Careless Whisper: Speech-to-text Hallucination Harms | Montreal AI Ethics Institute

Montreal AI Ethics Institute

Factuality

Ownership

Careless Whisper: Speech-to-text Hallucination Harms | Montreal AI Ethics Institute

🔬 Research Summary by Allison Koenecke, an Assistant Professor of Information Science at Cornell University. Her research focuses on algorithmic fairness in online services. Overview: OpenAI’s speech-to-text service, Whisper, hallucinates entire sentences in addition to producing otherwise accurate speech transcriptions. These hallucinations induce concrete harms, including (a) perpetuating violence, (b) claiming inaccurate associations, and (c…