ITU researcher receives funding to improve human-centered AI
PhD student at IT University of Copenhagen’s Department of Computer Science Elisa Bassignana has received an International Postdoc grant from Villum Fonden to gain a better understanding of how large language systems are encoded with social, cultural, demographically defined information and values.
Most of us have had tiring encounters with customer service chatbots and other voice-controlled technologies. AI is becoming a staple of everyday life, and the technology has recently taken another quantum leap, but when interacting with human users, AI applications still fail to grasp the nuances of human communication – the sentiments and emotions expressed between the lines give machines with their categorical approach to information exchange a hard time.
“I want to improve the quality of human interaction with AI technology, so perhaps the technology at some point will be better at understanding what users expect and need from it,” says Elisa Bassignana who has just received a grant from Villum Fonden’s Villum International Postdoc-programme.
The researcher’s project is entitled Personality and Emotion-Aware Computing for Human-Centered AI and focusses on the Natural Language Processing aspect of artificial intelligence. By critically examining large language models from a linguistic, socio-demographical perspective, Elisa Bassignana hopes to be able to further AI’s capabilities in human-centered interactions.
“My project will focus specifically on two aspects: personality and emotions. Ultimately, I want to integrate these factors into language models so the models can be personalised to a greater extent. The scope of this project is core research, but obviously this would benefit a wide range of practical applications, such as chatbots, voice-controlled systems, and other language-based technologies.”
As the researcher also points out, there are some obvious pitfalls to avoid when improving AI’s capabilities in this arena. After all, AI technology capable of adapting to personality and emotional aspects could have harmful effects if used for nefarious purposes. Elisa Bassignana’s project is not focused on the ethical aspects of the problem, but according to the researcher it is something to be acutely aware of.
However, the benefits outweigh the downsides at this point in AI development:
“If we can develop AI that is capable of examining its own biases – the culturally, sociologically, and demographically specific values in the language models – we will be able to create a fairer technology that is free of, for instance, gender bias. Ultimately, I want to find out what is encoded in the language models and how we can diversify this type of latent information, so a system isn’t limited to one specific social demographic.”
As part of the project, Elisa Bassignana will conduct research at ITU and at Bocconi University in Milan, Italy. Theis Duelund Jensen, Press Officer, tel: 2555 0447, email: email@example.com