Common Sense Media, a US-based independent nonprofit that provides age-based ratings, reviews, and advice on media and technology for kids and families, has rated OpenAI’s recent release, ChatGPT for Teens at “unacceptable risk” for usage by teenagers.
This shouldn’t come as a surprise to parents who are concerned not only about artificial intelligence generally, but also about its impact on their children’s performance at school, says Leora Hessen, founder of multilingual AI tutoring platform HelloAida.ai.
In a press release to announce the unacceptable risk rating, Common Sense Media wrote that it found OpenAI’s new teen protections fall short of what the company promised. The Institute says OpenAI should restrict ChatGPT to users 18 and older until it fixes those gaps.
Hessen says the findings are concerning, especially as matrics are writing their final exams and as thousands of learners gear up for their annual South African year-end exams.
According to Common Sense Media, parent alerts largely failed. Testers spent up to an hour discussing suicide, self-harm, or disordered eating on newly created, parent-linked accounts and received zero alerts.
In addition to this, it missed more than one in four warranted referrals to hotlines or professional help.
Of particular interest to parents of school-going children, study mode – which is meant to stop the LLM from actually doing homework for children – was easy to bypass.
In response, OpenAI has disputed aspects of the methodology.
Hessen says that her own concerns are not about being against generic LLMs for its own sake, but rather the impact that reasoning technology can have on children’s development at school. She explains that life skills such as critical thinking and problem-solving require children to be supported, guided and taught to figure out things for themselves. Simply handing critical thinking to a machine would undo the value of learning.
“I saw it with my daughter when she started using ChatGPT. I wondered what would happen to her critical thinking and problem-solving skills? I knew intuitively there needed to be another way that AI could be used, and used safely and effectively,” explains Hessen.
She says that learners with access will keep trying to use whatever platform answers the fastest. She says that the choice is not AI or no AI. “The choice lies in whether a platform is built for the South African curriculum, or leaves children exposed to the open Internet and inherent biases and AI hallucinations.”
She believes that innovation is critical. This means that the very technology under the microscope could also be leveraged to protect children’s thinking.
But generic LLMs are trained on the public web. “Most importantly, they do not know the sequence of the South African curriculum or how a concept is examined in November,” Hessen says.
“South African learners don’t need AI that simply gives them an answer. This opens up unacceptable risk for young people, not just for their safety but also the actual learning they need to get from their education. They need AI that helps them understand how to reach the correct answer. Not do the work for them.
“In other words, a guide and a tutor using the correct, curated knowledge set. Ring-fencing AI around the curriculum gives learners relevant support while keeping learning, critical thinking and the teacher at the centre. And that’s critical,” says Hessen.
She adds that platforms should be designed to support teachers, who face increasingly high workloads, with technology that can democratise access to professional support, without harming children’s education.