ITB Expert Warns AI Can Still Err, Should Not Be Primary Reference

An artificial intelligence (AI) expert from the Bandung Institute of Technology (ITB), Ayu Purwarianti, has warned that the ability of AI to generate text, images, and videos resembling human work will become increasingly difficult to distinguish in the future. She urged the public not to use AI as a primary reference and to remain critical of every piece of information it produces.

Ayu stated that in the future, it will be harder to tell the difference between AI-generated and human-made output with the naked eye. “But for now, if we frequently use an AI tool, we usually become more familiar with the characteristics of that AI tool’s output,” Ayu told Republika on Tuesday.

Ayu explained that the workings of AI Large Language Models (LLMs) like ChatGPT are referred to as next token prediction, where each word is generated based on a probability value. Because of this, some call it a stochastic parrot, akin to a parrot that can answer because it has been trained. “Like a parrot taught to say ‘waalaikumsalam’ when someone says ‘assalamualaikum’, even though the parrot does not understand that the utterance is a blessed greeting,” Ayu said.