How DFS, poker, basketball, Starcraft and strong opinions, weakly held (SOWH)can help your startup [Part 1]
Since my co-founder Ricky introduced me to the concept of “strong opinions, weakly held”¹, I’ve been thinking a lot about why this concept seemed to resonate with me and how to deliberately practice it. For brevity’s sake, I’m going to henceforth refer to this concept as SOWH (pronounced SOW as in you reap what you sow).
For many worthwhile endeavors, whether that means figuring what to study in school, what to build first for your startup, or how to find the right partner, you typically start with a very limited set of information. It’s rarely clear at the outset what the best course of action is, or even what the first step should be. In school, we were encouraged to withhold judgment, keep an open mind, and try not to decide on anything until we had gathered enough information. Moreover, in group decision-making settings, when we’re aware that nobody has all of the information, we tend to try to defer to consensus building. However, when the amount of available information is so lacking to the point of feeling debilitating, we often choose to punt the decision—to make a non-decision. We think: why make a choice now when you can make one later with more information? What if we make the wrong choice? Worse yet, what if we make a choice that limits our optionality or completely shuts the door on an opportunity? It is much safer to try to gather information before committing to any course of action. However, SOWH would argue that action based on a clear hypothesis is the most efficient and effective way to get information, so long as you are willing to let go of your initial hypothesis in light of evidence to the contrary.
So WTF is SOWH and SO-WHAT? Let’s decompose and start with “Strong Opinions”.
Having a strong opinion² means that you look at the information given to you and formulate your best hypothesis h₁. You then proceed to act upon h₁ despite uncertainty. Moreover, you act as if your level of confidence c in your hypothesis h₁ is greater than your true confidence level ᴦ to start to gather evidence to increase ᴦ. To be clear, in practicing SOWH, you must first acknowledge that uncertainty does indeed exist, and also that it is generally useful to communicate your level of uncertainty or confidence in your hypotheses. Nevertheless, for many endeavors, there simply isn’t enough information for anyone to form any kind of hypothesis in which they have more than a modicum of confidence. If every party involved in the decision-making process expressed their opinions without any conviction, you would often end up in a stalemate without any clear path forward.
To move forward, you try to quickly gather all of the evidence in a single direction, then “pound the table” with your opinion so that 1) you invite everyone to come up with all the reasons why h₁ may not be correct or may not work, and 2) you can quickly devise experiments to try to prove that h₁ is not correct. The central thesis of SOWH is that action is usually the best way to gain more information, and action based upon a strong opinion brings about the best information.
So what about the “weakly held” part?
While gathering evidence in support of your hypothesis h₁, it can be tricky to collect and judge evidence objectively— after all, we’re only human, and we tend to prefer to see evidence that confirms our views³. But the trick here is that we’re not looking for evidence to necessarily confirm h₁ — in most cases, proving h₁ right is a long process that requires lots of evidence-gathering. What we’re really looking for is evidence to reject h₁ so that we can backtrack and formulate a new hypothesis with additional data. Backtracking a bit, let’s define what we mean by backtracking in the context of SOWH.
Let’s go back to the “what to study in school” example, and say that you’re a college freshman and you aren’t really sure what you want to study. You like math, you like physics, and you like computers. You have an inkling that it might be cool to be a theoretical physicist, but you aren’t sure what that encompasses. You could just take a broad range of classes to get a general feel of what you like or don’t like. Or by practicing SOWH, you proceed with the assumption that you’re going to study theoretical physics until proven wrong, so you take an introductory physics seminar, take the prerequisites for a theoretical physics class, or try to do research with a professor in that department. After a few classes, you may find that you actually can’t stand, well, how theoretical it all is. Now you’ve explored and ruled out one possibility definitively, and you can backtrack, but that doesn’t mean you have to start from scratch. You really enjoyed some aspects of those physics classes, especially running the experiments involving computer simulations, so maybe your next hypothesis is that you want to do experimental physics.
If you were to walk into an introductory computer science class, you would probably hear the terms “breadth-first search” (BFS) and “depth-first search” (DFS). They’re both graph traversal algorithms. In the former, you search all neighboring nodes at the same depth before moving onto the next level, whereas in the latter, you traverse down a single branch until you either find what you were looking for or are forced to backtrack because your current branch cannot possibly have what you were looking for.
Think of your hypothesis h₁ as the first branch you traverse in the graph—you might get lucky and find what you were looking for without exploring any other branches. But more likely, at some point you’ll hit a dead end and need to backtrack. As soon as we see enough evidence that contradicts h₁, we need to backtrack. Sometimes you’re able to immediately prove your hypothesis wrong, and other times you may traverse down what seems like a really promising path only to have it turn out to be a dead end. Sometimes, you only need to backtrack a little bit, and other times you may need to backtrack all the way back to the root and start all over. Practicing the weakly held part of SOWH means that you’re just treating truth-finding as a graph algorithm— algorithms don’t feel like their egos are hurt when they need to backtrack⁴.
When you search breadth-first, you may gather lots of evidence, but it’s likely the evidence will be orthogonal or worse yet, contradictory. Let’s say you go out to talk to users for your startup’s product. You decide to talk to five users matching very different user profiles for 20 minutes each. You’ll likely find that they all have different use cases, get different value from your product, and therefore want different features. Or you could spend a few hours with a single power user, build a prototype for her to try solve her problem really, really well, get more feedback from her, and then continue to iterate on solving her biggest problem. Now, it may turn out that the job that user wants to hire your product for isn’t a good fit, or that there aren’t many other users like her. But until you find that evidence, you proceed assuming that that user’s needs represent exactly the power users you want to build for. The key insight here is that as soon as we accumulate enough evidence that a path is not going to yield the correct solution, we must immediately backtrack. If we stay attached to our initial strong opinion or continue looking for evidence to prove h₁ even though there’s a lot more evidence to reject h₁, we’re not practicing SOWH, we’re just setting ourselves up to be wrong more often than not.
That all sounds great in theory, but in the real world, unlike a true graph search, we rarely have the luxury of collecting enough evidence to show with 100% certainty that we should reject h₁ and that we need to backtrack, so how do we know when to revise our initial strong opinion? And is this better than just exploring breadth-first or gathering evidence differently before formulating our initial hypothesis? Why do we need to form a strong opinion at all? To answer those questions, I’m going to explore three examples from three seemingly very different games: poker, basketball, and Starcraft in Part 2.
- From most sources I could find pertaining to its origin, “Strong Opinions, Weakly Held” was first coined by Paul Saffo.
- We’ll use opinion, hypothesis, and bet somewhat interchangeably for the purposes of this essay. They’re not necessarily the same thing in most contexts, but an opinion within SOWH should be akin to your alternative hypothesis. I apologize in advance to anyone from a more scientific or statistical background if I misused any borrowed terminology here. Please do correct me if I’ve done so!
- I apologize to every teacher and professor I’ve ever had for all these links to Wikipedia, but as a primer for most of these concepts, I think Wikipedia is a great place to start. Just don’t cite it in your papers 😂
- At least none of the algorithms I’ve ever heard of. Although maybe neural networks will someday develop egos as part of their cost functions? Gradient descent into depression and anger