AI Skill Report Card

Inventing Impossible Objects

A87·Nov 7, 2026·Source: Web

Generate ideas by colliding distant domains, inverting defaults, and following the result past "useful" into "actually new."

14 / 15

Pick two domains with zero obvious connection. Force a merger. Push until it stops being a pun and becomes a functioning concept.

Domain A: Grief counseling
Domain B: Weather systems

Collision: "Emotional climate forecasting" — a practice where instead of
predicting rain, a trained reader predicts the next 72 hours of a grieving
household's emotional pressure systems, issuing advisories ("high chance
of sudden anger Thursday, avoid major decisions") based on anniversary
dates, sleep patterns, and social load. Not therapy. Not astrology.
A new third thing: meteorology of feeling.

Notice what makes this work: it's not "grief is like weather" (a metaphor). It's a proposal for an actual practice with actual mechanics (advisories, forecasts, pressure systems as structure) that didn't exist as a named thing before.

Recommendation▾
Add a brief note on calibrating output length/quantity when a user wants many ideas vs. one deeply developed concept
14 / 15
Progress:
- [ ] Step 1: Pick two+ source domains with no natural overlap
- [ ] Step 2: Force collision — find the structural (not surface) link
- [ ] Step 3: Invert at least one assumed default in the result
- [ ] Step 4: Push past the first-draft pun into mechanics/rules
- [ ] Step 5: Stress-test: could this exist? What would break it?
- [ ] Step 6: Name it precisely — the name should do real work

Step 1 — Pick distant domains. The further apart, the better the eventual idea. Good pairs are mismatched in register too: one sacred/serious, one mundane/technical. (Grief + meteorology. Bureaucracy + dreams. Plumbing + memory.)

Step 2 — Find the structural link, not the surface pun. Surface: "clouds are fluffy like grief is heavy" (bad, shallow). Structural: both involve systems with pressure, fronts, and prediction windows (good, generative — it gives you mechanics to build on).

Step 3 — Invert a default. Take the obvious version of the idea and flip one assumption:

  • Obvious: a device that helps you remember things.
  • Inverted: a device that helps you forget things on a schedule, so memory decays at a rate you choose rather than randomly.

Step 4 — Push past the pun. The first version of any collision idea is usually just wordplay. Ask three times: "okay, but how would this actually work?" Each answer should add a rule, a constraint, a texture — something that makes it feel like a system rather than a sentence.

Step 5 — Stress test. Ask what would make this fail, who would hate it, what the weird edge cases are. Objects/systems that survive scrutiny feel real; those that don't were just a phrase wearing a costume.

Step 6 — Name it precisely. A vague name ("Emotion Weather App") kills it. A precise name carries the mechanics inside it ("Grief Advisories" implies issuance, urgency, expiration — more interesting than "app").

Recommendation▾
Include a short 'bad example' showing a failed collision (pun-stuck, not pushed further) to contrast with the strong examples
18 / 20

Example 1: Input: "Invent something new involving libraries and locks (as in, door locks/locksmithing)." Output: The Overdue Key — a locksmithing practice where keys are issued on loan, not owned. Every lock in a building is rekeyed on a rotating schedule (like book due-dates), and your key silently expires. To keep access you must "renew" it — a small ritual (return to a central kiosk, state why you still need entry) that creates an auditable, intentional relationship to space instead of permanent, forgotten access. Inversion: normally keys are permanent and books are temporary; here that's flipped. Mechanic that makes it real: a "late fee" for expired keys isn't money — it's a mandatory 24-hour lockout, so the consequence matches the domain (access, not cash). Edge case that proves it's real: emergency services need an override — so there's a "reference key" that never expires but logs every use publicly, like a book that can never be checked out, only read in the room.

Example 2: Input: "Something completely new, no domain given." Output: Inheritance Auctions for Habits — when someone dies, instead of (or alongside) willing objects and money, they will three habits — a morning ritual, a way of greeting strangers, a private superstition — to be auctioned among their descendants and friends. Winning a habit means committing, on record, to perform it for a year, with a small community of other "habit-heirs" checking in. Structural link: inheritance law (claims, executors, contestable wills) collided with behavioral psychology (habits as the real unit of a personality). Inversion: normally you inherit things that outlast you; here you inherit actions that require you to keep a dead person's rhythm alive through your own body. Mechanic: a "habit executor" verifies compliance, the way an estate executor verifies debts are paid — except the debt is to memory, not money.

Recommendation▾
Consider trimming some repetition between Workflow step explanations and Best Practices, which restate similar points
  • Force yourself to pick the collision domains before you know what the idea will be — if you start from a vague "I want something creative," you'll default to familiar territory.
  • Favor collisions where one domain has rigid rules/mechanics (law, meteorology, plumbing, auctions, locksmithing) — rigid systems are what let you build real texture into the new idea.
  • Invert only one thing at a time per idea. Inverting everything produces incoherent nonsense; inverting one default produces something recognizable-but-strange, which is the sweet spot.
  • Always ask "how would this actually work" at least three times before presenting the idea — this is what separates a real invention from a clever one-liner.
  • Give the idea a name that contains a mechanic, not just a mood.
  • When asked for multiple ideas, make sure the source domains differ across ideas — don't just vary the surface theme while reusing the same structural trick.
  • Stopping at metaphor. "Love is like a garden" is not an invention — it's a simile. Inventions need mechanics: rules, consequences, roles, objects, procedures.
  • Collision domains that are secretly close. "Combining cooking and baking" isn't a real collision — pick domains with no natural adjacency.
  • Inverting everything at once. This produces word salad, not a coherent weird idea. Change one load-bearing assumption; keep the rest recognizable.
  • Vague naming. "A New Way to Think About Memory" is not a name, it's a hedge. Names should be concrete nouns/phrases that do work.
  • Skipping the stress test. An idea that can't survive "what breaks this?" usually dissolves back into being a pun once someone pokes at it.
  • Defaulting to tech-product framing ("an app that…") — this flattens genuinely strange ideas into startup-pitch shapes. Let the idea take whatever form it wants: a ritual, a law, a profession, a piece of furniture, a unit of measurement.
0
Grade AAI Skill Framework
Scorecard
Criteria Breakdown
Quick Start
14/15
Workflow
14/15
Examples
18/20
Completeness
14/20
Format
14/15
Conciseness
13/15