Bad decisions in life arise from having optimized for the wrong metric

Something I have noticed:

• Unfulfillment arises from using metrics chosen by others as proxies for personal success (e.g. a new car as a proxy for happiness).

• Sustainability problems arise from using short-term success as a proxy for long-term one (e.g. boosting sales this quarter by delivering false promises to customers).

• Failed relationships arise from choosing a person using widely-desirable traits as a proxy for personally-desirable ones.

In general,

Proxy optimization

I once wrote:

“Humans are extremely good at succeeding at their priorities, and extremely dishonest about them”.

With that, I meant that we are great at acting in such a way to succeed at the metric we chose, but often choose the wrong one or do not realize we subconsciously chose another one. I call this phenomenon commitment to failure.

Every time we fail at something we had the resources to succeed at, it is due to the fact that we considered an internal success to fail at it (because we were uncomfortable with the consequences of succeeding at it). In other words, we acted optimizing an internal metric (Expected Emotional Outcome) instead of an external one (success at whatever activity we were doing).¹

Even when we choose a proxy that is initially correlated with the ultimate outcome we desire, we often fail at misunderstanding a key concept:²

As an example, imagine an author in the business of selling books online. He discovers that about 2% of his Twitter followers bought his book. He decides to run an advertisement to get more followers (the number of followers becomes a proxy for book sales). Such ad selects followers based on their sensitivity to advertisements, not on their propensity to buy. In other words, it tends to select people who are interested in following the author on Twitter, regardless of whether they are interested in buying his books. As a result, if before the advertisement 2% of followers would end up as customers, after the ad it it is likely on 1% will. The proxy optimization (running an ad optimized to get new followers) diluted the correlation between the proxy and the ultimate metric (the % of followers buying the book).

A useful heuristic: if there is a way you could maximize the proxy without contributing to your ultimate goal, it will happen. Proxies are not chosen because they are more impactful; they are chosen because they are easier to measure or to influence. And the reason they are easier to influence, is generally because there is a population of low-hanging fruits which contribute to the proxy but not to the ultimate metric. Collecting these is a waste: they do not do anything to improve your standing towards your ultimate goal. Optimizing for a proxy often brings a dilution in quality.

Proxies promote unwanted behavior by introducing unwanted rewards towards wasteful actions. It is the same thing that happens when you are not specific in rewarding someone. Let’s say that you told your son “Congratulations for having gotten an A on the test!”, without knowing that he cheated. By rewarding a proxy (the mark) instead of the real thing (the study), you would have reinforced an unwanted behavior (the cheating).

Define success without using proxies

Naval Ravikant said: “A cockroach is just as evolved as we are, just across different fitness functions.” What metric you choose to define success, i.e. your fitness function, matters. It will apply evolutionary pressure on your behaviors and personality traits and shape you, by reinforcing those traits that led to a progress on your chosen metric and by weakening those who did not. Choose your proxies wisely, don’t become a cockroach.

Another reason why proxies should not be used, is because circumstances change over time. It might sound useful to choose a proxy to focus on what appears to be the most beneficial sub-objective now, but this might lead to tunnel vision or to a lack of periodic reevaluation to determine whether the current proxy is still beneficial to follow. What got you here won’t get you there³. Memorizing concepts might be a useful skill to get your degree, but it is a terrible one for a successful career (other than in acting). Money is great to increase your life quality when you’re poor, but it doesn’t do much once you’re rich. What looks a great proxy now will almost invariably be a bad proxy then.

A last reason not to use proxies: they bring noise (because they are partially decoupled from the ultimate metric which we do want to measure) and they bring cognitive dissonance (for the same reason; if we ever take an action which is good from the point of view of the proxy but not of the ultimate metric, we’ll end up asking ourselves “Why did I did it?”.

Following proxies instead of the real thing is similar to drug addiction. Addicts become reactive to the cues (the proxy). Living a diverse life increases happiness because it avoids addiction and tolerance due to repetitively associating rewards to the same cues⁴.

(I am inclined to believe that humans are fundamentally good and happy; evil and unhappiness emerge when they follow unnatural proxies or when they encounter constructs or products engineered to hack metrics, such as advertisements and drugs, and become addicted to them).

Eliminating proxies

Prioritization is the art of spotting proxies and exclude them from our to do list. Proxies are fought with not-to-do lists. A good example is Warren Buffet’s 25–5 rule. Write down the 25 most important things you should do. Circle the 5 most important. Move these 5 to your to-do list, and the other 20 to your not-to-do list (these 20 tasks would be busywork, which, by the way, is a proxy for real work).

Intelligence is the skill of optimizing a given metric; wisdom, the skill of choosing the right one to optimize for.

Some heuristics on how to spot proxies and get rid of them:

• Everything which causes addiction is a proxy.

• Any metric which is not weighted by the impact it generates is a proxy. (And any metric whose correlation/impact to your ultimate goal is approximated to be static in time, is a proxy.)

• If a proxy is chosen because it is easier to improve, rather than easier to measure, don’t choose it.

• If a process describes what it is, rather than what it is for, it has been optimized for a proxy.

• If it won’t be important in five years, it is a proxy.

• If it weren’t important for people one thousand years ago, it is a proxy.

• Doubt everyone whose core competency is a proxy (students whose competency is to pass exams, which are a proxy for knowledge; teachers whose competency is to get papers published, which are a proxy for teacher quality; financial advisors whose competency is looking trustworthy; the salesman who’s great at presenting a slide deck but not at actually making sales happen; speakers whose competency is to use words rather than choosing them; and so on)⁵.

These heuristics won’t be always true, but you’ll be better off by acting like if they did (that’s the point of heuristics).

This was an excerpt from my book “100 Truths You Will Learn Too Late.”.

Many thanks to Damian Ocean for providing insights to the first drafts of this post, which appeared on Twitter.

Notes

Footnotes

1. Alexander Wolfe noted that “What people who don’t risk don’t understand is that it is better to risk and fail at something that resonates deep within than to do nothing”. To which I replied: “[That’s] the difference between external failure (proxy) and internal success (actual metric to consider).” ↩

2. This is not (only) a repetition of Goodhart’s law. Goodhart focused on people gaming the system once a regulator introduced a policy. My heuristic is more general and works in single-player scenarios too. It is because the easiest way to improve an unweighted metric is often to include items whose weight towards the ultimate goal is lower than the past (e.g.: in the example in the text, the easiest way to improve the number of followers is to attract those who won’t buy anything). ↩

3. As far as I know, Marshall Goldsmith made the expression famous, in his homonymous book. ↩

4. Living a diverse life means acknowledging that because of the problems of addiction and tolerance, pursuing a single proxy will yield diminishing returns over time and lead to dissatisfaction. ↩

5. Did you notice any similarity between this list and the contents of Nassim Nicholas Taleb’s Incerto? That’s because the Incerto is, too, about how to spot and get rid of proxies. Fooled By Randomness is about spurious correlations between proxies and real metrics; The Black Swan is about frequencies used as proxies for impact; Antifragile is about averages used as a proxy for distributions; Skin In The Game is about incentives (and proxies are incentives). ↩