AI in Education: Are We Asking Students to Drink from a Firehose?

There is little doubt that artificial intelligence can be useful in education.

AI can explain difficult concepts, generate practice questions, provide immediate feedback and adapt learning materials to suit individual students. In EdTech, this creates opportunities that would have been difficult—or prohibitively expensive—to offer at scale in the past.

However, there is another risk that deserves more attention: information overload.

Much of the discussion surrounding AI focuses on accuracy. Does the AI hallucinate? Are its explanations correct? Is the feedback reliable?

These are important questions. But even when an AI system produces completely accurate information, it can still generate far too much of it.

A student might submit a short answer and immediately receive several paragraphs of corrections, explanations, examples and suggestions.

A single composition written by my daughter can generate a few pages of corrections, with one line for each grammar or vocabulary error.

 Every individual point may be valid. Yet can the student realistically understand, remember and apply all of them at once? Probably not. Actually, definitely not. Because even I can’t do that myself. How can I expect my 11 year old to do it.

Human beings are not computers or AI. We cannot simply download a large volume of information and immediately incorporate it into our thinking. When students are presented with too much feedback, their attention may begin to fade. Eventually, they may skim the response, become discouraged or simply stop absorbing the information. I know I would. And if you think about it honestly, you probably would too.

There used to be a common expression in management consulting, when I was in Deloitte: “drinking from the firehose.” It described the experience of having to absorb an enormous amount of information from a client within a very short period.

The imagery is appropriate. A firehose fires a stream of water at high pressure, and the consultant tries to absorb as much of it as possible. His or her job depends on it after all.

If absorbing so much information is difficult for an experienced adult consultant, imagine what it is like for a pre-teen or teenage student.

This principle guides the the different grader tools developed by Articulate Intelligence. Rather than presenting students with an unstructured wall of corrections, the tools organise their feedback into clear areas. For example, with Composition graders, feedback is organized into content development, language use, organisation and task fulfilment.

For Oral Exam reports, feedback is broken down into PEEL sections to give improvement points for each part.

Students can see what they have done well, understand their most important areas for improvement and receive specific suggestions they can apply to their next piece of writing. For situational writing, this also includes whether they have addressed the required content points, used an appropriate tone and fulfilled the purpose of the task.

The aim is not simply to identify every possible weakness in a student’s work. It is to turn AI-generated analysis into focused, actionable guidance—helping students improve one manageable step at a time.

Feedback should also be appropriate to the student’s age, ability and emotional readiness. A struggling learner may need one clear correction and some encouragement. A stronger learner may be ready for more detailed analysis. Giving both students the same exhaustive response simply because the AI is capable of producing it is not personalisation.

True personalisation is not only about deciding what a student needs to learn. It is also about deciding how much the student can reasonably absorb at that moment.

AI gives us the ability to generate almost unlimited content, feedback and data. But unlimited output is not the same as unlimited learning.

The real challenge for AI in education is therefore not simply to produce more information. It is to moderate, prioritise and pace that information so that students can act on it.

Good teaching has always involved judgment: knowing what to explain, what to leave for later and when a learner has had enough.

AI in education needs that same restraint.

Learning happens a little at a time—not through a firehose.