Everything Is Obvious Once You Know The Answer
“ How common sense fails us ”
by Duncan J. Watts
- Goodreads
- Rating: 5 / 5
I read this book years ago before I started taking notes. Back then this opened my eyes on a number of biases, the unreliability of common sense and the necessity of data. I’m not making it justice here ; it’s a great read ot help seeing the world differently, I strongly recommend.
summary
Judge on intent rather than on results. Base your decisions on data, not common sense. Stay adaptable.
Common sense is referred to only as a positive thing. It’s indispensable for day to day life. But for deciding policy or strategy in any complex systems, it’s highly unreliable.
That what is self-evident to one person can be seen as silly by another should give us pause about the reliability of common sense as a basis for understanding the world.
~ Duncan J Watts, Everything is obvious
We want a reason for things to happen. We want to explain good results with positive attributes, and bad results with negative attributes. In truth context and luck are much more important that we want them to be.
When inspecting in hindsight we tend to have circular reasoning: looking at the attributes of the successful thing and using this as the justification for its success. Harry Potter is successful because it melts magic with young adults with fairy tale with fantasy. In other words: Harry Potter is successful because it has all the attributes of Harry Potter.
Decouple analysis of decisions and attributes from the actual result: explaining something from the vantage viewpoint of hindsight is highly prone to bias, successes tend to be post-rationalized to good decisions and reciprocally ; while in reality a lot of this is due to context. We can only describe, not explain. We only look at what happened, not what could have happened. Instead of judging the results, look at the actions and the actual attributes.
Future is unpredictable and complex systems have a proven irreductible portion of randomness. Complex prediction systems consistently only marginally outperform the simplest ones. Instead of investing in better predictions, invest in better contingency plans and adaptability.
Success brings cumulative advantage - something successful will easily be more successful. The reason why the Mona Lisa is the most famous painting in the world is a) because it’s fine, b) because it’s already the most famous painting in the world.
notes
We are oblivious to our own fallacies. It’s easy to accept other’s logical contradictions, but we feel like they don’t apply to ourselves.
common sense
We live in a set of unspoken rules, that are sometimes hard to describe. Common sense is the set of practical knowledge that we apply in particular situations without questioning where that knowledge is coming from. A lot of common sense is contradictory (e.g. far from eyes is far from heart, vs. with distance the heart grows fonder). Something that is common sense for someone might seem stupid for someone else.
That what is self-evident to one person can be seen as silly by another should give us pause about the reliability of common sense as a basis for understanding the world.
~ Duncan J Watts, Everything is obvious
Since common sense works so well to solve day to day problems, it only gets positive feedback and we want more of it.
However it’s dangerours when used to solve more complex problems (e.g. politics), since it doesn’t rely on data but only intuition.
thinking about thinking
We are bad at understanding history. We look back mostly at large events (big successes and big failures, not triviality). We look back at the event and its actual outcome, rather than more holistically at what could have happened. We attribute too much of the outcome to a certain person, or a certain action, rather than also looking at the context.
When trying to understand actions from someone which look silly to us, we should try to understand their incentives. Actions are often rational once we understand the rationales.
Incentives are a good example of common sense that fails often. Paying employees more does not make them work harder. Incentives plans often don’t work as expected, such as employees effectively being demotivated after receiving poor bonuses rather than working harder to get good ones. Incentives also create bad side effects, over optimizing for what gives good returns. Yet common sense dictates that the way out is designing better plans, discarding that this was already the goal in the first place.
The problem gets worse once the plan works. After the fact analysis concludes that the components of the plan were precisely those that were required, and could have been found at the very beginning provided that enough thinking and common sense was applied
circular reasoning
Mona Lisa is the most popular painting in the world because it has all the attributes of the Mona Lisa. This is an example of how common sense is looking for intrinsic attributes to explain the success of something, while the reason why something is successful is also thoroughly contextual
Given that we can’t run experiments and validate predictions, we can only describe, not explain.
We also don’t look at what did not happen. For example, attributes contributing to a crash can result in a list of factors where thousands of other flights also qualify, but did not crash. For each kid who does a shooting, thousands have the same social profile and don’t shoot anyone. Same for terrorist profiling.
wisdom of crowds
Something at the macro level cannot be trully explained until we understand it at the micro level as well. For example, psychology can certainly not be understood by only looking at neurons and connections, but understanding the chemistry and underlying of the brain is critical to explain why people are acting a certain way that doesn’t seem rational. This is the micro macro problem.
In sociology this takes the form of methodological individualism vs representative individual. Representative individual is what’s used in common sense. It’s a whole used as a shortcut to explain the individuals. E.g.: families, poor, rich, mob, etc. Methodological individualism claims that the behavior of the whole isn’t explained till we understand behavior at the individual level.
This compounds with after the fact reasoning, where we try to explain through common sense the outcome of a social gathering by providing general qualities to the group, disregarding the individual. Example: a 100 person mob is composed of people with different thresholds to start violence. Each person has a different threshold between 0 and 99, being the number of people they need to see act violently before they start themselves. This mob would act entirely violently. Compare to another one exactly the same, but there’s no 2, and instead we have 2 3s. This mob would have 2 violent people. An outside observer would qualify the first mob as violent, the second as ordonate, while they in fact differ only by one individual. This illustrates how generalizing to the group based on after the fact observation doesn’t provide useful data.
Common sense is trying to explain success through intrinsic value, while in fact the social context plays a crucial role. Accumulated success is that something already successful will become even more successful. Back to the Mona Lisa, the reason why it’s popular is mostly that it’s already popular. Science experiment creating separate world with same music but isolated rankings yields different rankings per worlds. Most popular are never last and worst are never first, but there are significant disparities between worlds.
In other words we assume that things are successful because of a few critical attributes, but in reality no-one can tell why something is. Facebook didn’t get successful because people were waiting for it, but because a few people started using it, then drew a few more people, and so on.
special people
Common sense tends to find a few special people to explain things. A particular politician for economic crisis, a particular military leader for winning a war, a particular influencer for new trends. This is particularly a consequence of common sense where we prefer having a specific person to blame for problems or admire for success, and to whom we confer special attributes like charisma, anger, strength, determination, etc. In reality this is also highly context dependent, and if history could be replaced, some of the heroes would be replaced by others, because there ought to be one.
history
History is looking at things from after the fact. Even when trying to remain factual, this vantage point is making it hard to actually explain, because a) it concentrates on what did happen, neglecting everything that did not happen b) it focuses on interesting facts, rather than everything benign and standard c) a story better told is more credible d) decision quality is often confused with outcome.
On that last point, outcome is hard to determine because history never ends. Company performance for example is largely attributed to ceo, and the same person who gets praises one year will be lynched the next depending on the stock performance. This is because of the standpoint that we know what the outcome was, decide that it must be because of decisions, and attribute the performance to those decisions. But then if the year after performance recovers, maybe it’s because of those decisions, or maybe because of the context. Since history can’t be repeated there’s no way to tell.
predictions
We’re making a lot of predictions but rarely taking accountability for these.
We’re bad at understanding what we can predict. Simple and repeatable systems can be predicted. For example, when a comet will fly by, how to get to Mars.
The more complex the system gets, the harder it gets to make predictions. On repeatable systems, it’s possible to determine a reliability. For example, weather is relatively reliable: chance or rain with 60% confidence predicts rain 60% of the time. However we want to know whether it is going to rain, and when receiving such a forecast and seeing no rain, common sense dictates that prediction did not work, while it just falls on the 40% of the time where prediction was that it would not rain.
Unrepeatable systems, of which social systems are largely a part of, are extremely complex to predict. When the outcome of an election is determined to a certain confidence (e.g. Obama predicted at 90%), the confidence doesn’t mean much, since we can’t run the election again.
It’s also hard to know what to predict. We label major events by their biggest cause but they’re the product of much more. The French revolution is not just the Bastille, Katrina wasn’t the largest hurricane on that year, but the destruction, the social tensions, displacements that resulted made it what we call Katrina. Since we look at happened and not what didn’t happen, we also miss the account for all predictions that did not happen.
best laid plans
So what can we predict?
We can’t predict that we’ll catch the flu, but through aggregates we can predict a probability that we will get it. Same for credit risk, or general probability of a book to succeed. However the particular probability that a specific book will succeed is harder to predict.
For complex events, aggregating the wisdom of crowds give reasonable predictions. They get better with money put on the table.
Experimentation shows that complex events are highly subject to diminishing returns. That is, the difference in success rate of the simplest prediction model and the most sophisticated one is usually marginal, incremental at best (a few %). Examples are provided for football and baseball which are prone to statistical analysis, but the increment of complex analysis over simple prediction based on a) recent performance and b) considering that home teams win 60% of the time is 4%. Looking at a dice with a microscope will not increase the quality of predictions much.
This tends to suggest that complex event behave like random events, and so there is a maximum threshold for predictability
strategy
A strategy that fails in one setting works in another. The iPhone worked only because the internet turned out to become popular in the year it came out. The minidisc failed for the same reason. The fact that it was a good strategy or a poor one can be decided only from the ventagw point of history.
Instead of putting too much on strategy, solution is to develop multiple strategies, and keep adapting as story unfold. Decentralizing decision by making it local (eg market to allocate carbon emissions, prize research, etc.)
Make sure that data doesn’t get clouded by common sense. That plans don’t get stopped by people making common sense judgement
justice and judgement
Judging acts based on results is complex. For example a drunk driver who killed people and ruin lives may be sentenced with life in prison. Such a person will be called a criminal. But a difference of a few seconds might mean a close call, not killing anyone and not even being caught. Yet the same act will be judged in a much harsher way in the first case than in the second, just because the conclusion is different. If the intention is justice, not revenge, the philosophical question is much deeper than just “they killed someone they deserve jail”. The reason why the act, the behavior, requires jail time in one case but not the other needs to be analyzed with more than just common sense application of justice. It is a demonstration that only referring to common sense by analyzing acts through their impact rather than their intent yields to different conclusions
Similarly, judging on performance yields bad results. The same CEO can be called a genius for their strategy one year and lynched for their bad decisions the next, even though the only thing that changed is unpredictable changing context. Same goes for everyone, where success is attributed to all people part of a successful team (halo effect), and failure must be because they did something wrong. The solution to that is to judge on actual attributes, intent, and demonstrated process.
Accumulated advantage also makes judgement harder.
- A researcher who gets onto a successful research early on, or associated to it, will draw more funding, and therefore more results. They will also tend to get more prominent positions in the resulting papers even when their participation is low (see Find your why). A similar person whose research didn’t succeed will not benefit from such advantages. Yet looking at them we will judge that one is competent and the other not as much.
- Products built by successful companies tend to be more successful, not by their attributes, but by the sheer fact that the company already benefits from notoriety. Yet its success will largely be attributed to attributes rather than context. Their CEO will also be judged as visionary or not based on the company’s success. We need to attribute success to something. Not to say that such geniuses don’t exist, but they’re rather the exception than the rule.
- Successful songs released by famous singers will be judged a special, judge based on the fact that they’re successful.
Compensation is also hard to determine.
- Traditional packages will reward good results rather than performance itself (even if it’s usually called “performance”). Yet those results are largely dependent on context. It’s also very hard to determine what constitutes good results because there is no finish line.
- The same person can be very successful one year and very bad the next for things beyond their control (one of their customers going bust, or its CEO marrying the competitor’s CEO, …), by applying the same strategy and effort. Determining compensation based only on results is then rewarding luck rather than actual performance.
- Finance is largely prone to this, where results are largely dependent on unpredictable factors and luck. There are countless examples of brokers beating the market several years in a row, even for decades, and then loosing so much that overall they lost to the market. Yet judging them during their success streak we will decide that they have a knack, then judge harshly their demise, even if the strategy remained the same all along. - Dogma is especially prone to that. Financial sector is famously against taxation, and judging to be deservent of good compensation in times of success. Then during crisis rely on public money for bailout. Then once bailout has been reimburse debate the justice of having to pay interest and new regulations. Yet looking back at when they were having a choice between brankrupcy and survival, they would probably agree to conditions. The solution for that is to decide a clear philosophy for public insurance and pre-agree on conditions rather than relying on common sense application after the fact.