Against Objectives
How come Einstein accomplished so much in his early years and failed to produce anything meaningful in his later years?
Did his cognitive abilities simply decline as he got older? Maybe.
Was the frictionless environment at the Institute for Advanced Studies actually harmful? Possibly.
But I now think the main factor is something else: Einstein fell prey to the grand objective fallacy.
It roughly goes like this: Grand things are accomplished by visionaries who set their sights on a clear objective and relentlessly keep going until they get there.
It's a narrative the human brain loves.
It's like a comforting blanket in an otherwise overwhelmingly confusing world.
The only problem? It's completely fake.
In hindsight, it's always possible to connect the dots in a way that fits the grand objective story arc.
But that doesn't mean this is really what happened.
The real story is often far too confusing, weird, and frankly boring.
Big things are usually accomplished by people following their hunches and intuitions with no clear plan or objective in mind.
People fixated on some grand objective tend to get stuck and never accomplish something meaningful.
That's precisely what happened to Einstein in his later years.
His grand objective was to develop a unified field theory. That never worked out.
Sure, if you aim high it's not too surprising when you fail.
But the real tragedy is that none of the ideas he developed while working on his unified field theory had or have any relevance to other researchers.
Einstein's approach earlier in his career was way different.
He didn't have a grand masterplan to revolutionize our understanding of energy and matter or to overthrow classical mechanics or Galilean relativity.
Instead, he was just following his hunches and intuitions, solving problems that were right in front of him (quite literally in the form of Henri Poincaré's book Science and Hypothesis).
There are experiments backing up this idea.
Researchers training robots tested two different approaches:
Train using objectives like "get out of the maze" or "walk".
Train using novelty search where the robots simply try to do something different than they had done in the past.
Interestingly, novelty search outperforms traditional objective-based algorithms in most scenarios.
Robots learn faster and more reliably to navigate out of a maze by always trying something new instead of focusing on the objective "get out of the maze".
This works because after crashing into enough walls, crashing into a wall stops being novel. So the only way to do something novel is to not crash into walls.
In contrast, robots focused on getting out of the maze often get stuck.
A deeper force at play here is that objective-based algorithms provide an ordering from "good" to "bad".
With a clearly defined objective you can directly assess any action using it.
The issue with this approach is that the stepping stones that actually lead to the objective are often unpredictable.
Just like Einstein with his ambitious vision of a unified field theory, robots get stuck in cul-de-sacs because the steps leading there look promising when comparing them to the objective.
In contrast, novelty search, provides an ordering from simple to complex.
Once all simple behaviors are exhausted the only ways to behave that offer something novel are complex.
Novelty search can't of course guarantee that you will reach a specific objective.
But it usually leads to interesting outcomes. That's a direct result of the implicit ordering from simple to complex.
Objective-based approaches, on the other hand, can't guarantee success either. But they typically don't lead to interesting results along the way.
This also matches my personal experience.
Every time I set out to do something grand I failed.
Everyone I know who set out to do something grand didn't.
And again, the real tragedy is that nothing meaningful came out of these periods as a byproduct.
In contrast, everything meaningful I created was the result of open exploration without a plan or objective.
For example, my first book, Physics from Symmetry, was not the result of me trying to write a book.
I was just taking notes for myself, trying to figure out how far I can get in deriving the theories of modern physics just using symmetries as my starting point.
Only after many months when I talked to a friend about the project he mentioned "so this is like a book, you should publish it".
That's when I realized "He's right!". I had indeed written a book without realizing what I was doing.
I always made more money and had more people reaching out to me to say thanks whenever I wasn't trying to make money or help people.
The lesson that novelty search trumps objective-based approaches is an incredibly hard lesson to execute on.
Virtually every biography reinforces the myth of the grand masterplan, the big vision.
So we feel like if we also want to accomplish something big, we should follow a grand vision and develop a masterplan.
You can't just follow your hunches and hope that something good comes out of it.
That's just not how things work.
Except it is.
As soon as you let anything but curiosity be the driver and rudder you are on the path to failure.
But for how many people is curiosity the sole driver?
Few.
Which is why few meaningful things get done.
Which is why letting yourself be guided by curiosity feels so weird.
For most people objectives like money, status, doing something that sounds cool, are the driver and rudder.
And no human is totally immune to these mimetic forces.
Even, for example, most researchers are driven by other factors like status above all else.
You can't get funding for open, curiosity-driven exploration.
Which means there is no place for curiosity-driven researchers in today's academic system.
Which is why scientific progress is stagnant in so many fields.
At one end of the spectrum you have academics carrying out incremental research purely driven by short-term objectives.
At the other end of the spectrum you have crackpots driven by ambitious objectives never producing anything relevant.
You have startup founders with big visions failing to change the world and millions of people solving incremental problems for a regular salary.
There's simply no one "chasing hunches and interesting problems without narrow material and objective constraint”.
Yes, this is all a bit sad to think about.
But it also means that if you're able to fully embrace an open-minded, objective-free approach you have an insane competitive advantage.
There's no competition because everyone else is getting stuck in local maxima.
The big secret is that a purely curiosity-driven approach works.
You might not get rich or famous or make a ground-breaking discovering.
But the odds are good that something interesting will come out of it.
To me this feels like forbidden knowledge.
The whole school system is designed to brainwash every member of society to always have clear objectives.
You're not supposed to wake up each day and fly by your own creative compass.
You should have an objective like making enough money to buy a nice house or do something good for humanity.
Goals have an "irreproachable and unimpeachable status".
At any time you should have a good answer to the question "What are you doing?".
"Just exploring" is not an acceptable answer unless you're 23 and on a backpacking trip through Europe.
The key is to realize that drifting through life and exploration while looking similar from the outside are actually polar opposites.
Drifting is passive and leads to a narrowing of perspective while exploration is, well, active and widens it.
A stable society, of course, needs processes driven by clear short-term objectives.
Short term objectives are important when the path from A to B is clear.
But paradigm shifts, dents in the universe, those genuine leaps forward come from poking around, exploring, and following hunches, not grand visions or masterplans.
Einstein lost track of that in his later years.
So did you.
So did I many times.
And that's why I wrote this.
Really as a reminder to myself.
It's time to go explore again.
Best thing I have ever read.