26 Oct 2021

This is a work in progress. If you have feedback - comments, questions, or pointers to related stuff - please email me.

• Introduction
• The scientific method
  • The scientific method in the social sciences
• The state of government
• Scientific government
  • Figure of merit
  • Democratic selection of the figure of merit
  • Policies and experiments
  • Examples
• Implementing scientific government
  • The organizational structure of science
  • The structure of scientific government
  • How to implement scientific government
  • How can we get there from here?
• World government
• Possible outcomes
• Related work
• Feedback

Introduction

Governments around the world have failed us. They haven't adequately addressed our most dire problems: climate change and human overpopulation. Most people in the world are poor and/or miserable. In the U.S., democracy has been subverted by billionaires; systems designed 200 years ago have broken down in the presence of concentrated wealth and ubiquitous social media.

Suppose we could start from scratch and design a brand new kind of government. What might we do differently?

I argue for government based on ideas borrowed from science, namely:

• Scientific theories are judged on the basis of experiments. To the extent possible, government policies should be evaluated and selected using the same "scientific method". We should use policies that can be proven to work, rather than ones based on traditional beliefs and prejudices, or on the economic interests of a minority.
• Science has developed an organizational structure - a tripartite system of journals, research institutions, and funding agencies - that has done a good job of enabling and implementing the scientific method, and has been remarkably resistant to corruption. It's pretty close to a meritocracy. The organizational structure of government should be modeled on this - and in fact should be built on top of it.

Current government systems are based on choosing "leaders" who then select and implement policies. This is a flawed model - it's like having the Pope decide the laws of planetary motion. As Adam Grant points out, the worst people run for office.

In the system I propose, there are no leaders of this sort. Policy decisions are made on the basis of experimental results, not the intuition and prejudice of individuals.

I'm interested here in long-term government policies: taxes, health care, land use, environmental regulation, the penal system, maybe even small-scale things like where to put traffic lights. Government also has to deal with unexpected short-term issues and crises; for those we need competent leaders who can make good decisions. That's outside of the scope of this essay (though some of its ideas apply there as well).

The scientific method

The scientific method was developed to explain the physical world. It involves several related ideas:

• Instruments that make quantitative measurements, with known precision and error.
• Theories that try to explain measurements, typically mathematically. Theories are predictive.
• Experiments that use instruments to test theories. Theories are disproven if they fail to predict the result of an experiment. If there is no possible experiment that could disprove a theory, the theory has little value.

New theories may meet resistance - organized religion tries to suppress science that contradicts its belief systems; oil and tobacco companies fight science that threatens their profits. The scientific community itself can form internal power structures that suppress research that threatens dominant paradigms. The scientific method is designed to resist these pressures. Disproven theories must be discarded, no matter how entrenched and powerful their supporters. The truth - even if there's initially overwhelming opposition to it - eventually wins.

It's worth noting that:

• The scientific method is stable; it's existed with the same basic principles for hundreds or thousands of years.
• Science is the same in all cultures. There is no separate "Bolivian Physics" or "Hindu Chemistry".

The scientific method in the social sciences

The scientific method has been successful in a range of "hard science" domains: physics, chemistry, astronomy, engineering, biology, medicine, and so on. It has converged to universally accepted core theories in these areas.

In other areas - economics, psychology, sociology, humanities - there have been efforts to use the scientific method. The results have been less successful, because in the social domain it's hard to:

• make accurate quantitative measurements;
• design experiments that control for factors other than the one being studied;
• design experiments that don't harm human or animal subjects.

Using science in government will face these same difficulties.

The state of government

Governmental systems may include electoral, legislative, and executive structures. Governments define economic systems: regulation of industry and commerce, who owns what, taxes and distribution of wealth, access to natural resources, and so on. So by "government system" I include the economic system.

Many forms of government have been tried: various forms of democracy, oligarchy, dictatorship, monarchy. With a few exceptions, they've all failed in various ways.

They've chosen bad policies, with bad results. They haven't addressed the overwhelmingly most important issue: climate change (and the related issue of overpopulation). They've allowed an ever-increasing wealth gap. Most countries, including the U.S., have policies that maximize the short-term wealth of a few Plutocrats, at the expense of everyone else and of the future of the Earth. The majority of human beings lead unsatisfying and hopeless lives. Stress, anxiety, depression, and societal discord are rampant.

The systems themselves often are unstable - they often don't work as originally designed for more than a few decades. All communist governments have quickly been taken over by opportunists and transformed into dictatorships or oligarchies. In the U.S., corporations and billionaires have figured out how to buy the electoral process, and have created an oligarchy with the facade of a democracy.

The American 'founding fathers' did their best to define a stable system. But they couldn't anticipate that wealth would become so concentrated, that the negative effects of economic activity could be so extreme, and that popular information systems (e.g. Facebook) would become so pervasive and susceptible to corruption.

Governments and the processes behind them - even in democracies - often damage societies. This is particularly acute in the U.S., where the right wing has adopted a strategy of manipulating their supporters a) to hate people of other races, other countries, and other opinions; b) to distrust science, and to be unable to distinguish facts from lies; c) to put personal short-term interests ahead of everything else. America is an angry, unhappy place. I think that scientific government can change this.

Scientific government

Can we borrow and adapt ideas from science to create governmental systems that work better - that make people happier, and that adapt to changes in the world?

Science has theories, government has policies. In the same way that we can use (objective, quantitative) experiments to evaluate theories, we can (in many cases) use experiments to evaluate policies. The general idea:

• Define a "figure of merit" - a quantitative measurement of how well government policies as a whole are working.
• Develop alternative policies designed to increase the figure of merit.
• Perform experiments that test these policies - i.e. that see how they affect the figure of merit.
• Based on experimental results, use the policies that work best.

My thinking about this was inspired by the documentary Sex, Drugs, and Democracy, which describes Holland's approach to social issues such as sex education, drug use and prostitution. They identified figures of merit: the rates of drug abuse, crime, teen pregnancy, STDs, violence against women, poverty, and so on. They experimented with novel policies, such as legalizing and regulating drug use instead of criminalizing it. They found that these policies greatly improved the figures of merit; for many of them, Holland is the best in the world. Sadly, few other countries learned from this success story; American social policies are still rooted in Old Testament principles of punishment and revenge, and they often exacerbate the problems they try to solve.

Inspired by Holland's example, I propose a new form of government - "Scientific Government" - which uses scientific principles to find effective policies. In the U.S., Scientific Government would largely replace the executive and legislative branches, at the national, state, and local levels.

Figure of merit

At the core of Scientific Government is a "figure of merit" M(t) - a single quantitative time-varying measure of how well governmental policies are working. M(t) might include components such as

• Health (physical and mental).
• Happiness.
• Poverty, and wealth distribution in general.
• Crime.
• The health of the environment (for humans and other species).

The definition of M(t) must specify precisely how each component is measured. For example, for happiness we'd need to specify

• which people are included (perhaps only people above a certain age);
• how the happiness of an individual is measured;
• how these numbers are aggregated: probably a robust statistic like median or RMS, rather than the mean.

We then must specify how these components are combined; some sort of weighted average, again with a robustness mechanism. We never try to optimize its components separately - e.g. a repressive policy might reduce crime to zero but decrease happiness.

It could be argued that M(t) should have only one component, happiness. Perhaps the other components are all reflected in this, and measuring them separately is counter-productive. This is possible, but I suspect that measuring a subjective quantity like happiness is noisy compared to other components, and would make it difficult to perform experiments.

The "figure of merit" idea is inspired by the gross national happiness index concept articulated by Bhutan's King Jigme Singye Wangchuck in 1972. He proposed that policies should be evaluated based on this index, but this was not done in a systematic way.

Democratic selection of the figure of merit

An ideal government gives people what they want, and doesn't tell them what to want. So it must include a democractic component.

In the current U.S. government, the democratic component involves voting for candidates and for ballot measures (i.e. policies). Both of these are fundamentally flawed. Voting for candidates devolves into identity politics and demagoguery; voting for policies perpetuates bad policies.

In Scientific Government, the democratic component is the selection of M(t): it's decided via periodic popular elections. There must be constraints on how M(t) is defined:

• The definition must be simple enough that the average person can understand it.
• Policies can't be directly included in it. For example, we can't have "availability of guns" as a component of M(t). If gun proponents think that guns make people safer, they can propose weighting safety more heavily in M(t). Experiments would then decide how changing the availability of guns affects safety, and hence M(t).

Any society will inevitably have disagreement. One of the goals of scientific government is to move this to the highest level - well-defined differences of opinion about what society should be - rather than personalities, identity politics, and propaganda.

Policies and experiments

Once M(t) has been established, the general flow of scientific government is:

• People (or groups, or companies) can propose a new policy P.
• A government agency made up of scientists (see below) decides whether it is feasible that P could significantly increase M(t), and whether it is possible to design an experiment to prove this.
• If so, the government carries out this experiment.
• If the experiment shows that P increases M(t), the policy is adopted (possibly replacing existing policies).

If other countries or societies have adopted P, it may not be necessary to do an experiment; it may be possible to estimate P's effect on M(t) from existing data.

There are a number of potential problems in doing social experiments:

• The experiment's subjects may have a prejudice about P.
• It may be hard to control other factors.
• The experiment may create hardship for some of the subjects. But letting people opt out of experiments would undermine them. Scientific government needs global buy-in; people need to be willing to make sacrifices for the common and future good.
• Some policies may not impact M(t) right away (or may impact it negatively), but will increase it in 20 years. How to evaluate such policies?

Scientific government must address these issues.

Examples

Some areas of government policy that might be addressed by the scientific approach:

• Medical care.
• Other potentially socialized services, e.g. car insurance.
• Safety nets: social security, welfare.
• Minimum wage.
• Environmental regulation.
• Anti-trust regulation.
• Foreign trade policy (tariffs).
• Taxation structure.
• Gun control.
• Drug policy.
• Immigration policy.

Example: Murder

We can assume that M(t) contains terms that encourage a low murder rate. What policies are likely to achieve this, and how can they be assessed?

We think of this in terms of punishment, with the idea that harsher punishment is a deterrent, with capital punishment at the extreme. But some studies suggest that harsher punishments don't result in lower murder rates. Who knows - maybe it works best to give murderers psychotherapy and - if needed - job training. This may be difficult for some people to swallow.

In any case, I suspect that systemic factors have a vastly larger impact on the murder rate than does the choice of punishment. If people don't have economic opportunity, if paths to wealth are criminalized, if mental health care is not freely available, if guns are easy to get - then we're going to have lots of murders, regardless of deterrents.

Scientific government would do experiments involving these causal factors. It would find and fix the reasons why most murders happen.

When a plane crashes, the NTSB investigates the hell out of it. They find why it happened - a mechanical failure, a problem with pilot/co-pilot communication, whatever - and they make a recommendation to prevent that kind of crash from ever happening again.

My personal view is most murders are like plane crashes: they reveal a systemic failure of some sort. There should be an NTSB for murders. Each murder should be investigated to find its root causes, and policies to address these causes should be explored.

Example: Abortion

In the U.S., abortion is primarily a wedge issue created by the right wing. However, they succeeded, and now any government in the U.S. needs to address the views of a big chunk of the populace, regardless of how those views got there.

How does the availability of abortion impact M(t)? It has been shown that it lowers crime - not surprising, since there are fewer unwanted children. But this doesn't address the concerns of anti-abortionists, who view fertilized ova as being fully-privileged people.

I think what it comes down to is: do the measures of happiness embodied in M(t) apply to fetuses? This needs to be specified in M(t). Which means that in scientific government, the abortion issue would be put to a popular vote. Which is about as good as we can do, I think.

Implementing scientific government

The organizational structure of science

The "scientific method" described earlier is an ideal. In practice, scientists needs money for salaries and equipment. Most modern societies, recognizing the economic value of science, have created organizational structures to support science.

These structures are essentially global, though they vary a little between countries. Generally are three main pieces, providing a form of checks and balances:

• Journals publish papers. Papers cite other papers. The number of citations of a paper measures its "impact". The "impact factor" of a journal is the average impact of its papers. A reputation of a scientist is estimated by the number and impact of their publications. Journals are owned by companies and professional societies. Each one has an editor-in-chief and an editorial board, who generally are well-connected scientists with good reputations. When a paper is submitted to a journal, the editors pick a set of reviewers who have expertise in the area.
• Research institutions such as universities and research labs provide a framework for training scientists and for conducting research. Hiring and tenure decisions are based largely on publication record, and on reputation in the scientific community.
• Funding agencies (such as the NSF and NIH in the U.S.) fund research. Typically the funding agencies have "program managers", who are academic researchers. They are have limited terms (2-3) years. Decisions about what areas of research to fund - the topics listed in "calls for propsals" - are made by panels convened by program managers. Grant proposals are reviewed by researchers selected by the program managers. Funding agencies check for conflicts of interest - for example, research that might benefit a company in which the researcher has ownership. The criteria for funding generally emphasize novelty. Other countries have somewhat different structures: for example, Germany has a set of "Max Planck Institutes", each of which receives long-term funding, and each institute has a permanent director. Both the U.S. and German systems have worked fairly well.

The scientific organization has been fairly corruption-resistant - certainly far more so than government. Why is this?

• Meritocracy. It's hard to get a foothold in science without being smart and doing something good and useful. A few charlatans have attained prominence, and a few great scientists became nut-cases later in life, but there have been no Donald Trumps.
• Transparency. Major processes - like hiring decisions at universities and grant proposal decisions - are documented, and these documents are public.
• Power is distributed across a lot of people.
• Positions of power - like department chairman and NSF program officer - are held for limited periods.
• The general public is involved only indirectly; riling up the masses via propaganda has little impact on science.

Of course, there have been attempts to corrupt the scientific structure.

• The Koch brothers have undermined climate research by creating a propaganda mill (Fox News) and electing an anti-science president.
• Monsanto has given lots of money to universities in return for power that enabled them to suppress research (e.g. about the spread of genetically engineered corn in the wild) that conflicted with their business goals.
• There is a certain class of academics (I can think of a few) who do lousy research, but do just enough of it to get into positions of some power (conference organizing, journal editorship, funding agency activity).

The structure of scientific government

What are the functions of scientific government?

• Conduct elections for figures of merit. Decide what goes on the ballot. Run elections in such a way that they can't be bought: no advertising? Keep the figures of merit short and simple enough that people can understand them.
• Decide what new policies will be considered.
• Design and conduct experiments, interpret the results of experiments, select winning policies.
• Enforce policies.

How to implement Scientific Government?

What is a plausible organizational structure for Scientific Government? I think a good starting point is to 1) piggyback on the existing organizational structure of science; 2) in extending this to a government, use the same underlying principles: meritocracy, distribution of power, transparency, etc.

Here's a possible structure for scientific government:

• There is a "policy experiment agency" (PEA). It is analogous to a research lab. Its staff includes permanent "researchers", analogous to professors, who have PhDs in political science, statistics, or related fields. Researchers carry out policy experiments and write up the results in research papers, which can be published in peer-reviewed journals. Hiring and promotion of researchers is based on their publication-based reputation.
• There is a "policy decision agency" (PDA). Its task: given a budget for experiments, decide which policy proposals to study. This is a complex problem; proposals may differ greatly in the cost of studying them, and in their potential payoff. The PDA is analogous to a funding agency. It is divided into "directorates" for different policy areas. Each directorate is staffed by rotating "program managers" with limited-term appointments. Program managers can be academics or PEA researchers, and are chosen on the basis of their reputation.
• Anyone (e.g. special interest groups) can policy proposals to the PDA. These proposals are analogous to grant proposals. They are expected to make a compelling argument that the proposed policy would increase M(t), and to outline a plausible experiment that would prove this.
• Proposals are evaluated similarly to NSF grant proposals. The relevant program manager identifies 3-5 blind reviewers (academics or PEA researchers).
• On the basis of these reviews, the PDA decides which proposals to accept. Each accepted proposal is passed to the PEA, where it is assigned to a group of researchers. These researchers design an experiment to test the policy change (usually but not necessarily based on the proposed experiment). They conduct this experiment (which may take months or years) and they write a paper describing the results.
• When the results paper for a proposal has been completed, the PDA reviews it and decides what policy to implement.

In designing the above structure, we need to anticipate various kinds of corruption.

• People in the PDA and PEA may have opinions and prejudices about policies. We need to prevent these from unduly affecting funding decisions, the design and execution of experiments, and the final decisions.
• Special interest groups my attempt to influence the outcome of experiments.

Note: it's possible and desirable that a new academic field of "government studies" arise, which would focus on how to conduct policy experiments. People who want to go into government could major in this.

Self-sizing government

Bloat is general problem of governments. Once an agency has been created, there is no incentive for it to downsize or eliminate itself even if its function becomes irrelevant. This if doubly damaging because it promotes general distrust of government, and the mistaken belief that the less government the better.

Scientific government provides a theoretical basis for deciding how much government is best. M(t) will have some component that reflects disposable income. A given policy (or governmental function) costs money, and therefore decreases disposable income. If a policy's net effect on M(t) is negative, it should be discarded.

This is analogous to the practice in many corporations of estimating the revenue brought in by each employeed, and firing those whose salary exceeds this.

What about bloat in the mechanisms of SG itself, i.e. in the PEA and PDA? We need a way to estimate how large these agencies need to be in order to do their job well; the reputation system described earlier provides a basis for deciding who to fire.

How can we get there from here?

It's unlikely that scientific government would rise out of the ruins of a completely failed existing government, so we need to think about a continuous transition. This would have to start small; some possibilities:

• A candidate for some office - say, mayor or governor - announces that, if elected, they will use scientific methods rather than relying on prejudices. Or an incumbent starts doing this.
• Identify an area of state or national government policy that's not highly political, and that's amenable to experiments. Set up a small version of the above structure, and use it optimize policy in the area. Maybe when people see how well it works they'll think about expanding the domain.
• A community or small town (of enlightened people) jointly agrees to govern itself scientifically. Their success would spread to other towns and eventually to larger scale.

In any case, scientific government will have a hard time getting started in a society where lots of people hate each other, hate government, are uneducated, and don't understand or trust science. These conditions currently exist in the U.S. So a necessary first step is to reduce these factors in the context of existing government (if that's possible). The number one thing is to improve education and make it universally available.

World government

The most important government issues - e.g. environmental policies - are now global. Dealing with them on a national level doesn't work; no country is willing to drastically reduce its carbon emissions because doing so would place it at an economic disadvantage. For such issues, scientific government at the national level is insufficient, especially if it's adopted only by a few countries.

So ideally we should have a global scientific government, whose domain is global issues: resource usage, environment, population, immigration, trade etc.

I'm on the fence about whether the idea of "nation" has any place in the future. I don't think we need national governments to preserve cultural diversity (which I view as a good thing). In any case, national governments can continue to exist separately from the global government, and could determine policies that are purely internal to that country.

Possible outcomes

What policies will scientific government converge to? It's impossible to say. The data will decide - that's the whole point. But my intuition is that the policies will be something like:

• The economic system will be highly regulated capitalism. The regulations will ban or heavily tax sources of external costs (carbon emissions, pollution, resource depletion).
• People will be guaranteed a subsistence-level living (food, housing, medical care) regardless of whether they work. People will have an incentive to work and contribute to society, but those who can't do so won't starve to death.
• People who work will be guaranteed a living significantly better than subsistence level.
• The wealth of individuals will be capped at a certain level - say $100M. Inheritance will be limited. People will have an incentive to create large, efficient businesses, but they won't become demi-gods.
• The government will use economic incentives to achieve and maintain a target population size.
• Arts and music will be heavily subsidized.

Related work

"Freakonomics" by Dubner and Leavitt examines a number of social-science issues - some big, some small - through a scientific lens. The conclusion is that when you look at data carefully and objectively, you often find surprises.

Nicholas Gruen writes about evidence-based policy. This is the idea that policy-makers in the current elected-leader framework should be expected to provide evidence that justifies their policies. Apparently this has been proposed, and people claim to do it, but it hasn't actually happened. It seems to me it's unlikely to ever happen, or to affect policy decisions, in the current framework.

The V-Dem Institute in Sweden maintains data on governments and their outcomes.

Feedback

Dave W. points out that the scientific organization is not as much of a meritocracy as I make it out to be, and that many funding/hiring/publishing decisions are political and bogus. This is certainly true, but I conjecture that:

• Corruption occurs more in the fringes of science, and that in the "center" (e.g. CERN, Harvard, Nature) it's less prevalent.
• The corruption occurs in part because - at least in the U.S. - the size of the scientific community is artificially inflated. There are hundreds of universities charged with being "research institutions", creating tens of thousands of research jobs, and there aren't enough actual scientists (or ideas) to go around. Hence there are lots of charlatans in the fringes. Scientific government ideally wouldn't suffer from this, because it would be only as large as necessary (see above).

My son Noah asks: would Scientific Government fund basic science research, if it doesn't directly contribute to M(t)? No. If we want basic research, we need to define M(t) to allow it. One way to do this is to observe that historically, basic research leads to beneficial technology after a few decades. Quantum mechanics in the 1930s enabled lasers and microchips in the 1960s. So if we evaluate policies based on their expected long-term effects on M(t), we'd fund basic science. And we'd leave it up to the scientific community to decide what kinds of basic science to pursue, as we do now.

What about research in things like pure math? And what about space exploration? It's hard to argue that these will ever put food on anyone's table. But maybe understanding the universe make people happy, and M(t) can include a clause for it.

Noah also points out that experimentation means that some people will suffer from sub-optimal policies. This is true, but it's no different from clinical drug trials. A small amount of suffering during an experiment is better than perpetual and universal suffering because of a bad policy. Also: the size and duration of experiments can be minimized by using modern (Bayesian) statistics.

Several people thought I was proposing putting scientists in charge of government. I can only conclude that they didn't actually read the essay. Perhaps they looked at the title, formed a mental model of what the essay must say, and moved on.