, most development is also done by private companies rather than public and academic institutions. Petri was worried about the transformation of the production of science in that it is being taken over by commercial private entities.
ChatGPT and machine learning might not be sustainable in ecological terms as their development starts to include even larger datasets, requiring more energy. Costs of use was discussed by both Petri and Jaana, pointing out the unknowable nature of costs as it increases along with development but also the value of information and data it collects from its users such as university students and researchers who actively or passively train it.
Finishing the panel Emilia wondered if using ChatGPT contribute to sustaining the existing quantity over quality mentality and academic capitalism in universities and production of research as language models make production text easier than ever?