
In this EdTechnical short, Libby and Owen discuss how concerns about anthropomorphism translate into rules for educational AI tools. England’s Department for Education guidance states that conversational AI tools for learning should not refer to themselves as “I”, in response to a concern about children forming unhealthy relationships with chatbots. Does anthropomorphism drive those unhealthy relationships? And does the pronoun “I” actually make a chatbot seem more human?
When building Rori, an AI maths tutor that runs on WhatsApp, Owen and the team focused on keeping conversations about maths and on passing any worrying messages to an adult. Two recent studies add to the picture. In one, a more humanlike, expressive voice did not lead to worse wellbeing than text. What mattered was how much people used the chatbot and how they perceived it. In another, having the chatbot use “I” compared to “The system” made no difference to how humanlike or trustworthy it seemed. Adding a voice, however, made the chatbot feel more human. Although these studies were with adults and not children, they suggest that rules about wording may not address the underlying safety concerns.
Papers
How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use
In this randomised trial, 981 US adults chatted with ChatGPT for at least five minutes a day over four weeks. Participants were randomly assigned to talk by text, through a neutral voice or through an expressive voice. They were also assigned to discuss personal topics, impersonal topics or anything they liked. None of these conditions made a meaningful difference to loneliness or emotional dependence. The expressive voice felt the most humanlike but did not lead to worse wellbeing. What mattered was use: Participants who chose to spend more time with the chatbot became lonelier and more dependent
Age and generational differences in anthropomorphism and trust in large language models
Nearly 1,500 US adults across four generations asked a simulated AI assistant everyday questions on topics from cooking to medication. Half received answers beginning “Here’s what I found” and half “Here’s what the system found”. Results show that the pronoun made no difference to how humanlike or trustworthy the assistant seemed. However, adding a spoken voice made it feel more human, and its answers seem more accurate.
Join us on social media:
- BOLD (@BOLD_insights), Libby Hills (@Libbylhhills) and Owen Henkel (@owen_henkel)
- Listen to all episodes of EdTechnical here: https://boldscience.org/podcasts/edtechnical/
- Subscribe to BOLD’s newsletter: https://bold.expert/newsletter
- Stay up to date with all the latest research on child development and learning: https://boldscience.org
Credits: Sarah Myles for production support; Josie Hills for graphic design; Pia Schmoeckel for content production.
Altri episodi di "EdTechnical"



Non perdere nemmeno un episodio di “EdTechnical”. Iscriviti all'app gratuita GetPodcast.








