AI is disrupting sectors as diverse as healthcare and marketing. Now it is the turn of education. Governments sense an opportunity to save costs and improve outcomes for students. Schools are exploring tools that can adapt exercises to individual students while parents spot a chance to improve their children’s education without the expense of sending them to private tutors. But as with anything AI-shaped, there are rewards as well as risks.

Over-dependence and cognitive laziness are often touted as parental concerns. But there may also be security and privacy issues, especially if you’re using a glorified chatbot. Parents should understand the limits of machine intelligence.

The AI tutor market is growing fast

The market for AI tutoring tools is booming. According to one estimate, it’s set to grow from around $2bn to over $17.7bn between 2025 and 2033. The British government has already invited edtech and AI companies to design tools for the classroom. It claims that as many as 450,000 disadvantage pupils could benefit, saying AI could help families that are unable to afford private tutors. In the US, the picture varies from district to district but the direction of travel is broadly similar, with AI and tech firms introducing specialized tools for students and teachers.

The market is maturing fast. Today, you can find various types of AI tutor tools to buy and use. Indeed, the term “AI tutor” is applied to a remarkably broad range of products. Some are little more than general-purpose chatbots with an educational label while others are structured learning systems built around a curriculum.

At the bottom end of the market are simple chatbots and AI assistants. They might be able to summarize and answer general questions, but they don’t necessarily offer much value beyond that. In fact, they may end up answering questions on behalf of their users. A more sophisticated option would be a chatbot purpose-built for teaching. These are programmed not to answer questions directly but instead offer “Socratic-based” tutoring.

More advanced still are Intelligent Tutoring Systems (ITS) that offer a pre-determined curriculum designed by humans. They “scaffold” learning to provide hints when students aren’t sure of an answer, and make it virtually impossible to surface inappropriate content.

Do AI tutors even work?

The jury is still out on whether any of this works. The evidence is encouraging in places, but it doesn’t seem to support broad claims that AI tutors will improve learning in every setting. Sometimes, the technology may be making things worse. In one recent study in the UK, two-thirds of teachers said they’d already observed a decline in writing and problem-solving skills among students due to AI use. A separate study identifies “a significant negative correlation between frequent AI tool usage and critical thinking abilities.” It notes that AI use for tutoring children “may inadvertently diminish users’ engagement in deep, reflective thinking processes.”

This is one of the biggest developmental risks associated with encouraging children to use AI tools. If the AI is too quick to provide an answer, or at least makes the process too easy for the learner, they could develop a cognitive laziness which is harmful in the long term.

There are other potential psycho-social risks. Children may end up placing too much trust in their confident-sounding AI tutors, and fall for hallucinations more readily than adults. They may even come to rely on them as quasi-friends, which could stunt emotional bonds with their peer groups. AI is also a poor teacher in many respects. It might know the “answer” to a question, but lack the ability to manage a child who is frustrated with the learning experience. The empathy and soft skills of a human teacher are very difficult if not impossible for a machine to imitate.

One pertinent question is therefore how the tool makes the child engage. Does it ask them to show their reasoning, attempt a problem and respond to feedback? Or does it make completing the homework faster while quietly doing most of the thinking?

Security and privacy risks

There’s more. For users of ungated chatbots and similar tools, there’s a risk that conversations may take a detour into age-inappropriate chat. An AI tutor may also collect far more than a child’s answers. Depending on the product, it could retain prompts, schoolwork, performance data, interests, account information, device identifiers and voice recordings. Anything shared with the model might be used for training purposes and possibly leaked to other users. And it might also be shared with advertisers and other third parties.

This doesn’t just represent a privacy risk, it could turn into a full-blown security issue if the model provider or one of those trusted third parties that have your child’s data is breached. If it’s a voice-activated tool, the provider will have biometric data from your child which would be highly valuable if hackers were able to link it to their identity.

This results in several questions for parents, including:

  • Is the child old enough to use the service under its terms?
  • What information is collected, why, and how long?
  • Are conversations used to train or improve AI models?
  • Can a parent access and delete the child’s information?
  • Which service providers or other entities receive the data?
  • Does the product show advertising or use data for profiling?

Finally, there’s the risk of prompt injection. This happens when AI tools are tricked into ignoring their safety guardrails and display untrusted content. Your child may be savvy enough to do this, enabling them to access age-inappropriate material. Or a hacker may hide instructions in legitimate web pages, so that when the AI interacts with that page, it takes them to a phishing site or performs some other malicious action.

Some precautions for parents

There’s plenty for parents to think about. If you’re keen to get your kids in front of an AI tool to help with private tutoring, first understand the difference between a simple co-pilot and a dedicated ITS. You want tools that will guide your child’s learning journey rather than provide simple answers.

Also consider:

  • Tools which are compliant with data protection regulations (e.g., COPPA, CCPA, GDPR), meaning they won’t share your child’s data with the underlying AI engine.
  • Opting for a tool approved by your child’s school or supported by educational evidence.
  • Reading the privacy information, especially the parts dealing with model training, data retention, advertising and third-party providers.
  • Using the tool with your child. As good as AI is, your child will probably benefit more from learning alongside a human.
  • Teaching your kids never to share personal information with the AI, and how to spot hallucinations. This will stand them in good stead when using AI later in life.
  • Talking to your kids to find out how they feel about the AI, and ensuring that they understand it doesn’t have the power to feel or emote.
  • Explaining that the tutor can sound certain even when being wrong. Important answers should be verified with teachers or against textbooks or other reliable sources.
  • Toggling on any age-appropriate settings, if available.
  • Limiting “AI time” per day, just as you would screen time.
  • Checking your child’s prompt history to ensure they’re using the tool correctly and safely.

The technology is maturing at a rapid pace, with new offerings coming to market all the time. So if you’re keen to try AI tutors, be sure to stay updated on what’s available out there. And remember it may not be a substitute for good old-fashioned offline learning.