Kenneth Stanley: Innovation insight

I recently enjoyed listening to Kenneth Stanley - Greatness without goals on the “Invest like the best” podcast with Patrick O'Shaughnessy. Schedule a quiet 75 mins and listen to it in full if you can, or have a read of my summary below.

It is always dangerous to gravitate towards information that “rings true”. We are hard wired for selection bias. But Ken’s take on the nature of complex problem solving, gels with both my personal, first hand experience, and much of the literature I have studied at length from complex system science, broadly speaking.

Ken is a Professor in Computer Science and a pioneer in the field of neuroevolution. He is also the co-author of a book called, Why Greatness Cannot Be Planned, which details a provocative idea that setting big, audacious goals can reduce the odds of achieving something great. My summary of the podcast goes a bit like this:

Core hypothesis:

• Goal-orieted, or objective led behaviour may be problematic when trying to solve certain problems (not all problems). For example, goal oriented behaviour to improve health or fitness can be quite effective (simple problem with known pathway), but goal oriented behaviour to invent a fundamentally new product or service may be doomed to fail complex problem with unknown pathways);

Research context/genesis:

• The finding stems from AI research by Kenneth Stanley and colleagues:

  • See Picbreeder for an overview of the core research, summarised as:

    • Picbreeder lets you pick the blobs/patterns that you like the best, and then the picture you choose has “children”

      • Children are not exact duplicates, but look a bit like the parents

      • Eventually blobs start looking like something recognisable. For example, blobs may end up looking like a butterfly after sufficient breeding events

      • New users then don’t have to start from scratch, they can choose to start from something they think looks a bit like an object they like, and breed new images from there

  • So Picbreeder is a metaphor for the process of innovation and discovery, as it builds on what has gone before, but sometimes throws up surprising/novel changes

    • Ken states that In 99%+ of cases in picbreeder, when you go back through the history of images that end up looking like a real world object/thing, go back far enough and you will end up finding an object that looks nothing like the finished object

    • In other words, people that discovered these objects that looks a bit like something in the real world, were not trying to discover them or select images that would lead to them

    • Conversely, when tasked with trying to pick images with a goal in mind, that goal could not be achieved. For example, give users the task of breeding to create a butterfly, and it is never achieved

    • They even tried using a machine machine learning algorithm designed to pick blobs that look like an end-goal, for example, a skull. It failed.

So, why does this occur? And is there empirical evidence/data to support the research?

• The research finding is similar to the “like causes like” fallacy, which comes from philosophy/cognitive science. E.g see - https://en.wikipedia.org/wiki/Fallacy_of_the_single_cause

• For example, when you look back at the history of stepping stones taken on the path to creating major inventions, they always rely on stepping stones completely unrelated to the end-goal. For example, see history of vacuum tubes

• Interesting anecdote - if you include a consensus mechanism that lets users pick blobs by voting into picbreeder, blob evolution stagnates and doesn’t evolve into anything interesting/recognisable. Is this a metaphor for “design by committee” diluting good design? It would seem so

How can we usefully apply the research?

• The answer is not to do random things and hope they connect with something useful, which is a common misinterpretation of the research. So how do we action the insight?

  • Need to recognise/accept the likelihood that there is no probably good/perfect strategy to achieve a specific thing that requires complex stepping stones. This first step helps liberate your thinking

  • So, what can you do?

    • Need to keep an open mind and keep multiple stepping stones available to you, so you can pick and choose which to step to next

    • Why? because it is the interesting stepping stones that we could take next, that have power in the creation process - so the more you have available, the more chance you have of getting somewhere useful

    • So, collect stepping stones and honour what is interesting to the individual - what an individual finds interesting or useful, as opposed to the crowd, helps reduce the risk of consensus crowding out useful

    • Recognise stepping stones when they snap into view (need to let go of old ones). Stepping stones could be novel/new technologies/services/trends that you notice.

Did you find this article interesting or helpful? Do you need help refining your innovation process, or figuring out how to collect useful innovation stepping stones?

Use the comments below for feedback, or the contact page to get in touch