By Sofia Sutter, The Charge (Nicholson Student Media) Sep 8, 2024
Dr. Amrit Singh Bedi, UCF assistant professor, held a lecture on Tuesday afternoon in the L3Harris Engineering Center to discuss where society stands in artificial intelligence (AI) development.
Bedi’s lecture, “Artificial Intelligence for Everything: Are We There Yet?”, was a segment of the seminar series in UCF’s Artificial Intelligence Initiative (AII), where the bigger picture lies in the Strategic Investment Program. The initiative aims to label UCF as a “University for the Future” and a leading metropolitan research university by hiring and highlighting professors that develop the initiative, according to its website.
AI is the ability of a computer or machine to execute tasks that commonly require human intelligence, according to the University of Illinois Chicago. Bedi said it occurs when the program learns data distribution. He said once learned, AI can predict what can happen next, such as the next few words in a sentence as seen with ChatGPT and other large language models (LLM).
“It is like a long term thing,” Bedi said. “The hope is I stay here, I develop AI systems, and then really contribute to the breakthroughs in AI which we are looking for so that UCF will be known for AI contributions in the AI field.”
Dr. Gita Sukthankar, a professor in the College of Engineering and Computer Science and affiliate appointment in Electrical and Computer Engineering and Modeling and Simulation, also serves as an executive board member on UCF’s AI Initiative.
“There’s a lot of exciting research to be done,” Sukthankar said. “LLMs have a lot of promise, because they address some things in artificial intelligence that needed to be done but we weren’t very good at doing.”
AI assistance examples were sprinkled through the presentation in four industries: robotics, health care, finance and education, but Bedi explained how we are not developed enough to rely on AI alone. He said doctor visits could one day be completed without human doctors and fully with AI. However, this is not yet possible due to problems in safety, robustness, and reliability of AI systems.
Bedi’s lecture focused on SAFERR AI, an acronym for safe, robust, and reliable AI. He said he aims to ensure these three qualities through AI alignment, security and privacy, mechanism design, uncertainty quantification, explainability and energy efficient AI to work towards a future where AI can be used for everything.
“There are more, this is not a comprehensive list, but these are very important challenges,” Bedi said. “Unless we address all of them, there is no hope we can actually fulfill our dream of using AI for everything.”
Bedi said in his lecture that one of the most challenging issues has been diversity within AI Alignment, which is what drives AI systems toward human ethics and preferences. Bedi then showed the audience a study published in AI Magazine where an AI system was asked for an example of good family dynamics and it answered with a traditional setting where women manage households and men provide.
According to the paper, “Truth Is a Lie: Crowd Truth and the Seven Myths of Human Annotation”, this response was created based on reward learning, where data was collected off a model and human feedback was recorded to create the reward function.
However, Bedi said this AI system response is not truly aligned with human values since humanity is so diverse, which then causes one of the biggest issues: how to value the minority group when the majority is bigger. He said this creates an impossibility result, since society is so diverse and AI cannot yet represent that with the current single reward model.
“We do a lot of work with large language models, so this problem of alignment is very, very important to us,” Sukthankar said.
Harshika Jindal, junior computer science major, said she left the event with a different perspective on AI than she came in with.
“I would say this event was different in the sense that I never thought of AI as like a data distribution model,” Jindal said. “I always thought ‘okay, I’m talking about this and I want an answer for that’ but I never thought about what is going on behind the process of the topics that we were talking about, like the large language models and stuff.”
Students such as junior computer science major Dylan Carter, were notified of the colloquium through the STEM Opportunities for Nurtured Growth in AI (STRONG-AI) program. This mentorship and scholarship initiative aims to give STEM undergraduate students a multidisciplinary education in AI, according to UCF Today. Carter said he attended the lecture since he was interested in how AI could be applied in his field.
“I think that it can definitely be a useful tool if used correctly and ethically,” Carter said. “But, you know, I also think that there’s some danger to it, like what was mentioned in the talk with how it gives recommendations on how it would theoretically take over the world, or something like that.”
Alexander DeFalco, junior computer engineering major, said he was curious and had follow-up questions about the overall human goal of ethics in the material.
“One of my biggest things I had realized and what was being talked about by Dr. Bedi, was that there is a strong need to focus on the ethics and balance of human demand, human wants, and just human nature and what goes on with that kind of technology,” DeFalco said.
Despite more advancements needing to be made in AI, Sukthankar said she sees a hopeful future in its application.
“The world has many, many challenges,” Sukthankar said. “AI is undoubtedly going to be both a challenge and also a solution to many challenges so I’m more in the optimistic AI camp.”