According to a research team in the Netherlands, sarcasm is being trained into AI systems. Matt Coler, a researcher at the University of Groningen’s speech technology lab, mentioned to The Guardian that they have been successful in reliably recognizing sarcasm and are looking to further develop this ability.
Coler highlighted the importance of non-verbal cues like delivery and facial expressions in detecting sarcasm. These cues were found to be crucial for the AI to differentiate between sarcasm and sincerity.
Meet Mustard: The Sarcasm AI
Researchers from the US and Singapore compiled a database (called Mustard) filled with video clips showcasing instances of sarcasm from popular TV shows. They trained the neural network using this database, leading to the AI being able to detect sarcasm 75% of the time.
Xiyuan Gao, a PhD student, explained that incorporating more visual cues like smirks and eyebrow raises could enhance the accuracy of the AI. Gao acknowledged that achieving 100% accuracy is a challenge even for humans.
Although not perfect, this development is a step forward and certainly less scary than AI models deliberately rejecting instructions.
Featured Image: CBS
Leave a Comment