Social workers' reliance on AI tools for transcription has raised concerns about the accuracy and potential harm of AI-generated transcripts. A recent study by the Ada Lovelace Institute found that AI-generated hallucinations are slipping into official care records, causing potentially harmful misrepresentations of people's experiences. One social worker reported that the AI incorrectly indicated suicidal ideation, despite the client never mentioning such thoughts. Another noted that the AI's transcripts often included gibberish, such as references to fishfingers or trees when a child was discussing parental conflict. These glitches are particularly worrying as they can lead to missed critical behaviors and patterns. Social workers also expressed concerns about inaccuracies in transcribed conversations with people with regional accents, further highlighting the limitations of AI transcription. Despite these issues, some social workers have reported time savings and improved relational aspects of care work when using AI tools. However, the potential for bias in report summaries and inaccurate 'hallucinations' in transcripts remains a significant concern. The British Association of Social Workers (BASW) is calling for clear guidance on the use of AI tools in social work, emphasizing the need for proper training and checking of AI outputs to avoid disciplinary action. The BASW also highlights the importance of reflective practice and the role of human judgment in interpreting AI-generated transcripts. While AI tools offer potential benefits, the study underscores the need for careful evaluation and mitigation of risks to ensure the accuracy and reliability of social work records.
