Diagnosing government incapacity: alignment, dysregulation, capability

Now seems like a moment of reflection who work on some form of “improving government.” I see a lot of people wrestling with ideas and theories of change and where we are now. There is a lot of Discourse in and around “civic tech”, DOGE (both what it has done and what it implies), “Abundance” and “state capacity.”

I think good faith wrestling with reality is always useful. If you never mark your ideas to market, proverbially, I’m not sure how effective you can be in the long run.

So I want to share a bit of where my mental model is today around this big cluster of things.

Any attempt to improve government capacity is predicated on a theory of the constraints on said capacity. Sometimes this theory is never articulated explicitly, and I believe that leads to unclear and less effective thinking.

I’ve come to a conceptual model of state (in)capacity with roughly three distinct sources. This is informed by a deep conviction that to understand government one must view it as a complex system of many decentralized moving pieces and recurring feedback loops, rather than a monolith.

A complex systems lens not only defines “government” as having multiple dimensions, it also suggests that sources of feedback loops (in politics/governance, in civil society, in culture at large) are endogenous and a part of the system, rather than separate.

(A concrete example of this: I view the private sector interests that comprise the political economy of SNAP as a part of “government” in the sense of it as a system one might seek to influence. This is the sort of subtlety that seems obvious to any practitioner, but seems to me often elided in theories of change that focus more narrowly on, e.g. a particular government agency.)

The three distinct buckets of state capacity constraints I think about today are:

1. Alignment problems

2. Dysregulation problems

3. Capability problems

If you know me, you know I spend much of my time empirically poking at AI models. “Alignment” is a robust research topic in AI that — quite roughly — can be thought of as the problem of aligning the system to human values and goals.

I think this is a useful lens because, candidly, much of what I see in the government capacity world is in fact a fairly straightforward disagreement on goals or objectives.

This is indeed one form of misalignment! But a problem is when people confuse disagreeing with the goal the state has been directed to pursue with the ability of that state to execute it.

I go out of my way to make this clear because I genuinely think attempts to change goals through pure capacity work are misguided. There can be more philosophical objections, too, here around democracy and legitimacy.

I also find the inverse is often taken to be quite obvious: it’s generally viewed as bad when an attempt is made to stop a passed-policy from working by hampering state capacity. The problem is it's really tempting, because it can feel like more within one’s reach. But I think the long term consequences are degradation of capacity overall.

My more direct perspective: If you want the government to pursue different goals, that is a policy (and politics) problem and that is the appropriate (legitimate) level at which to work it. It creates negative externalities to either try and pursue different goals through solely capacity increases or by attacking capacity.

There’s a different flavor of an alignment problem, which is when policies are simply not written in such a way that they map to the goals behind them.

I see this mostly clearly in incoherence: unimplementable policy. Or less sternly, policy that fails to make tradeoffs against goals that, in implementation, will necessarily be arbitrated and tradeoffs made.

In some ways, this is merely an accountability sink for the policymakers. Getting policy passed is messy — the sausage factory is real — and compromise is often necessary.

But when compromise leads not to a compromise of goals but rather to incoherence of the operationalization of those goals what you end up with is policymakers (legislators, executives) getting a perceived upside/win in the short term, followed by a downstream rug-pull when the incoherence is implemented and that is what people actually experience.

If the first version of an alignment constraint is more fundamentally political, this one is a bit more technical. Interventions likely live at the level of legislative staffers, and institutions like those that analyze policy on behalf of legislators; in California I think of the committee staff analyses, DOF, LAO.

An underrated part of this is that a new policy — well-aligned to goals — must necessarily grapple with all the existing policies that govern an implementing agency.

I heard of one example recently: of California’s major housing bill SB 79 requiring cleanup because the agencies are told to issue guidance on a timeline that the California’s Administrative Procedure Act makes infeasible.

There are often good feedback loops in the legislative process for budget changes that must accompany a new policy to make it work. But we, structurally, lack much beyond that — on the dimensions of HR, technology, pure process steps, etc.

(Aside: this one is where I have a bit of an extreme take, which is that most US-federal-government-wide policymaking is a bad idea. But it persists because that is the prestigious work to do. In reality, pushing decisions down as close as possible to the problem is better, but that’s not what people wanting to do Big Things set out to do!)

A different flavor of constraint on government capacity is what I call dysregulation problems.

What I mean here:

• Government is a complex system

• Complex systems are, in the long run, carved by recurring feedback loops

• When those recurring feedback loops are broken — or distorting vs. stated goals — the state capacity apparatus has bugs

A concrete example: I worked on unemployment insurance during the pandemic. One thing I heard from career folks in that world, when the bright attention was on them, was: this happens every time.

What did they mean? Something like: “Every time there’s a big recession (or other flashpoint on the UI system) we get a ton of attention and energy and then a year or two later it all goes out to sea, and we’re where we were before.”

And so they resisted newcomers pushing big changes. I don’t think they were wrong!

What would a breakout of this dysregulation have been? Likely, a change to funding of the UI system where — rather than dollars flooding in as soon as a recession hits (but with no time to hire up and train with those new dollars) — funding is more stable and provides resilience over time.

There are other versions of this: a really-existing recurring feedback loop in federal agencies that, if you have a bad outcome nothing really happens, but if you don’t follow process you have a GAO report written about you. (And people can be fired on the latter much more than the former.)

The point here is meaningful change is seeing the recurring feedback loops, changing them, and adding new ones.

(Note that this is quite different from a lot of prior efforts. It points less to things like hiring software developers and more to things like changing government IT audit standards.)

This is, in some ways, one I don’t have a lot to say about. Why? Because I — perhaps diverging! — think the underlying capability constraints of government fall downstream from the constraints above.

State capacity as capability is, to me, the object level. Yes, one must look at it, and likely work backwards from concrete capability problems. But it’s the rock to be carved, not the level of intervention.

Medium? I wouldn’t impose sweeping, High Modernist change predicated on it.

But these are the conceptual shapes I find myself coming back to when people talk about XYZ problem, particularly related to government’s technology (in)capacity.