Rule induction · Learning
Rule Learning Test
One hidden rule splits eight objects in two. Sort them, and the feedback tells you when you have it.
No IQ score. No cognitive age. No percentile we can’t source.
Rule induction · Learning
Rule Learning
One hidden rule splits these eight objects in two. Sort them, and the feedback will tell you when you have it.
Your best · —
Trains working out a hidden rule from examples and feedback, in this task.
Sign in and Train keeps your records. Go to Train
Keep going
All games- Chronoscope · FocusReaction TimeHow fast are your reflexes?
- Sequencer · MemorySequence MemoryRepeat the growing pattern
- Plate · MemoryVisual MemoryFind the tiles that flashed
- Lexicon · MemoryVerbal MemoryHave you seen this word before?
- Reticle · FocusAim TrainerHit the targets, fast
- Chromatograph · InhibitionStroopName the colour, not the wordCore
- Delay Line · MemoryN-BackHold the stream in mindCore
- Gate · ControlGo / No-GoTap — but hold when toldCore
- Router · FlexibilityTask SwitchChange the rule on the flyCore
- Goniometer · SpatialMental RotationSame shape, or mirrored?Core
- Register · MemoryNumber SpanHow long a number can you hold?Core
- Collimator · FocusFlankerAnswer the middle arrow, ignore its neighboursCore
- Comparator · MemoryChange DetectionOne item comes back — did it change?Core
- Gantry · ControlTowerPlan the whole sequence before the first moveCore
Free to play here. On Train, part of Core.
What this measures
How many examples you need before you can name the rule that splits them — rule induction, in this task. One hidden rule puts four of eight objects in one bin and four in the other, and the only thing you get is whether each call was right. The number is a count of examples, not a speed and not a score, and the eight objects are redrawn from new features every run so a rule you have already met cannot help you.
Play for your trials to the rule, climb the six structures in the order the research measured them to be hard, or sprint the clock, then challenge a friend.
How to get better
You mostly get better at this task — holding one guess at a time, testing it against the next object rather than the last one, and noticing when a feature you had written off is the one carrying the rule. That is a real, honest thing to practise. It does not make you quicker at working out rules anywhere else, and we would rather say so than sell you the other thing.
Questions
- What is a rule-learning task?
- You are shown one object at a time and put it in bin A or bin B. A hidden rule decides which bin each object belongs in, and the only information you get is whether each call was right. Working out the rule from that feedback is the task — it is the concept-attainment paradigm Bruner, Goodnow and Austin described in 1956.
- What does this test measure?
- How many examples you needed before you had the rule, in this task. That is the number: a count of examples, not a speed and not a score. It is one reading on one screen on one day, and it moves from run to run like any single measurement.
- Why are there exactly six structures?
- Eight objects varying on three yes-or-no features can be split into two groups of four in exactly six structurally different ways. Shepard, Hovland and Jenkins established that in 1961 and Nosofsky and colleagues replicated it in 1994, including the order people find them hard: one feature deciding is easiest, an exclusive-or is next, the three rule-plus-exception structures sit together in the middle, and the one with no rule below the whole set is hardest. The Six types mode is those six in that order, so its later stages are harder because they are harder for people — not because we added a shape.
- Why did my run end with no number?
- Because you did not reach eight correct calls in a row inside the 64-example budget, and a run that never reached the endpoint has no trials-to-criterion to report. We print the rule in plain words instead. A consolation score would be a claim about something that did not happen.
- Can I get better by remembering the last run?
- No. The three features and their values are redrawn from a new seed every run, and so is which of the eight objects the rule puts in bin A. The structure underneath is the same; everything you could memorise is not.
- Does my device or my speed affect the number?
- Not in the Quick test or in Six types. Both are untimed and both count examples, so the number is the same on a phone as on a laptop. The 30-second sprint is the one mode that reads a clock, and its records are kept per device class and never compared across them.
- What are the three modes?
- Quick test hides one exclusive-or rule and reports how many examples you needed before eight in a row came out right. Six types walks the six structures in their published difficulty order with a shrinking example budget, and continues on four features once you clear all six. The 30-second sprint states the rule in words and asks you to apply it — a wrong call subtracts one, so guessing nets you nothing.