We need teachers who tell us what not to learn

Where a learning plan gets too large

Consider a software founder who wants to help small manufacturers spot defects in their products using photographs. They can already build an application, but they know little about manufacturing or computer vision. As they research the idea, they find material about cameras, lighting, machine learning, factory operations, and quality control. Each subject is relevant. Taken together, the recommendations could become a substantial course of study before the founder has tested the idea with anyone.

The immediate question is more specific. Can photographs show the defects that a particular manufacturer cares about, and could a tool help the people inspecting them? The founder needs enough understanding to investigate that question properly. They could begin by looking at examples with an experienced inspector, learning how defects are identified, and finding out how photographs would be taken during production. That work could reveal which technical knowledge they need next.

Suppose some defects are difficult to distinguish because the lighting changes between photographs. Learning about image capture may then deserve attention before experimenting with more complicated models. Or the photographs may be clear, while the definition of an acceptable product depends on details the founder does not understand. In that case, more discussion with the manufacturer could be more useful than another machine-learning tutorial.

The founder may eventually need considerable depth across several subjects. The question is how to build that depth in an order that helps the company progress. Preparing for a small supervised trial and preparing a system that customers will depend on are different stages of the work. Good guidance should make the requirements of each stage clear.

There is also a cost to copying how an established company operates before reaching that stage. In Do Things That Don’t Scale, Paul Graham describes founders recruiting users and handling work manually before developing more scalable methods. Applied to learning, that suggests a useful question: are you studying something because it will help you now, or because a much larger version of your company might eventually need it? Paul Graham

Teaching includes deciding the order

A topic can be important without needing to come first. Sometimes people need an introduction that lets them try the work, followed by deeper study once they have something concrete to connect it to. Other concepts need to be understood before an attempt will make much sense. Choosing between those approaches is part of teaching.

Fast.ai’s Practical Deep Learning for Coders provides one example. Its stated starting requirements include coding experience and high-school mathematics. The course introduces a working model early and teaches the calculus and linear algebra it needs along the way. Students still study foundations, but prior university-level mathematics is not a condition for beginning the course. Practical Deep Learning for Coders

That choice has a defined audience and purpose. It gives people who can already program a way to begin applying deep learning while developing more understanding. A person preparing to research a new mathematical method would need a different plan. The useful lesson is that prerequisites should be tied to what someone is being prepared to do.

Google’s Machine Learning Crash Course makes some of these boundaries explicit too. It names the mathematical concepts and programming experience learners should bring, identifies calculus as optional for advanced topics, and explains that the course does not teach specific machine-learning APIs in depth. A learner can see both what is expected and what the course is designed to leave out. Google for Developers

Existing experience changes those decisions further. Research reviewed by Slava Kalyuga and colleagues found that instructional methods effective for beginners can lose their usefulness, and sometimes have negative effects, for more experienced learners. This finding concerns how people are taught; it does not give us a universal list of topics to skip. It does support the need to reconsider guidance as a person’s knowledge grows. Taylor & Francis Online

For someone crossing into a new subject, that means looking carefully at what they bring with them. An experienced programmer may need an introduction to experimental design while having little use for another explanation of loops and functions. Treating them as a beginner in everything adds work without necessarily addressing the part they are missing.

Leaving something out needs a reason

There is a risk in promising to remove unnecessary learning: a beginner may not know which missing ideas are important. They can follow instructions successfully and still misunderstand why the result works. The difficult judgment is deciding which gaps can remain open and which would prevent them from doing the work responsibly.

The scientific-computing guide is useful here as well. Although its authors leave out some advanced practices, they retain requirements such as preserving raw data, recording processing steps, and testing software before relying on it. They reduce the initial burden while keeping practices that help a researcher understand and reproduce the analysis. PLOS

A recommendation to skip something should therefore have a specific reason. Perhaps the person has already demonstrated the skill. Perhaps the topic belongs to a different kind of project. Perhaps an existing tool handles the task, and the learner needs to understand how to use and check that tool. Those are different situations, and each one changes what should happen next.

Suppose a researcher uses AI to write a script that produces a chart. They may have little reason to memorise every command used to format it. They still need to understand which observations went into the chart and whether the comparison answers their question. If they cannot explain those choices, the finished chart is not enough evidence that they are ready to continue independently. The learning need becomes clearer when someone examines the work with them.

Some foundations will need sustained study. If a missing concept repeatedly causes mistakes, a sequence of quick fixes may take more time than learning it properly. A useful teacher should be willing to recommend that effort and explain what it will make possible. Cutting unnecessary learning should leave more room for this kind of depth.

Nor should every interest have to justify itself against an immediate project. Someone may choose to study mathematics because they want a deep understanding of it, or explore another subject before deciding what to build. That is a valid goal. The waste appears when a person is led through material they believe is required for a different outcome, without anyone checking that assumption.

Make postponement specific

“Come back to this later” becomes more useful when later has a meaning. For the founder working on visual inspection, detailed knowledge of camera hardware might wait while they test photographs taken with existing equipment. If the project reaches a point where those cameras cannot capture the necessary detail, that decision deserves another look. The reason for postponing the topic also tells them when to return.

The same approach can guide how much they learn about a tool. Before a small trial, they may need to understand its inputs, common failures, and how to check its output. Before customers rely on it, they may need a deeper investigation of reliability and operating limits. A teacher who remembers those conditions can change the advice when the responsibility changes.

This also gives AI assistance a clear place in the plan. If a tool can handle part of a task reliably, there may be less reason to spend time practising that part by hand. The remaining learning should cover whatever the person needs to make effective use of the result. A change in the tool, the task, or the required level of independence can change that decision.

Before committing to another large block of study, it helps to connect it to a piece of work. What will you be able to attempt after learning it? What currently prevents you from making that attempt? A small exercise, a review of previous work, or a conversation with someone experienced may provide enough information to answer. Where the answer remains uncertain, the plan should preserve that uncertainty instead of turning every possible gap into a mandatory course.

Over a long project, some topics will disappear from the plan, others will return, and a few will require much more attention than expected. Keeping track of the reasons makes these changes understandable. It also gives the learner a way to question the advice rather than simply following another list.

This is one of the responsibilities we want Easable to take on. Guidance should account for what a person is trying to do, what they can already demonstrate, and where tools or other people can help. When we recommend leaving something out, we should be able to explain the decision and recognise when it needs to change. The value of a shorter learning path depends on whether the person can still do the thing they came to learn.