ITU researcher: Generative AI needs a reality check

Assistant Professor in Computer Science, Anna Rogers, is a new face at IT University of Copenhagen. She conducts research on Natural Language Processing (NLP), and her research makes it easier to determine whether Transformer-based language models, which are used in systems like ChatGPT, are reliable or not.

Anna RogersComputer Science Departmentartificial intelligence

Written June 23, 2023 11:32 AM by Ditte Ørsted Johansen

On March 15, Anna Rogers started her new position as Assistant Professor of Computer Science at ITU. Her field of research has garnered a lot of attention in the past months. She specializes in NLP and Transformer-based language models which is the core technology behind systems like ChatGPT.  

Anna Rogers’ career journey has taken her all over the world. She completed her PhD in Computational Linguistics at the University of Tokyo, and she was a post-doctoral researcher at the University of Massachusetts and University of Copenhagen. What brought her to ITU is the university’s thriving research environment. “The university also provides me with a strong research support infrastructure and a research group which inspires me to dig deeper into NLP,” says Anna Rogers.   

Analyzing Transformer-based language models

Anna Rogers analyses and evaluates Transformer-based language models. This technology is a specific type of deep learning networks, usually rather large in terms of parameters and pre-trained on large volumes of text. It enables computers to process human language and conduct certain tasks, like translating, transcribing, answering questions, or summarizing large amounts of texts in a matter of seconds.  

“I am analysing transformer-based language systems. I study how these language models are trained, how they learn so well, and how we can tell that they are reliable and do what people tell them to do,” says Anna Rogers. “As part of the model training, the models are asked to predict missing words in sentences. The idea is that after doing this task for a long time, the language models learn enough language and enough facts about the world to be useful in real tasks.”

Programme Chair at AI-conference

Anna Rogers has a busy summer ahead of her. Alongside her work at ITU, she is one of three programme chairs at the world-renowned AI conference, The 61^st Annual Meeting of the Association for Computational Linguistics (ACL). Association for Computational Linguistics is the biggest professional organisation for NLP, responsible for most of the top conferences in this field. ACL 2023 will take place in Toronto in July. “We received over 5,000 submissions at this conference, and my role as a Programme Chair is to organize the peer review process and select the best papers,” says Anna Rogers.

This conference is an opportunity to discuss both challenges and possibilities related to NLP and in particular Transformer-based language models. “Generative AI is a remarkable accomplishment, but it is not as advanced as many news headlines suggest. The special theme this year is called “Reality check” because we would like to direct attention to what it really can do, and when it can be used appropriately”, says Anna.

Anna looks forward to working with ITU students on safe and reliable NLP technology. In the next two semesters, she will teach Reflections on Data Science and Seminars in Data Science.

Further information