Each fall, the Prix Ars Electronica festival transforms the cityscape of Linz, Austria, into a spectacle of electronic and digital art. In September 2024, the festival drew over 100,000 visitors to the city and artificial intelligence (AI) was a big draw for this year鈥檚 registrants.
The festival, known as one of the world's largest media events, buzzed with AI art installations and high-profile tech conferences. European Union regulators were particularly prominent, wrestling with the complex implications of rapidly evolving artificial intelligence technologies.
One attendee was the winner of an Honorary Mention in the category of Interactive Art +, the iSchool鈥檚 Assistant Professor Dong Whi Yoo. He was there, along with colleagues Younah Kang from Yonsei University and Soonho Kwon from the Georgia Institute of Technology, where Professor Yoo had earned his doctorate.
Their piece 鈥,鈥 was born out of Professor Yoo鈥檚 research on the intersection of human-computer interaction in the healthcare field. Predictive modeling of treatment shares some features with AI, and Yoo鈥檚 research had developed a model that examined social media posts for signs of mental health issues. That work highlighted a critical challenge for AI: the 鈥渆xplainability problem.鈥 As Yoo puts it, explaining these predictive models sometimes feels like modern-day fortunetelling.
The AI model he was working on used complex and intricate linguistic markers 鈥 examining post length, complexity and even specific vocabulary patterns such as frequency of profanity 鈥 in order to predict the likelihood of a coming mental health crisis for the user. Because of his work鈥檚 nature, Yoo is concerned with some terminology surrounding AI, such as the term "hallucinations" to describe errors generated by AI, due to potential stigmatization of mental health experiences.
Judges at the festival recognized the value in sparking conversations that 鈥淎I Fortuneteller鈥 raised. They discussed its exploration of issues surrounding human autonomy, trust and motivation.
The festival also underscored a key limitation in AI development: the coming affordability crisis. Currently, Large Language Models (LLMS, what are popularly referred to as AI) require so much computer processing power that the costs in energy consumption and in environmental impact are immense. This could lead to a future scenario in which deep-pocketed tech companies like Google or Microsoft dominate the field of AI to the exclusion of others.
Another issue Yoo recognizes with AI/LLMs is excessive information input leading to errors in output. Scanning the entirety of Facebook, for instance, leads to misinformation, opinion, sarcastic posts and satire being absorbed into a language model incapable of making the complex linguistic distinctions between types of information input. One possible solution Yoo proposed to both problems would involve small-to-medium language models where the AI is tasked with more limited input to provide more focused results tailored to certain fields.
In that vein, the National Institute of Health have made significant investments in medical language models, representing one promising pathway while demonstrating how closed, carefully curated datasets might help overcome current technological barriers and lead to better predictive modeling outcomes.
As AI continues to evolve, Professor Yoo鈥檚 work and research is right there at the forefront 鈥揵ridging technology, healthcare, art and critical social analysis.