Learning With AI: Measure How Quickly the Skill Comes Back

A month after learning to add chapter markers to a recording, you sit down to edit another interview. You remember why the markers matter, but the sequence of steps has vanished. An AI assistant offers to walk you through it. You could take the blank as proof that the lesson was wasted. Or you could ask a more useful question: how much help does it take before you can do the job independently again?

The two questions measure different things. Your first attempt tells you what you can do without help at that moment. The work it takes to get back to a clear standard tells you something about relearning. For a creator who uses a skill only between projects, both answers matter. If the skill comes back with a short refresher, you can plan for that. If it needs a full walkthrough every time, you may need a different way to practise or a better reference at the point of use.

What the savings method actually measures

In a 2015 replication of Hermann Ebbinghaus’s forgetting experiment, psychologists Jaap Murre and Joeri Dros used a method known as savings. They compared how much work it took to learn lists of syllables with how much work it took to relearn them to the same criterion after a delay. If it took less work the second time, the difference counted as savings. This is a different question from how many of the syllables a person could immediately produce without help.

The replication used one participant who learned and relearned lists over several delays. The study helps explain the measurement; it cannot tell us how quickly a particular designer, student or business owner will recover a multi-step skill. Nor does it show that using a chatbot speeds learning. The practical idea is to borrow the distinction between independent performance and effort to recover that performance, without borrowing the paper’s curve as a personal timetable.

Use the same finish line both times

Return to the interview example. Define a pass before opening the chat: place three chapter markers at the correct points in a practice recording, give them meaningful labels and verify them in the exported file. During the first learning session, keep a rough record of active practice time and the amount of help you needed before you could pass on a fresh clip. The labels and times can differ between clips as long as the task calls for comparable work.

Suppose you originally needed a walkthrough and two practice runs. A month later, you cannot recall the menu command, but one hint is enough to finish your first practice run. That would be a useful observation. You still needed help getting started, so it would be premature to say you could do the whole job unaided. Yet the better use of practice time may be to rehearse finding that command, rather than to sit through the entire lesson again.

The comparison breaks down if you change the finish line. If the earlier session included checking the export while the later one ended when the markers looked right on screen, you measured less work alongside any possible learning. A software update, a shorter clip or a more revealing hint can change the result too. Keep a brief note about what differed. For everyday learning, you are looking for clues to guide practice, not trying to claim a precise measure of memory.

Let the AI help in steps

A full answer can be useful when you need to finish a real job. When you want to measure what comes back, ask the assistant to offer one clue at a time. Prepare a known answer or use the actual software to verify the result. If a clue sends you to the wrong menu or a feature that is absent, the extra time says more about the quality of the clue than the state of your memory.

• Try it cold. Take a comparable practice task and see which parts you can finish before opening notes or asking for help.
• Ask for the smallest useful hint. Start with the place to look or the rule to apply. Request a worked step only if you remain stuck.
• Check the output. Finish the same kind of task you used for the first session. Verify the result against the same standard, even if the answer now feels familiar.
• Record what came back and what did not. Note the sticking point, the help used and what you could do on a fresh attempt. Keep work interruptions and AI mistakes separate from practice time.

MaryChuks.com’s guide to letting a worked example fade offers a useful way to reduce that help over successive attempts. For the studio, that might mean keeping the export check beside the desk but practising the marker sequence without instructions. If the task is too infrequent to stay fluent, a well-kept reference may be the more sensible choice. For tasks that must be performed independently, keep the unaided attempt prominent and retest it on another day. Relearning can help you choose the next practice step; it cannot stand in for being ready to work without help.

The experiment can stay small. One line in a notebook is enough: date, task, where you got stuck and what let you continue. Over several projects, recurring sticking points will give you a concrete list of things to rehearse. If the list stays the same, move the practice closer to the work or improve the reference you reach for, rather than adding more pages of notes.

Source

• Jaap M. J. Murre and Joeri Dros, Replication and Analysis of Ebbinghaus’ Forgetting Curve, PLOS ONE (July 2015).