Surfing the fitness landscape

A few notes I took reading Tim Harford’s Adapt, mingled with my own thinking.

Evolutionary “fitness landscape”

For an explanation of “fitness landscape”, see here

The fitness landscape is constantly changing, and it and the networked system that defines it are so complex and chaotic, that it’s impossible to predict the future or to answer what-if’s about history. (According to P. Romer, “An equally, if not more important stumbling block has been the deep philosophical resistance that humans feel toward the unavoidable logical consequence of assuming that genuinely new things can happen and could have happened at every date in the past. We are forced to admit that the world as we know it is the result of a long string of chance outcomes…. Once we admit that there is room for newness—that there are vastly more conceivable possibilities than realized outcomes—we must confront the fact that there is no special logic behind the world we inhabit, no particular justification for why things are the way they are. Any number of arbitrarily small perturbations along the way could have made the world as we know it turn out very differently.”)

But soothsayers and high-priced consultants are perennially popular. (Studies show we prefer advice from a confident source, even if the source has given us lousy advice in the past! Think for yourself, don’t follow anybody, because “Thinking the guy up ahead knows what he’s doing is the most dangerous religion there is.” — F. Wirtanen)

Changing the world means changing the rules

Most of us are small individuals, just a point in the distribution, so realistically we need to find good tactics for surfing the fitness landscape.

If you are in position with some impact on policy for your organization, you likewise need to find good tactics for it to surf its fitness landscape. The organization is just competing at a slightly higher scale than the individual. But a big difference from an individual is that an organization has some control over the local fitness landscape the individuals within it are surfing. A system, even of just a big company, let alone an entire economy, is too complex to command and control as if it were a machine. Manipulating its internal objective function is one of its major tactics for surfing at the higher-level it can’t really control.

How to best reshape the part of the fitness landscape you have some influence on? Change the objective function to make the peaks and valleys as consistent as possible with your objectives. This is necessarily done through leveraging proxy metrics and if they are not carefully chosen and measured, you will get a landscape filled with perversely deformed creatures that are optimized for the proxy metrics but don’t meet the intended objectives.

One such tactic recommended by many economists is to “tax bads, not goods” (externality) , such as making polluters pay for the damage they cause. That’s the idea behind a carbon tax. Likewise, there should be natural built-in reward for producing social benefit. No one is capable of predicting in advance the best way to achieve big objectives, but evolution coordinated by a market can get the job done in surprising ways, if the appropriate price signals are in place. Prescriptive (or proscriptive) mandates don’t work.

At any scale, you are not entirely limited to surfing, you can try to have some influence on the landscape you’re surfing, such as by lobbying governments for different rules.

Rules for intelligent evolution, climbing the fitness landscape

Try many survivable controlled experiments, both incremental and some crazy ideas, and heed the evidence.

Much of Harford’s book is how about challenging each of the pieces of that strategy is. There are many countervailing forces to intelligent evolution, especially social/psychological forces.

Hayek, gung ho, delegation, decentralization

The economist Hayek argued that the value of centralized knowledge was grossly overrated vs. ‘knowledge of the particular circumstance of time and place’.

The powers that be love to fantasize otherwise, and the social/psychological forces have usually let them get away with it, making others pay the price of their grandiosity or control-freakness. But some of the most successful modern organizations delegate/devolve/decentralize, so tomorrow may be different. Tribes of tribes outlearn, outexperiment and outcompete the old centralized T. rexes.

The biggest example of this in Adapt is the story of the US recent war in Iraq, so it’s interesting to consider that at the very time Hayek was writing his famous paper, Evans Carlson of the Raider Marines was demonstrating the same with his tactic of “gung ho”

The definition of failure

One of the most problematic shortcomings (in my opinion) of Harford is his accepting the antique definition of ‘failure’, because it’s contrary to everything else he’s saying. It’s critical to get straight on what we mean by to “fail”.

What’s a failed experiment? In science it’s an experiment that’s so badly controlled/designed, or so badly performed, that you can’t learn anything from it, including perhaps about whether someone else’s experimental results are reproducible. An experiment fails if it doesn’t give useful data, if it doesn’t make a useful distinction.

The measure of success/failure of a scientific experiment has nothing to do with your expectations or desires. And it’s generally more delightful to a scientist to be surprised anyway. According to Tina Seelig, “There’s a huge problem with the word failure. As a scientist, when I do an experiment that doesn’t work as I expected, what do I call it? Data. It’s not a failure. In fact, some of the most interesting scientific research comes from experiments that have unexpected results. The key is to look at the things that don’t come out as expected as data that provides interesting clues to what is really happening. If you are afraid of failure, you won’t try anything new.”

More generally, it’s ridiculous to be surfing a constantly changing fitness landscape with very limited information and to consider any outcome that isn’t consistent with your wishes a failure. If you don’t you apply your best effort and follow through and learn, then you’ve failed.

I’m having trouble expressing this, so I guess I can’t blame Harford. But it’s like with the market. You can’t predict the stock market and the financial market and the real estate market and the currency market and on and on, but if you have capital, it’s got to be somewhere. There’s no such thing as no choice, no such thing as doing nothing. And unless you use some arbitrary measuring stick (dollars, gold, wheat?) then you can’t even say for sure whether you’ve got more capital this week than last week, let alone predict what will happen in the future. How you define success and failure in that environment is very fuzzy.

Safe to fail

A lot of the book is about safety – safe to dissent, safe to be right, safe to tell the truth. “Jack Galvin also taught Petraeus that it is not enough to tolerate dissent: sometimes you have to demand it.” Confirmation bias, groupthink, and the rest of the usual suspects get in the way.

And safe to fail – in two ways 1) experiments that might not have results that conform to our wishes “It isn’t right to expect a Mario Capecchi to risk his career on a life-saving idea because the rest of don’t want to take a chance.” “The moral of the story is not that we should admire stubborn geniuses, although we should. It is that we shouldn’t require stubbornness as a quality in our geniuses. How many vital scientific or technological advances have foundered, not because their developers lacked insight, but because they simply didn’t have Mario Capecchi’s extraordinarily defiant character?” 2) decouple the system so that a little change in one place doesn’t set off a catastrophic cascade.

Lottery tickets, but better

A recent example of supporting crazy ideas is here.

Harford notes correctly that “lottery ticket” is a poor metaphor for the idea, because lotteries are a zero-sum game with an obvious expected value. So the metaphor he uses throughout is the Spitfire which helped win the Battle of Britain and had a very unconventional and inexpensive birth story. The idea is that organizations and societies should fund lots of diverse experiments, because maybe a few (and we can’t predict which) will help us win big when the fitness landscape changes. As they say, “Dig your well before you’re thirsty.”, but water is changing course underground, so you better dig a lot of wells.

Thinking can’t replace action

Theory is important, but as a rule you learn more from many pilots and controlled experiments than you do from theorizing.