Artificial Intelligence is a hot topic these days, inspiring both excitement and fear in equal measure. One of Maseeh College’s impressive Computer Science faculty members, Ameeta Agrawal, recently discussed her unconventional journey to the field and her hope to develop computational technology that will serve diverse communities. In addition to Agrawal's recognition as an esteemed instructor and a beloved figure among computer science students, she maintains a commendable track record as a successful researcher. Her most recent grants include one from the National Science Foundation (NSF) to address persistent disparities in Large Language Models (LLM) treatment of text across linguistic and writing styles. Her focus is on bridging these gaps by crafting models capable of equitable handling of text from diverse social groups. Another noteworthy project, also funded by the NSF, looks to re-imagining the Role of Humans in Security Education. This collaborative endeavor with Wu-chang Feng, a computer security expert and professor at PSU, promises an exciting exploration of the intricate dynamics among educators, students, and AI within the educational realm. Their aim is to uncover innovative approaches to harness AI's potential for enriching student learning experiences while addressing lingering concerns regarding its long-term impacts on the educational landscape.
With a multilingual background and fluency in English, French, Hindi, and understanding of several Indian and one African language, Agrawal's expertise extends to programming languages as well. Agrawal always wanted to combine her love of computer science and languages and eventually found her passion in natural language processing (NLP), the branch of artificial intelligence dealing with giving computers the ability to understand text and spoken words in much the same way human beings can. She stumbled upon a project focused on emotion detection from text during graduate school, which sparked her interest in this subfield of computer science. She and her dedicated graduate students in the PortNLP Lab continue pushing the boundaries of research in efficient and multilingual NLP, with the goal of harnessing the potential of large language models to better address complex societal challenges.
She highlighted the significance of ChatGPT, a large language model released in late 2022 by OpenAI, which garnered widespread attention due to its low barrier to entry, both in terms of price-point and its user-friendly interface. Agrawal pointed out that ChatGPT is neither the first nor “the only large language model or generative AI model out there. . . . The thing that really propelled ChatGPT to where it is right now is just having that very, very accessible interface and just making it available to anybody."
However, she also pointed to its limitations, particularly in accurately dealing with text beyond English. Moreover, Agrawal cautions against placing blind trust in AI models. Her “biggest concern [is] how limited people have an understanding of what's going on. Why is it working? And then you see people, especially students, really blindly trusting everything.” While LLMs can be valuable educational tools, users must be cautious and corroborate information from various sources. She raised concerns about the dangerous territory of relying on proprietary models like GPT-4, where researchers have limited access to understand the inner workings, data sources, and potential biases or the algorithms.
The fluidity and flexibility of her educational pathway is also reflected in her AI-inflected collaborations, which are diverse and extend beyond the Computer Science department. Within PSU, Agrawal has mentored students from URMP (Undergraduate Research and Mentoring Program and LSAMP (Louis Stokes Alliance for Minority Participation) Program, and is looking forward to working with REU (Research Experiences for Undergraduates) students this summer. These collaborations showcase her commitment to fostering a diverse and inclusive research community.
Agrawal's collaborative efforts extend to Applied Linguistics as well. Through her interactions with researchers in this field, she has been able to explore the intersection of NLP and corpus linguistics, where the potential applications of NLP in linguistics are explored. One project involves a comparative study between Computer Science and Applied Linguistics using ChatGPT, analyzing how students from different fields utilize the same technology. As Agrawal's work continues to evolve, she recognizes the importance of joining forces across disciplines and building bridges between different areas of expertise. Her collaborations with researchers in social sciences, linguistics, and other AI-related fields highlight the potential of interdisciplinary research and the power of combining diverse perspectives to tackle complex problems.
She co-founded the Compassionate Computing (CoCo) Lab with Maseeh College’s Assistant Dean of Research, Antonie Jetter and Mrinalini Tankha from PSU’s Anthropology Department. The lab aims to bring together social science, linguistics, policymakers, and businesses with computer science to explore broader real-world implications and ethical considerations of NLP technologies. CoCo Lab represents a unique lens for NLP research and expands the focus beyond traditional technical metrics to consider the social impact of the technologies. Agrawal believes that while NLP researchers used to be mainly interested in performance measured through numerical accuracy, the field is now evolving to address ethical and broader societal considerations. This shift reflects the increasing recognition of the need to understand the potential implications of NLP models in the real world.
Large Language Models are very powerful tools, but also very new tools. We're still learning how to use these responsibly.
Coco Labs held a successful first Ideathon last year (see the link below for a first-hand account of participating in the event), asking students and PSU stakeholders of different disciplines to “explore the emerging landscape of AI tools and come up with creative ideas to communicate the effects of climate change in Oregon. The objective is to further foster community engagement, outreach, and education about the topic.” Prizes, rather than being hierarchical, were given for categories ranging from “Most Risk-taking” to “Best Pitch to Target Audience.” Plans for more events are in the works.
She continues to collaborate across the PSU ecosystem and beyond, organizing events that lower the barrier of entry for NLP and AI, while keeping an eye out for ethical and inclusivity concerns. This summer Agrawal will be serving as one of the Ethics chairs for an upcoming NLP conference of The North American Chapter of the Association for Computational Linguistics (NAACL). Next month, a paper she co-authored with one of her students looking into efficient conversational models will be published at a Towards Ethical and Inclusive Conversational AI (TEICAI) workshop.
As for the future of AI language models, Agrawal expressed optimism about their potential impact on education and research. However, she acknowledged that we are venturing into uncharted territory with these models, and caution is essential: “Large Language Models are very powerful tools, but also very new tools. We're still learning how to use these responsibly.” Researchers, educators, students, as well as the broader public need to adapt and continue exploring a future in which compassion and inclusion become integral aspects of technological innovation.