BLACKSBURG, Va. (WDBJ) – Ask artificial intelligence anything and it will likely give you an answer.
“These are amazing technologies, right?” Virginia Tech Assistant Professor of Computer Science Bimal Viswanath said. “A lot of useful applications, so our goal is to enable these benign applications while making it safe, popular for people to use it.”
Along with the good in AI, there can be some negatives, such as toxic language.
“If someone is giving other people a hard time, speaking in a way that’s offensive, causing harm, emotional harm, and so on, a model that’s trained on the data can pick that up and can be racist, can be extremely toxic and pretty much anything you can imagine that you’ve seen online,” Viswanath said.
Researchers at Virginia Tech have received a grant to help measure and mitigate harmful language from the bots.
“An attacker can actually target these chatbots specifically to make them produce toxic content,” Virginia Tech PHD Student Aravind Cheruvu said. “I think that is an important aspect to look into, but as far as the chatbots are concerned, ultimately, they just learn from the data which is being given to it, and how they’re actually learning it.”
The researchers are studying how prevalent toxic language is in AI and how to reduce it going forward.
“Our future aspects of this particular research is to investigate, can we actually mitigate the specific categories of the toxicity that is either there is a hate speech or there is a propaganda,” Cheruvu said.
The grant is worth $600,000 and covers research over the next three years.
Copyright 2023 WDBJ. All rights reserved.
