There’s this concept in AI called overfitting (don’t be impressed by my knowledge, I had an amazing dinner with Brian Christian
last night and he told me about it), which I found absolutely mesmerising and so relevant to explaining one’s childhood.
When AI is being programmed to learn something, it’s put in a test environment: it’s given data to solve a problem and it tries to solve it again and again until it learns to do it properly. When AI is doing this, it often takes shortcuts. It programs itself to behave in ways that make the overall process easier and faster to do.
The problem with this is, once you then launch the AI in the real world to work with real data, it doesn’t work. The shortcuts it programmed for the test data don’t work with all the different types of real inputs. So the AI is basically useless.
The smarter and the more powerful the AI, the more and the bigger shortcuts it will make, and the more likely that it won’t work in the real world.
To me, this sounds like how we are programmed.
As a child, we’re like a little AI machine programming ourselves. And the test-data and input we use is (mostly) our family home. We create shortcuts to adapt to our parents needs, behaviour and reactions, the interactions and systems set in our family, we program and reprogram ourselves around the shifting-test-data that is our environment until we’re fully set. We’ve fit our identity, beliefs and truths around the tests that we had to run again and again.
We have a drive to survive and thrive, and we’re all built to survive and thrive in our test environment, but just like AI, I think we can over-fit to it.
When we’re out in the ‘real world’, away from home, which for me has been almost over 10 years now, I find with every passing year that I’m uncovering more and more ways that I’m programmed un-ideally to cope with world.
My familiar ‘shortcuts’: beliefs, behaviours, thoughts are perfectly created for an environment that has data very different to the variety I’m having to face. Unlike AI though, I can’t just ‘not work’, I’ll have to function and take the hit of maladaption in negative consequences to myself, emotional pain, bad results, distress.
Not everyone is my parents and siblings, not every system is my family system, and my overfitting to that test-data leads to behaviours that in reality, just glitch.
I don’t like calling people ‘smart’ or not, but the label does easily transfer a description of intentional-targetted-effective-thinking, which the more a child probably has to do to fit their environment, the more they probably overfit it. And the more inflexible/counterproductive they might be in the real world. Maybe some children can be ‘too smart’ to be happy adults?
Frameworks like this are so valuable to me. I’m not passionate about arguing if they’re right or wrong, they don’t have to be right at all: the point of them is, sometimes they just are so useful for reframing, understanding, and highlighting the moving parts and the possible reasons for things being the way they are. And therefore they can be the way to enact positive change.
Accepting that your now unhelpful behaviours or ways of viewing the world are most likely ancient, reasonable, maybe even intelligent coping mechanisms for a test-environment you’ve long left, can help you get rid of them.
Creating your own frameworks for cause and effect: my parents did this, so I had to respond in this way, so now I might think this is true - don’t even need to be 100% accurate, but they can be 100% valuable in uprooting the shortcuts, reconsidering new data, becoming functional in our new environments, and maybe even, avoiding overfitting to whatever data we are being fed at the moment in our current setup.
The only constant is change :)