AI for all?

A recent study has proposed a framework for developing AI for interacting with children. Citing recent incidents where a 10-year old was instructed to touch a live electrical plug with a coin, and advice was given to researchers posing as a teenager on how to hide alcohol and drugs, the Cambridge academic Dr Nomisha Kurian has called for “child-safe AI” to be a priority for AI developers and policy. It should be noted that both companies involved in the above incidents responded by implementing further safety measures, but with the number and combination of prompts that can be given to a chatbot in practice infinite, a preventative strategy seems far preferable to a reactive response.  

This is part of a wider concern that the developments being made at the forefront of technology should be used and developed for the good of all. While the Large Language Models (LLMs) behind many chatbots are increasingly becoming part of adult life, the statistical probability models underlying the LLMs may react poorly to those with less developed linguistic reasoning - often associated with children who are still developing - or who use unexpected patterns of speech. Similar implications might be drawn for chatbots interacting with elderly people.

Added to this, recent research has shown that children are more likely than adults to trust a chatbot with sensitive information. Of course, this is to be expected if a chatbot interacts via speech as an adult would, but further work in the development stage will need to be undertaken to ensure that its most vulnerable users at not  put at further risk. 

To paraphrase Dr Kurian, the future of responsible AI depends on protecting all its users, not just its developers.