Natural Sciences: Discovery-first inquiry

What you'll be able to do: Be led to a scientific idea by cases and questions until you produce it yourself, which is the only reliable way to displace an intuition you already hold.

Why telling doesn't work for the core ideas

Article 01 established the problem: science misconceptions are occupied positions, not gaps. Here's the consequence.

When you're told the correct model, the wrong one isn't removed. It was never brought out, never tested, never seen to fail. So you end up holding both, the formal answer you can state, and the intuition you actually reason from. Under any unfamiliar question, the intuition wins, because it's the one that feels like understanding.

Displacing it requires the wrong model to be activated and then to fail in front of you. That's what discovery-first does: it gives you cases, asks what you think happens, and chooses the next case based on what your answer revealed.

The caveat that has to travel with this: unguided discovery learning has a poor research record and is generally worse than direct instruction. What works is the guided version, carefully sequenced cases with feedback, aimed at a specific idea. Handing someone equipment and hoping is not this method.

Where it's worth the time

It's slow. Reserve it for the two or three ideas per module where a strong wrong intuition exists:

• Natural selection: the teleological intuition is the default human explanation and has to be actively displaced.
• Newton's first and third laws: the impetus intuition is nearly universal and survives instruction.
• Equilibrium as dynamic: "the reaction has stopped" is the intuitive reading.
• Entropy as counting: "disorder" is the taught analogy and misleads badly.
• Genetic drift: everything that happens in evolution feels like it must be selection.
• Deep time: no amount of stating the number conveys it.
• Why orbits are falling: the intuition is that orbit is a balance of forces.
• Photosynthesis and plant mass: "from the soil" is the default answer for most adults.

Don't use it for nomenclature, conventions, or anything with no underlying logic to find. There's nothing to discover in the naming of alkanes.

A worked sequence

Natural selection, with a learner who holds the standard intuition:

> On an island, some beetles are green and some brown. The ground is brown. > Birds eat beetles. What happens over many generations? >: "Mostly brown." > > Do the individual green beetles turn brown to survive? >: "No, they just die." > > So what changes over time, the beetles, or the population? >: "..the population. The individuals don't change at all." > > Second island, grey rock, all the beetles are brown. What happens? >: "They get eaten.. but there aren't any grey ones. So nothing improves?" > > What would need to exist for the population to change? >: "Some variation. If they're all the same, selection has nothing to work on."

Two conditions derived, neither stated. Note that the second one came from a case the first idea couldn't handle: counterexample hunting operating inside an inquiry.

The prompt

I want to work out [TARGET IDEA] for myself. Don't tell me it and don't define anything. Give me one concrete situation at a time and ask what I think happens.
Choose each next situation so that my last answer reveals what I still haven't got. If I state the idea, tell me whether it's complete, don't let me stop at a version that's missing a condition.

That final clause matters: the beetle learner's first answer was right and incomplete, and stopping there would have left them with half the mechanism.

Pitfalls

1. Unguided discovery. Sequenced cases with feedback, or don't bother.
2. Reading instead of answering. The generation is the mechanism.
3. Using it for everything. Slow. Two or three ideas per module.
4. The AI confirming too early. It will want to explain. Re-issue the rule.
5. Stopping at the first correct-sounding statement. Ask whether it's complete.
6. The tell: you were led somewhere and can't say what evidence took you there. Then you were told, slowly.

Try this today

Pick the one idea in your current topic that students reliably get wrong.

Ask to be led to it with cases only, no definitions, committing to an answer each time. Then read the textbook version. The interesting part is usually a condition you didn't reach.