The Explanatory Leap
Criticize ideas in your head to jump gaps that trial-and-error can never cross.
- Difficulty
- Advanced
- Time to result
- ~ongoing to results
- Steps
- 6
- Confidence
- 66%
Deutsch separates two search processes. Biological evolution varies blindly and selects in the world, so it can only take small steps and every intermediate has to be a viable organism — which is why nothing ever evolved a way to deflect asteroids. Human creativity works differently because its conjectures are explanatory: they say why something would work, which lets them be criticized inside a model instead of tested in reality. That criticism is what lets a mind jump straight to an idea whose intermediate versions would all fail — a redesigned tailplane, a skyscraper — without paying for those failures. The practical mechanism is: state the problem situation, conjecture an explanation rather than a variation, criticize it hard in the model, then build only what argument could not settle. Recombination describes the parts; it never accounts for the leap.
Origin
David Deutsch's account of creativity from The Beginning of Infinity, worked through live on Naval's The Deutsch Files III against the popular 'creativity is just connecting things' claim attributed to Steve Jobs.
Core principles
- 01An explanation says why something works, and that is what makes it criticizable.
- 02Evolution has to stay alive at every step; a mind does not.
- 03'Creativity is just mixing things together' is false at the word 'just'.
- 04Universality means any idea can be connected to, and criticized against, any other.
- 05Real novelty is recognizably in the old style and recognizably new at the same time.
How to run it
- 1
State the problem situation
Write down what is actually wrong or missing, and what you currently believe about why. Without a problem situation there is nothing for a conjecture to be a conjecture about, and no generic advice can help you.
Pro tip If you cannot say what would count as an improvement, you are collecting ideas, not solving a problem.
- 2
Conjecture an explanation, not a variation
Propose a change together with an account of why it should work. A variation can only be judged by testing it; an explanation can be judged by argument, which is thousands of times cheaper.
Pro tip Force the 'because' clause into the same sentence as the proposal.
Watch out A proposal with no mechanism smuggles you back into blind trial and error.
- 3
Say it in one sentence and expose it to criticism
Compress the conjecture until someone competent can attack it in minutes. Deutsch's example is a whole aircraft redesign proposed in one sentence and then critiqued by an aeronautical engineer.
Pro tip The right critic is the one whose objection you cannot already predict.
- 4
Ask whether the path has to be incremental
Check whether each intermediate state has to survive on its own — physically, commercially, organisationally. If it does, you are stuck with evolution's constraint. If it does not, the gap can be leapt.
Pro tip Most 'we must do it step by step' claims are about organisational comfort, not about viability.
Watch out Incrementalism is genuinely mandatory wherever every intermediate really must function; do not leap over a live system.
- 5
Leap, then test only the residue
Build the version that survived criticism, and spend real resources only on the questions argument could not decide. The series of aircraft that would never have flown never gets built.
Pro tip Log which failures criticism saved you from — it is the cheapest evidence that the method is paying.
- 6
Judge the output by recognizable novelty
A real creative result looks unmistakably like your work and unmistakably unlike anything already in it. If it is fully inside the space of your prior output, you have interpolated, not created.
Pro tip Ask a knowledgeable outsider to place it: they should recognise the author and be surprised by the piece.
Watch out Systems trained purely on your existing work can only reach the space those works already span.
In the wild
Deutsch contrasts a designer saying, in a single sentence, that an aircraft might be better with the tailplane as a single object rather than a winged assembly — a claim an aeronautical engineer can immediately criticize — against making that same change incrementally, which would produce a whole series of aircraft that do not fly. The leap costs one sentence and one argument. The incremental route costs a fleet of crashes. The difference is not intelligence or speed; it is that the conjecture came with an explanation, so it could be attacked in a model rather than in the air.
→ A design change is settled by argument in minutes instead of by a sequence of unflyable prototypes.
No lineage ever evolved asteroid deflection, because an impact every few million years exerts no selection pressure and there is no viable incremental path to it. Yet humans can reach the idea directly, because the conjecture is explanatory rather than a mutation. Deutsch and the host push this into a SETI proposal: instead of scanning for biomarkers, look for a planet where asteroids are being repelled, or where rocks called buildings are conspicuously not eroding. Neither has any biological or random explanation — only explanatory creativity produces them.
→ A concrete search criterion for intelligence that follows directly from the difference between evolution's search and a mind's.
Common mistakes
Calling creativity recombination
If every idea were a recombination of earlier ideas, nothing about a skyscraper would be new that was not already present when our ancestors banged rocks together. The recombination story describes the ingredients and silently skips the step that matters.
Testing before criticizing
Running experiments on conjectures that a competent critic would have killed in a sentence spends real resources on failures argument was already able to eliminate.
Scaling compute instead of explanation
Deutsch's guess is that the missing ingredient is not processing power at all — throwing masses of computer power at a problem substitutes brute search for the explanatory step that actually does the work.
Is it for you?
Best for
Researchers, designers and founders working on problems where testing every candidate is impossibly expensive.
Not ideal for
Well-charted optimisation problems where cheap measurement already beats argument.
From the transcript
“it's not only faster it is explanatory which means that because it's explanatory it means it can leap over gaps in the knowledge space that…”
“if that's a good idea it could be criticized by an aeronautical engineer and so on but to make that change incrementally will probably produce…”
“if all our ideas are just recombinations of ideas that have happened before then there's nothing new about the skyscraper that wasn't already there when…”
From the episode
The Deutsch Files III