Policy as Computation

Definition of terms

There are four types of collective decisions:

1. There exists a convincing model/argument that supports it and no convincing model/argument opposing it exists. For instance, mass vaccination against smallpox. Before mass vaccination was adopted and the disease was eradicated, all we had were clinical trials, a model supporting the mass vaccination policy, and no serious objections to it.

2. No convincing model/argument supporting it and a convincing model/argument opposing it. For our purpose, this case is identical to 1., for the decision of not pursuing a policy.

3. A convincing model/argument supporting and a convincing model/argument against it.

4. No convincing model/argument supporting it and no convincing model/argument against it.

Let’s call 1. and 2. operational questions, and 3. and 4. policy questions. The term operational is commonly opposed to strategic (decision, once committed to, are hard/costly/lengthy to change). Here, policy questions are a subset of the strategic; they have additional properties. A model is rational (optimization of a stated, measurable objective). An argument is cultural/ethical (satisfies a stated non-measurable norm).

Moreover, the term “convincing” is doing a lot of heavy lifting here, but let’s say that the question of what is convincing is solved through the usual means (pure Aristotelian logic, peer-review, a vote, whatever the king say is convincing, which is best being itself a higher order policy question).

Operational questions can become policy questions and vice-versa, as new knowledge is becomes available. Policy questions are computationally irreducible

You can identify policy questions by the fact that the extremes are the most comfortable and commonly held opinions on them, because computing and explaining the nuances is too complex. You don’t have this for vaccines for instance. Being an absolute antivaxxer or unequivocally accepting all vaccines are fringe opinions at best, but the most intellectually comfortable, and easily communicable position on, say, academic research, is that it is either all useless or all equally important and valuable.

In newly independent Algeria, it was decided that what the country needed most were engineers and technicians. We ended up with excellent scientific and technical Universities, at the expense of humanities, social sciences, management. We now have a country that barely knows itself and what to do with its engineers.

Questions in categories 3. and 4. are typically there because they are computationally irreducible. The best way to answer them is to run the experiment, using the real world as a computer, over a long period of time. Before that the answer is a matter of faith.

Policy as a long-term computational experiment

What’s important with policy questions is that a lot of experiments are ran in parallel. So that within one lifetime a man can move to a municipality/country/continent where other experiments that are more aligned with his faith are ran. Reality is the computer model here.

Even more importantly, the policies/computations need to be revised if it turns out that they didn’t work. And here is the critical point: the experiments are sometimes ran for so long that people forget that it was an experiment and get attached to it. Or people try to force being proven right. Sunk cost fallacy, etc. Many reasons why people resist discontinuing a failed policy/experiment.

Also some problems with democratic elections: Mandates are not long enough to fully run the experiments and actually benefit from the actual point of the exercise.

The answers to policy questions often come in vague packages (red and blue team, with just an evocation of what the actual policy questions they are interested in, and you fill in the blanks about what their actual policy answers might be, taken to an extreme, this gives “Obama is nice and for peace because he is black, Nobel prize!”).

Political platforms often come in the form of answers to old questions (Communism vs Capitalism, a majority of Algerian politicians running on anticolonial platforms, 60 years after decolonization).

Questions can be artificially promoted from operational to political. E.g. Sarkozy let French suburbs burn in 2005 to get elected in 2006.

Answers to policy questions in the form known as “politics” manage to be simultaneously too short-termist, too vague, too permanent, and often times plain inadequate, when what is needed is simply running enough experiments over a long-enough time-horizon.