Domain Experts vs. Career Politicians

The case is easy to make and easy to make badly. Here it is with the three real objections attached.

The mismatch

A member of parliament is asked, in the space of one session, to have a view on aircraft certification, antimicrobial resistance, spectrum auctions, minimum support prices and the regulation of frontier AI models. No human being holds all of that. Nobody claims to. The system runs on the polite fiction that a mandate substitutes for knowing anything, and it is supported underneath by a civil service of generalists who rotate departments every few years, which means the institutional memory of a technical domain often lives in nobody at all.

Meanwhile there are people who have spent twenty years inside each of those subjects, who know precisely which rule is wrong and why, and whose entire route into the room runs through being consulted, at somebody else’s discretion, after the decision has taken shape.

That is the mismatch. Stated that plainly it sounds unanswerable.

It isn’t. There are three good objections, and the version of this argument worth holding is the one that has survived them.

Objection one: experts are bad at predicting things

This is the strongest empirical attack and most people advocating expert rule have never encountered it.

Philip Tetlock spent about twenty years collecting political forecasts from a few hundred specialists — academics, analysts, advisers — and scored them against outcomes. Roughly twenty-eight thousand predictions. The headline result, published in 2005, is that they performed barely better than chance, and that the most confident and most credentialed among them did worse than the moderately informed. The line everyone remembers is that the average expert was about as accurate as a dart-throwing chimpanzee.

If that is what expertise buys, the case for putting experts in charge collapses.

Except look at what was measured. Tetlock was scoring political forecasting: will this regime fall, will that country enter recession, will this conflict escalate. Predicting the future behaviour of a complex adaptive system is a task humans are reliably bad at, and being a professor of Soviet studies does not exempt you.

That is not the thing we are claiming a pilot has. Operating knowledge is not forecasting. A pilot cannot tell you the state of Indian aviation in 2035. A pilot can tell you why a particular go-around procedure produces confusion at a particular airport in monsoon conditions, because they have flown it. An anaesthetist cannot predict next year’s disease burden but can tell you exactly which line in a procurement rule will result in the wrong drug being on the shelf.

The distinction is sharp and it is operationally useful: we are recruiting for failure-mode knowledge, not for prophecy. Anyone whose pitch is that they can see what’s coming should be treated with the suspicion Tetlock’s data earned them. Anyone who can tell you the six ways the thing they have run breaks, in order of frequency, has something no election confers.

It also implies a limit we should say out loud. Expertise licenses you to describe mechanisms and their failure modes. It does not license you to make confident calls about outcomes years out, and a lab full of people who forget that will produce the same overconfidence in a better vocabulary.

Objection two: experts get captured

Also true, also well documented, and structurally more dangerous than the first.

George Stigler’s account of regulatory capture, published in 1971, is the standard reference: regulation tends over time to be acquired by the industry it governs and operated for its benefit. The mechanism is not bribery. It is that the regulator and the regulated are drawn from the same small pool of people who understand the subject, attend the same conferences, and will likely work for each other at some point. Everything that makes someone qualified also makes them entangled.

The 737 MAX is the case study nobody in this debate should be allowed to skip. The FAA had for years delegated much of the certification work to Boeing employees under its Organization Designation Authorization programme — expert review, in the most literal sense, by people who knew the aircraft better than any regulator could. MCAS went through it. Lion Air 610 went down in October 2018 and Ethiopian 302 in March 2019, and three hundred and forty-six people died.

Domain expertise did not fail there. Domain expertise was present in the room and was structurally unable to act, because the experts’ employer was the applicant.

So “put experts in charge” is an incomplete instruction. The complete one is: put experts in charge and assume they will be captured, and build for that in advance. Published conflicts. Fixed terms with no renewal. Reasoning published alongside every decision, so that a captured decision has to be argued for in public where the people who know the domain can see it. Rotation out on a fixed clock, so nobody sits in one seat long enough to become worth buying.

If we ever propose an expert body without a staging plan, we have failed our own test.

Objection three: nobody elected them

The oldest one, and the reason technocracy has an unpleasant history.

An expert body deriving authority from correctness has no answer when it is wrong, and no answer when the public disagrees with something it got right. Correctness is not consent. Every attempt to run a country on the first without the second has ended somewhere ugly, and it has done so with enough regularity that it should be read as a property of the design rather than as bad luck.

Which is why the first two words are not decoration. Democratically Elected System Hackers. We go through the front door: contest elections, accept the result, hold office by consent or not at all. Expertise determines what gets proposed and how proposals are examined. It does not determine who holds power. Those are separable, and confusing them is the whole failure mode of every technocratic experiment worth learning from.

Legitimacy is not a constraint on the rocket. Legitimacy is the launchpad.

What is actually being proposed

Narrower than it first sounds, and better for it.

Not rule by experts. A hard requirement that anyone proposing a rule for a domain has operated in it, that their reasoning is published in a form specialists can attack, and that they leave before they accumulate the mass that makes them worth capturing.

The Challenger inquiry produced the cleanest statement of why any of this matters. Thiokol’s engineers had told management the O-rings would not seal in the cold, and were overruled by people with authority and no relevant knowledge. Feynman put the finding in one line: reality must take precedence over public relations, for nature cannot be fooled.

Nature cannot be fooled. Aircraft, epidemics, power grids and irrigation systems are not persuaded by mandates. They respond only to whether the person who wrote the rule understood the machine.

Tell us where this is wrong.

desh>_