<\/p>\n
Dr. Anuji Gupta gives a keynote presentation on ‘AI-Slop’ at Pegasus Ballroom on Wednesday, opening students up to a deeper conversation about the ethical usage of AI.<\/p>\n
Allison Smith<\/p>\n
Dr. Anuj Gupta, professor of rhetoric and composition\u00a0at the University of South Florida, presented a lecture on “AI slop” during the Knights Write Showcase at the Pegasus Grand Ballroom on Wednesday.<\/p>\n
Gupta earned his Ph.D. in rhetoric and composition from the University of Arizona<\/a>, his work focusing on generative artificial intelligence.\u00a0Gupta said “AI slop” \u2014 low-quality AI-generated content \u2014 has become common on social media, from\u00a0harmless posts like entertainment to harmful content such as\u00a0\u201cdeep fakes,\u201d\u00a0which are designed to spread misinformation.\u00a0<\/p>\n \u201cBoth in quantity and quality, AI slop is AI content that is misaligned with author and audience expectations,\u201d Gupta said.\u00a0 \u201cA\u00a0lot of these AI models have been trained on content from the internet,\u201d Gupta said. “Those of us who have been on the internet\u00a0know\u00a0the\u00a0internet is not always\u00a0high-quality\u00a0content.\u00a0So,\u00a0it’s\u00a0basically trained\u00a0on a lot of garbage.\u201d\u00a0<\/p>\n Language Learning Models seek averages in human culture to create the most typical responses, creating issues in generating content, Gupta said.\u00a0<\/p>\n \u201cThat’s why all the content sounds sloppy or sounds mediocre,” Gupta said. “It\u2019s\u00a0because\u00a0they’re\u00a0really going for whatever is the\u00a0average of\u00a0human culture.\u00a0What we can do to combat that is to include AI only in\u00a0limited aspects of our writing workflows.\u201d\u00a0\u00a0<\/p>\n Apps centered on AI, Gupta said, are\u00a0seemingly launching\u00a0every day.\u00a0Keeping up with a new app every day can become exhausting and overwhelming, Gupta said.\u00a0<\/p>\n “I give it at least six\u00a0months to see whether\u00a0it’s\u00a0sticking or whether\u00a0it’s\u00a0fizzling out,”\u00a0Gupta said.\u00a0<\/p>\n Gupta said\u00a0AI at its most useful should support work with what the user already knows how to do, but not automatically do it.\u00a0Gupta said\u00a0after generating a response,\u00a0a\u00a0person\u00a0should\u00a0review\u00a0it and\u00a0repeat the process if need be.\u00a0<\/p>\n Students who use AI\u00a0should\u00a0revise\u00a0the results\u00a0before deciding whether to include any of the\u00a0materials, Gupta said.\u00a0<\/p>\n According to a study at the National Library of Medicine<\/a>, AI systems\u00a0regularly produced incorrect citations and fabricated sources, a problem commonly referred to as \u201challucination.\u201d\u00a0<\/p>\n Pedagogy and\u00a0policies made for AI\u00a0help prevent AI slop from entering\u00a0professional work.\u00a0Creating clear guidelines can help prevent the spread of AI slop,\u00a0Gupta said.<\/p>\n Adriana Carpenter, freshman\u00a0pre-management\u00a0major, attended the presentation with the goal\u00a0of learning\u00a0more about how\u00a0AI is currently being used.\u00a0<\/p>\n \u201cI\u00a0feel\u00a0like\u00a0AI is very relative to the society we live in,\u201d Carpenter said.\u00a0\u00a0<\/p>\n When it comes to her personal use of AI, Carpenter said she normally uses it to brainstorm.\u00a0<\/p>\n \u201cI bounce my ideas off people, and if people aren\u2019t available for me to bounce my ideas off of, I bounce it off of certain AI,” Carpenter said.\u00a0 UCF Faculty Center Director\u00a0Dr. Kevin Yee is an expert in\u00a0<\/a>educational development and AI<\/a> technologies. He said that students who pass the burden of their school work entirely to AI are participating in something called “cognitive offloading.”<\/p>\n He said cognitive offloading in an educational setting\u00a0causes students to miss the point of being at\u00a0a university. When a student allows the AI to do the thinking for them, Yee said, they don’t give themselves the opportunity to learn.\u00a0<\/p>\n
\u00a0
Gupta said\u00a0there are seven levels to AI misalignment in his presentation: training data, algorithms, infrastructure, apps, writing workflows and AI prompts, pedagogy and policies.\u00a0Preventing AI slop, Gupta said, starts with how the AI is trained, and the low-quality output can be prevented at any of these stages.\u00a0\u00a0<\/p>\n
\u00a0
As someone who just left high school, Carpenter said she saw a lot of students copy and paste AI-generated answers into their assignments.\u00a0She said that when other students used it to think for them, it became a crutch and they\u00a0couldn\u2019t\u00a0think without it.\u00a0<\/p>\n