Saving the world by separating ought from is
Why we need better fact-value separation norms, and what those might look like.
As I said in my intro post, we live in an age of widespread disillusion with institutions and widespread belief in declining institutional quality. Experts get it wrong, role models turn out to be corrupt and malfeasant, infrastructure crumbles, deliberations become shouting matches. All these failures are in our faces every time we read the news or look at social media feeds. Across corporate, governmental, nonprofit, and even interpersonal domains, it feels like our mechanisms for getting things done together are creakier than they used to be.
This feeling may or may not be grounded in reality. Certainly institutional failures are more visible than before due to modern communications technologies
, which could lead us to overestimate the extent of decline. And the particular problems of modern life may strain institutions more than the problems of prior generations. Still, even perceived decline can become a self-fulfilling prophecy, as losing trust in institutions can make them even worse, which makes us lose even more trust, etc. Conversely, if we could build visibly improving institutions, we’d make people both feel a lot better and actually be a lot better off.This post argues that one powerful thing we could do to improve institutional quality is develop better norms of separating facts from values. If that sounds vague to you, I hope reading on will make it more concrete.
Why separate facts from values?
The reasoning may be obvious to some, but is worth making explicit:
Methods that work to pursue consensus on facts don’t work to pursue consensus on values, and vice versa. Furthermore, in practice consensus on values is far more elusive and unstable than consensus on facts. Note that I’m not taking a position here on whether there are objectively correct values, and if I’ve written this post right, nothing in it should depend on that question.
In particular, in a secular society, there are no experts on values, only experts on various domains of facts.
Moreover, experts on facts tend to forget that they aren’t experts on values and that their values aren’t universally shared. This often leads to smuggling value judgments into what are presented as statements of fact. That in turn makes the statements less reliable and the experts less credible.
And to be a good expert in any domain, you need a sound process for pursuing, discovering, and refining factual knowledge. Laying values aside when pursuing facts can make that process more effective by reducing confirmation bias: the well-known tendency we all have to hold factual beliefs we want to hold because they are convenient for our values, and to reject factual evidence that is inconvenient for our values.
Having better experts, and better processes for pursuing factual knowledge, is one of our most important tools for achieving our values, regardless of what they are. To build a better world, we must get better at finding out what the world is like now and why.
Finally, division of labor and separation of concerns have proven throughout modern history to be powerful tools for improving institutional productivity. Fact-value separation is one key instance of that general pattern.
Existing separation norms
The above observations aren’t new; many institutions have norms that draw on them. Here are a few that come to mind immediately:
Recusal of decision-makers with conflicts of interest. When a judge who must decide whether a person committed a crime has to recuse themself if the person is a close relative, it’s in order to keep a fact-finding process free of one type of value judgment that can bias it.
Separation of editorial and news departments. Newsgathering journalism, in principle, should be committed to objectively finding out what is true; editorial writing, to making value-laden arguments for what should be true. Keeping them not only in separate sections of the newspaper, but staffed by separate people, aims to preserve that fact-value separation.
Separation of advertising from non-advertising content. We know advertisers have value-laden motives which can lead them to say things that non-advertisers wouldn’t. So knowing which content is advertising, and insulating non-advertising content production from advertiser motives, helps us perceive facts better. This one is dear to my heart from long experience at Google, where the wall of separation between search and ads divisions was something we took great care to try and maintain.
Some quick lessons come out of this list. First, whenever we strengthen the use of these existing norms, we strike a blow for better fact-value separation, thus for more effective social pursuit of facts, thus for a better future. In practice insisting on these norms often seems boring, schoolmarmish, pedantic, the sort of thing enjoyed by the type of person who says “Point of order!” a lot. But it’s Good Actually (tm) and people should take civic and professional pride in doing it.
Second, these norms are inevitably imperfect for general reasons that apply to whatever other separation mechanisms we might set up. Humans are flawed: reporters have preexisting notions and biases which will color their reporting, however hard they try to make it objective. There is a complex “attack surface”: advertisers can indirectly influence non-advertising content without ever telling anyone what to write. And fact-value separation can be in tension with other important conditions for truth-seeking: often the people who have conflicts of interest in a given domain are also the people who know that domain best, and there is no good way to advance knowledge in that domain if they recuse themselves.
Let’s separate scholarship from activism
Hopefully the claim is clear from the title. But to elaborate: if you’re a scholar, in or out of academia, whose job centers on finding out what’s true in some field, you should feel an institutional responsibility not to do advocacy work, especially advocacy work related to that field. And if you are doing advocacy work, you should not present yourself as a scholar of the relevant field(s), but should make clear that advocacy is your focus, and that people shouldn’t trust you as a reliable source of scholarly judgments. This all applies a fortiori if the “you” here is a collective body like a think tank or other nonprofit.
The arguments for this are much the same as the general arguments for fact-value separation. Activists who do scholarship naturally feel a powerful urge to look harder for evidence supporting their existing advocacy narrative than for evidence contradicting that narrative. Even if that doesn’t distort the factual conclusions they reach, it undermines others’ trust in the objectivity of those conclusions.
So it would be good for truth-seeking and for the trustworthiness of experts to have stronger institutional walls between scholars and activists. Yes, this applies to your favorite hot-button issue that you feel really strongly about. It means climate scientists shouldn’t be advocating for particular climate policies, scholars of various social inequities shouldn’t be social justice advocates, and so on. To give a recent example from my own interests: two advocacy organizations I support, Heterodox Academy and the Foundation for Individual Rights in Education, have done research on the extent of speech suppression in academic settings. I’m glad someone’s doing that research, but I wish other more purely scholarly organizations would do it instead, because the research results would be much more convincing coming from such organizations. This would especially help convince people who don’t already share the values that those organizations (and I) share.
To be clear, the point is not to denigrate activists or minimize the importance of advocacy work. Activism often helps us get to a better world! The claim is that the combined net effect of activism and scholarship will be more positive if the two are kept more separate.
And the caveats are likewise the same as for fact-value separation norms generally. Activists and scholars will inevitably talk to each other and influence each other’s thinking— activists have to talk to scholars to get their “ammunition”, and scholars are often motivated to do their scholarship by activist ambitions rather than a “mere” or “detached” curiosity. Thus, as with many recusal norms, if you exclude scholars in a field from doing advocacy work, you risk being left with too few people qualified and motivated to do the advocacy work. You can address this by having people consciously wear different “hats” at different career stages, but then you may create a “revolving door” dynamic of increased cross-influence.
So why try to do better anyway? In short, because it could broaden trust in truth-seeking institutions like universities and media, and could help us get a broader and more stable consensus on known facts. The lack of such consensus, the much-lamented meta-fact that we can’t get more people to agree on empirical propositions about issues like vaccines, climate change, and election integrity, is widely recognized to be a key driver of institutional paralysis. It surely has many causes, but one significant cause is that people who don’t want to believe a thing wave away evidence for that thing because the people presenting the evidence are advocates for values they don’t share. If you want more people to believe X, you need to make it harder for them to say “you’re just saying X as an excuse to advocate Y”. Better scholarship-activism separation can help make it so.
Governance reforms and fact-value separation
Rationalist types who like to think about better social decision mechanisms should recognize that many of these mechanisms likely operate by strengthening fact-value separation. The most obvious case is Robin Hanson’s proposed “futarchy” whose slogan is “vote on values but bet on beliefs.” The idea is that good governance requires
a deliberative process to discover what outcomes people want from government
an actually effective means of delivering those outcomes
and that the former is a matter of values while the latter is one of facts, so we should separate the two and assign each to a mechanism specialized for the relevant domain: democratic debate on values, market bets on facts.
Note how this template generalizes. Maybe betting markets aren’t the optimal way to choose effective means of advancing social welfare— or maybe deliberative democracy isn’t the optimal way to hash out what social welfare means. You don’t have to have faith in the efficacy of either one in order to believe that it would be better if we used different mechanisms for these two parts of the governance process!
Consider also the more incremental reforms proposed in Garett Jones’ book 10% Less Democracy. Many of Jones’ proposals focus on giving various government functions more independence from direct democratic accountability: the sort of independence many central banks have today. These functions tend to be exactly those where there is broad consensus on values— nearly everybody likes low inflation, low unemployment, and steady economic growth— and experts need the independence to find and act on the relevant facts about how to advance these values. Other proposals concern giving situation-specific voting power to entities who are more likely to have relevant factual knowledge, again separating the egalitarian domain of values, where no one is an expert, from the more “meritocratic” business of truth-seeking.
Observe that these kinds of separations in governance tend to be least effective in cases where both facts and values change quickly, and/or prior consensus breaks down. The obvious recent example is FDA regulation during COVID: what had been a relatively technocratic, independent agency making decisions based on a broad bipartisan value consensus is now deluged with criticism and hard-pressed to steer a consistent course. Now, I’m actually glad to see the FDA “politicized” because I’m a longtime dissenter from that broad value consensus: I think that even pre-COVID the FDA set way too high a bar for approval and thereby killed a lot more people than it saved, and COVID has just made the cost of their overcaution higher and more obvious. But whatever one’s position on that issue, the institutional squeeze they’re under should be a reminder that neither fact-consensus nor value-consensus is stable forever, and that robust rules for fact-value separation must be able to adapt to cases where what had been taken for granted can no longer be.
What did I miss?
The above account must be incomplete. Are there existing fact-value separation norms I didn’t list? Other reform proposals that fall under this umbrella? I’d love to know, and to update this post as I gain that knowledge, so don’t be shy!
In particular, two recent books that seem relevant to this debate are Julia Galef’s The Scout Mindset and Jonathan Rauch’s The Constitution of Knowledge. Both are on my to-read list and I’ll likely update this post after I read them. I’m publishing now anyway in the interests of not letting the best be the enemy of the good.
The person to read on this is Martin Gurri, starting with his book The Revolt of the Public.
Credit to my wife Catharine for making this point in discussion.