Lessons from Breeding Pictures Part I
Aka lessons from the book "Why Greatness Cannot Be Planned"
I wrote the whole long post while I was in design school in 2019, and still haven’t seen any reposts or convincing arguments against the findings of this book.
For Substack readers, I’ll be posting this in three parts.
There remains one a priori fallacy, or natural prejudice, the most deeply rooted perhaps of all which we have enumerated… that is, that the conditions of a phenomena must, or at least probably will, resemble the phenomenon itself.
John Stuart Mill, “A System of Logic, Ratiocinative and Inductive”
Recently during the spring break, I re-read the book “Why Greatness Cannot Be Planned” by Kenneth O Stanley and Joel Lehman. The book’s ideas are especially applicable for domains which I think broadly of as “search” i.e. one is exploring a “search space”, much like a treasure hunter. This contrasts with “execution” domains, where there is often much greater clarity about the domain, and the focus is just on doing. The recent book “Trillion Dollar Coach” talks about the innovation vs execution spectrum; the ideas of “Why Greatness Cannot Be Planned”are largely applicable to search & innovation.
I remain amazed that there aren’t more blogs, essays or articles summarizing this book, as I think the book’s ideas have HUGE implications for art, innovation, design, and society at large. That is why I will attempt to summarize, and spell out the implications of this book’s ideas here today.
To understand the background, you first have to understand the experiment that sparked the insights.
Background: Picbreeder and Open-Search
Picbreeder is an online website that allows any user of the website to “breed” pictures. From an initial population of blobs and lines, a user can select a desired picture (see blue circle above); the website’s algorithm then generates a new set of pictures, using the “genome” of the selected picture.
This goes on, ad infinitum, until you basically end up with a genealogy of images, with different branches of family trees (kinda like my extended family, which has some “dotty” and weird branches…).
Stanley explained that he got his key insight when he suddenly ended up with a sports-car picture. Prior to that, he had selected an alien face. How, he thought, did an Alien Face lead to a Sports Car? It was by accident! Most importantly, he realised that if he or anybody earlier had selected earlier images for a Sports Car, he would NOT have ended up with a Sports Car.
Because the Alien Face is a critical stepping stone to the Sports Car, but looks nothing like it (a la John Stuart Mill’s quote above).
If you see the bottom half of the picture above, you’ll see that if the selection process had “sports car” as a search filter, it would have filtered out the useful genes that ended up being the Sports Car’s “parts” (e.g. the wheels were derived from the Alien eyes). If you had sought out a Sports Car at the start, you would likely NOT end up with a Sports Car, because you would have missed the stepping stone (i.e. Alien Face) to the Sports Car. In other words, setting ambitious “objectives” is actually self-defeating, as the objective leads away from the objective. Objectives don’t work, unless you’re just one-step away (e.g. Alien Face to Sports Car). Stanley is quick to articulate that this concept is only applicable to ambitious objectives, i.e. objectives that are very far from where we are now.
The two catches that complicates the search for these stepping stones are:
It is impossible to know in advance exactly how many stepping stones away you are from the objective. Stanley made the comment that the Apollo moonshot happened to be one-step away from the technological boundary of that era, but they didn’t know it when they started. (It’s quite shocking to my mind to consider what the parallel universe would have been like, if we never succeeded in putting a person on the moon.)
It is also impossible to know in advance what stepping stones lead to the objective. I think the example of convergent evolution illustrates how different stepping stones lead to the same objective: fish, dolphins and ichythosaurs look superficially alike. However, the stepping stones to each streamlined shape from each type of animal were quite different, and this shows from the unique differences of each physiology e.g. fish have gills, dolphins have blowholes etc.
Next part, I will cover the recommendations of the book.
Key Learnings
Ambitious goals are complex, multi-step processes. There is very little clarity of which steps lead to what outcomes.
Optimizing for your end-goals can kill your chances of reaching the end-goals, because you might choose stepping stones that do not lead to your end-goals.
Looking for stepping stones is complicated by two facts:
we often don’t know we are exactly how many stepping stones away from the objective
we also often can’t know what stepping stones lead to the objective.
Question to think about
Which ambitious goal or objective have you achieved in your work or life?
What were the stepping stepping stones that led to the outcome?
Did the stepping stones look anything like the outcomes?