The Power Law, Combinatorial Progress, Narratives and Curiosity
This week’s research note is on friend-of-the-show Rohit Krishnan. Rohit is a VC and essayist who writes fascinating, thought-provoking essays on complexity, progress, innovation and technology over at Strange Loop Canon.
He previously appeared on episode 94 of Infinite Loops and is due to make a second appearance shortly this year.
Let’s get to it! As ever, if you have comments or questions please don’t hesitate to give us a shout in the comments or on our Twitter.
Thanks for reading Infinite Loops! Subscribe for free to receive new posts and support our work.
The Power Law is Everywhere
The Combinatorial Shape of Progress
Cultivating Curiosity. Encouraging Exploration. Accepting Failure.
How Narratives Constrain our Actions
Theme 1: The Power Law is Everywhere
Modern society is defined by the power law - the heavily skewed distribution of returns in favour of a very small number of market participants. This applies in the classic VC sense but the concept can also be used more generally as a way of understanding the world. Distributions in the political, economic, and cultural sector are all heavily skewed towards a very small number of participants.
Understanding the role of the power law allows us to adjust our policy solutions accordingly. In a broad sense, it means we need to be far more comfortable with failure and experimentation - the cost of an increase in the number of our mistakes will be dramatically outweighed by the returns generated by finding the ‘right’ successes. It also means that talent recognition and support is incredibly important.
As the world becomes more complex and interconnected the impact of the power law will increase.
Rohit’s discussion of the power law is based primarily around asymmetric positive returns. How do asymmetric negative returns fit into his worldview? Isn’t the risk of producing asymmetric negative returns also increased by expanding experimentation? This can be called the ‘Jurassic Park’ argument (“Your scientists were so preoccupied with whether or not they could, they didn’t stop to think if they should.”).
The power law framework implies that policymaking should focus more on the tails than the average. What would a tail-focused set of policies look like in practice?
Under the logic of the power law one person can make a huge amount of difference. There is a subtle tension in Rohit’s thinking between systematic accounts of progress and individualistic accounts. How does he weigh these relative to one another? What are his views of the Great Man theory?
Is the power law a theory of history? Or is it a unique symptom of the interconnected complexity of modern society?
What remains is a much starker realisation that we do inherently live in a power law world. And in a power law world getting the few things right is way more important than being right more often, even at the cost of being like Dr Doom. We should all try and be wrong more often.
We all fall somewhere on a bell curve that determines our abilities. Yet our exploits often follow a power law. It holds true for wealth, for accomplishment, for scientific discoveries and genius. And in the days when the status-bar for who all you knew was barbell-shaped, where you knew your locals and also knew the King, very few people broke out.
The problem with things that are stacked according to a power law is that wishing the top of the distribution was wider can't make it so. And the distribution of top performers is most definitely a power law, in economics, politics, even sciences. The only realistic option is to expand the whole distribution.
It sure seems like we're living in a world that's bifurcated. The rewards to being the largest, or number one, is extremely high, while the rewards to being the median is ever declining.
Theme 2: The Combinatorial Shape of Progress
The world is incredibly complex and interconnected.
Innovation (and therefore progress) is driven by the combined effect of growth in different sub-domains. Growth within different sub-domains happens at different speeds and is subject to different factors.
The growth of each sub-domain is in turn interdependent and combinatorial. New combinations of inputs in one field may produce an output which is in turn a preconditional input for a new technology in a separate field.
As a result, progress tends to take the shape of an ‘S curve’ – there are periods of growth followed by period of pauses where different technologies and innovations are catching up with one another. Different rates of progress amongst different technologies are a result of complexity, regulation, coordination taxes, risk aversion and bureaucracy.
Automation and computation could be the routes out of our current moment of stagnation.
Kurzweil argued that over a sufficiently long timeframe S curve progress cycles can be averaged out as an exponential trend. Does Rohit agree? (See Kurzweil 'the singularity is near').
What is the interplay between the S curve model of progress and Rohit’s theory of narratives (see below). Are narratives one of the ‘technologies’ that need to catch up in order for innovation to take place?
Elon Musk talks a lot about ‘jumping’ S curves (i.e. moving from one S curve to another). I’m interested to know how Rohit’s theory accounts for jumps between S curves – is this in of itself a combinatorial process or is there a separate framework needed to understand this?
The ‘gradually then suddenly’ model of progress has wider uses. Tanner Greer argues that ‘the culture war is a long war’, meaning that ideas that suddenly become popular today have been gradually percolating throughout society for generations, waiting for the optimum blend of socio-cultural-environmental factors before emerging.
The belief here at Strange Loop has been in agreement with the uneven distribution of the future but focuses more about the unseen power of combinatorial growth.
The thesis that we’ve espoused time and time again has been that progress is combinatorial, and needs the ingredients energy (to build stuff), materials (to build them with), information (to know what and how to build) and life!
As a solution, I suggest that innovation which leads progress is itself caused by combinatorial growth in its sub-domains, which first naturally creates accelerated growth, and later slowdowns as a result of a slowdown in any individual part or an increase in combinations to be tried.
New technologies emerge from the combinatorial interactions of existing technologies, as existing technologies act as complementary inputs to create a new technology, as long as the combination crosses a growing threshold
My thesis is that we're in a period of stasis simply because the number of technologies we need to progress don't all progress at the same rate.
Theme 3: Cultivating Curiosity. Encouraging Exploration. Accepting Failure.
Together, the power law and the combinatorial nature of progress mean that as a society we should encourage experimentation and curiosity. This will allow us to expand the “surface area of all possible ideas”.
In doing so we need to accept and embrace failure.
The route to innovation is likely to be unpredictable. We can only find such route if we act in unconventional and experimental ways.
Variance is good – the more variance there is the more opportunity we have to hit those power law returns.
Specialisation and bureaucratisation diminish our curiosity.
The flaw in effective altruism is that it seeks to reduce complexity into something more legible and measurable. But, the world is dynamic. By prioritising optimisation over experimentation EA is leaving itself vulnerable to unknown unknowns. Instead, EA should be encouraging a plurality of attempts and experimenting with far more diverse ideas that may not immediately pass the “QALY inspection”.
Has society become less curious?
Is information abundance an asset or a liability in cultivating curiosity?
What is the primary constraint on curiosity? This is important as until we identify this we cannot identify the correct policy solution. One could argue that the primary constraint is material factors (failure is a luxury some can’t afford), in which case focusing on curiosity / experimentation is focusing on the wrong problem.
Rohit’s case for experimentation is largely an instrumentalist one (by experimenting we can increase the chance of hitting outsized returns). What is the philosophical case for experimentation?
On a societal level, is there a marginal point at which experimentation becomes a negative rather than a positive? If every institution prioritised experimentation / curiosity over consistency / tractability would this be a good thing? One could argue that it is only via having some institutions which focus on consistency / tractability that we can have sufficient slack in the system to enable other institutions to experiment and embrace failure.
The above question ties into Rohit’s criticism of EA. In his piece, he argued that EA needs to experiment more with ideas that do not immediately appear tractable. He says that EA needs a market. But why does EA specifically need to fill this function? We already have a market – a huge number of non ‘EA’ businesses are aiming to deliver social good, just not under the EA banner. When viewed in a societal context rather than in isolation, is EA therefore performing a specific (and essential) function in focusing on tractable and solveable problems?
Effective Altruism is a constant fight against Knightian uncertainty
For us to have long-lasting success, we need to be aiming for far more experimentation and try our hand at new ways of encouraging innovation.
Innovation is usually the only way out to create a new S curve to capture and grow. To do that, you need to allow experimentation and create the right incentives for people.
The Humble Experimentalist tries things out to see if they work. If they work they are tried more often, or scaled up. And eventually we may reach a new equilibrium. This new equilibrium might comprise only one strategy, or both, or several mixed in varying degrees.
The opportunity to let people focus on that which tickles their curiosity is a process that lets us drastically expand the surface area of all possible ideas. The most intriguing notion is that this isn’t necessarily true for just the full time pursuits people do, but allow people to have much more diverse portfolio careers.
The point relevant to innovation overall is that the path that leads to discoveries, the ones that are most interesting to us, often are wildly circuitous and completely unpredictable. As Kenneth writes in his fantastic book Why Greatness Cannot Be Planned, objectives can therefore often be more counterproductive if what you wanted to get are ambitious and innovative outcomes.
The problem with expanding the whole distribution is more straightforward. Any program that promotes it will have to be far more okay with failure.
Theme 4: Networks Matter
Knowledge is an infinite set of interconnected lattices. By building networks we can connect different nodes of the lattice and therefore illuminate more of what we know.
By linking people and ideas we expand the visible scope of the lattice.
Clusters of skilled people have an outsized effect on the output of everyone within the cluster. The output of networks of talented people is therefore greater than the sum of its whole.
Networks are built on trust.
Our ability to craft developments are reliant on our ability to create distilled knowledge and disseminate it to others. AI performs the function of using distilled knowledge more effectively.
What type of socio-political environment does Rohit think is ideal for cultivating knowledge production and transfer? City states appear to have been a successful model (Florence, Athens etc). Does he agree with this, and if so why? What would a modern-day recreation look like? Could it be Balaji’s network state?
The lattice model of knowledge focuses on the production and the transmission of knowledge. In doing so, one could argue that it underplays the role of authority and legitimacy in the development of knowledge. For an idea to have currency it must not merely be located and transferred, it must also be espoused by someone who commands authority and legitimacy. How does Rohit’s worldview account for this?
What distinguishes a high quality network from a low quality one? What are the common features to high quality networks?
Are networks generally top down or bottom up phenomena?
Is genius culturally contingent?
Personally, I think of our knowledge as a crystal, dark, of unknown dimensions, infinite lattices linked to each other, and immeasurable size. What we get to see are small slices, and as we connect what we see in those sources slices, they light up and illuminate what we know. With every effort we expend the circle of light expands, and we learn more.
What that means is that our job in a society isn't necessarily to prompt getting more individual geniuses, or even to try and foster them through education or similar activities (at the margins), but rather to make the linking of people together easier.
All knowledge that we acquire gets hung on our own version of a lattice. Most of the time we might not even know that that's where it's getting hung, but nevertheless it exists. That becomes the definition of our causal-web. What each additional piece of knowledge does is that it helps expand the web. With the causal-web in place, created and interacting through multiple large clusters, then it stands to reason that the webs interact with each other to create a system.
The innovation landscape is more like a city of spires standing tall together rather than a solitary tower. The challenge is to find and encourage the formation of more clusters well in time where further expansion and attraction of talent can happen.
So the requirements become larger numbers of smart people, who have the time to cogitate and work on things that go beyond their daily bread, connected in networks that enable them all to rise together, rather than just succeed individually.
You see the same thing in history. Wherever there was a large enough cultural shift to create history-changing momentum, there usually was a large enough cluster. Ancient Greece, The Renaissance, Industrial Revolution. This isn't true of all cultural superiority or dominance, but it's oddly prevalent in areas where individual outperformance pushed human thought forward.
Theme 5: How Narratives Constrain our Actions
In an extremely complex world, stories and narratives allow society to coordinate actions and to enhance the efficiency of decision making.
Stories allow us to chart a path through complexity.
Stories constrain the 'option space' for decision making.
The creation of narratives requires agency and imagination. Bureaucratisation and specialisation negatively impacts our imagination.
We have a human tendency to find patterns where there are none. We use narratives to create meaning.
Rohit’s theory of storytelling appears to view it as a bottom-up process. He argues that the stories we tell ourselves ultimately guide policy. But one could make the counter argument that stories are a top-down process used to justify existing power structures. How much agency do we have over our stories?
What is more important to the power of a story – the story or the storyteller?
Rohit argue that stories and narratives are constraints. But the scope of possible stories is infinite. Almost anything can be justified via the right story. On this basis, one could argue that stories expand the potential path of actions rather than constrain them. How does his theory of storytelling account for this?
In a recent piece Rohit wrote about our tendency to find patterns where there are none, evoking the finance world as a particular example. He writes elsewhere how specialisation prevents us being able to understand how different pieces of information link together, resulting in us ‘extrapolating widely’. Finance is a highly specialised industry. Is there a link between specialisation and spurious pattern recognition?
One of the abiding strange loop beliefs is that navigating the complexities of the world requires the creation of narratives to guide you. And this idea that stories have shapes definitely makes you wonder how to deal with the world.
The second is that the stories we tell ourselves are important. It's what helps us triage the universe of possibilities into a few manageable narratives. And those narratives in turn guide our philosophical beliefs, which guides the policies we feel are appropriate and fair.
Regardless of which realm you look at, it seems to be clear that narratives constrain everything around us. It impacts how we view history. It impacts how we understand discovery, causality and change itself.
The collective fictions we believe in guide our actions, and also guide our options much like any environment guides the evolution of its organisms.
The pursuit of meaning is the noble goal that drives our species. We see this in the mundane, like asking the tea leaves of quarterly numbers to explain the present and predict the future for us, to the divine in seeing Jesus on a potato. But if you do find Jesus on a root vegetable or an incredible investing opportunity in a mass of equities data, that too is well and fine. Your sincere belief in that pattern guides an action, which has a consequence, that hopefully results in future actions being corrected.
Rohit mentions that the we confuse communication latency with analysis speed. Does this mean that efficiency is overrated?
Can one intentionally cultivate authenticity?
Is Strange Loop Canon a project of building or maintaining?
How did Rohit’s time at McKinsey inform his parenting style?
Rohit has written how some of the AI risk discourse is “a destruction myth. It’s apocalypticism crowdwriting its own biography”. My understanding of his argument is that we are focusing on the destination more than the journey – we are imagining a fatalistic future and then working backwards from that, rather than focusing on the issues that are actually in front of us today. A response to this is that by the time we have sufficient evidence that the fatalistic future is a tangible possibility it may be too late, and that in previously dismissing such possibility as a genuine concern we have overlooked offramps on the path to that fatalistic future. What does he make of this argument?
“I can’t write without a reader. It’s precisely like a kiss—you can’t do it alone.”
― John Cheever