Deep Honesty

Most people avoid saying literally false things, especially if those could be audited, like making up facts or credentials. The reasons for this are both moral and pragmatic — being caught out looks really bad, and sustaining lies is quite hard, especially over time. Let’s call the habit of not saying things you know to be false ‘shallow honesty’^[1].

Often when people are shallowly honest, they still choose what true things they say in a kind of locally act-consequentialist way, to try to bring about some outcome. Maybe something they want for themselves (e.g. convincing their friends to see a particular movie), or something they truly believe is good (e.g. causing their friend to vote for the candidate they think will be better for the country).

Either way, if you think someone is being merely shallowly honest, you can only shallowly trust them: you might be confident that they aren’t literally lying, but you still have to do a bit of reverse engineering to figure out what they actually believe or intend.

This post is about an alternative: deep honesty, and the deep trust that can follow. Deep honesty is the opposite of managing the other party’s reactions for them. Deep honesty means explaining what you actually believe, rather than trying to persuade others of some course of action. Instead, you adopt a sincerely cooperative stance in choosing which information to share, and trust them to come to their own responses.

In this post, we’ve leaned into the things that seem good to us about deep honesty. Writing while being in touch with that makes it seem easier to convey the core idea. We’ve tried to outline what we see as disadvantages of deep honesty, but we’re still probably a bit partial. We would love to see discussion of the idea, including critical takes (either that our concepts are not useful ones, or that this is less something to be emulated than we imply).

The rest of this post will be:

Some examples of where deep and shallow honesty diverge
Why and when you might want deep honesty
Various disclaimers about what deep honesty is not
A look at some difficult cases for deep honesty
What deep honesty might look like in practice

Examples of shallow (versus deep) honesty

Writing a very optimistic funding application which doesn’t mention your personal concerns about the project
- As opposed to being upfront about what you think the weaknesses are
Telling an official at border control that you’re visiting America to ‘see some friends’
- Rather than explaining that you’re also going to some kind of philanthropically funded conference about AI risk
Searching for and using whichever messaging makes audiences most concerned about AI risk
- Instead of whatever best explains your concerns
Saying that you totally disagree with the ideology of an extremist group
- And not that they are actually right about some important controversial topics, in a way that doesn’t justify their actions
Reassuring your manager about all the things that are going well and privately trying to fix all the problems before they grow
- Instead of telling your manager what’s going wrong and giving them an opportunity to make an informed decision about what to do
Rejecting someone from a programme with a note explaining that it was very competitive
- Rather than explaining what you perceived to be their weaknesses and shortcomings for the role
Telling yourself that you’re doing something for utilitarian reasons
- Instead of acknowledging that you also have a pretty weird kludge of motivations which definitely includes being recognised and appreciated by your peers
When a friend asks how you are, smoothly changing the topic because you don’t want them worrying about you
- Rather than opening up about private difficulties, or even just giving a wry smile and saying “well enough” in such a way as to provide a trailhead for a conversation about those difficulties if they want to pursue it
Sharing the fact that you have some frugal habits (like driving a Corolla), because you think they will make people think well of you
- And not also mentioning that you frequently spend large amounts of money on luxuries you consider time-saving

Why deep honesty?

In all the above examples, it’s easy to see how deep honesty could go wrong — your boss thinks you’re a moron, your friends don’t care about your feelings, you have a weird existential crisis about whether you’re even a good person, and instead of going straight to a conference you get dragged off for half an hour of interrogation by government officials^[2].

But what if it went right? It seems like when deep honesty is well-received, it leads to better outcomes, often in ways you can’t foresee. This isn’t a trite or mystical claim: rather, there will always be information you’re lacking that other people have. When you’re deeply honest, you equip them to make best use of their private information as well as yours. Perhaps your manager knows of a clever workaround to the problems you have. Even when they can’t make special use of the information, deep honesty makes it easy for them to rely on your reports, and so strengthens the relationship.

There are good reasons to refrain from deep honesty: it is a risk, and sometimes a large one. In the case of admissions, usually when you reject someone from a programme, you have a pretty clear sense of why, and actually explaining that to the applicant could be very helpful. But as well as being time-consuming, telling someone why they didn’t meet the bar can provoke quite a negative reaction and sometimes even reprisal. Deep honesty is an act of trust in the recipient.

However, it’s also sometimes quite hard to realize how much you’re missing when you stick to shallow honesty, and is easy to overestimate how successfully you’re crafting your message. Shallow honesty involves some amount of optimization, and so it falls prey to all of optimization’s classic failures.

To take the example of public messaging, particularly smart and capable people are especially likely to spot disingenuous sales pitches, and when they do, they have basically no reason to tell you that you seem manipulative to them, and so there’s no feedback loop. From their perspective, there’s a chance you might be provoked to switch to deeper honesty, but you also might just optimize more carefully. So you learn nothing, and you end up missing out on some of the best people without even noticing.

Shallow honesty works well enough in cases where, in some sense, the other person wants it. The border patrol official would be happy to accept that you’re here to see friends and wave you through, without working their head around the subject of the conference. But in domains where people are actively trying to resist adversarial optimization, they can catch on pretty fast.

Indeed, people with experience running admissions rounds generally learn how to spot applications that are shallow, because so many people do it. Likewise with funding applications. It is very natural to want to put your best foot forward and pitch people on why you’re so shiny and polished and great, but often it is a mistake. Sometimes the person reading your application wants to understand what you’re actually like, and if you clearly only give them half the picture then they still have to figure out the other half, only now they’re much less certain.

What deep honesty is not

Having made the case for deep honesty, it’s worth laying out some pitfalls, both in the application, and in how you might interpret the concept.

It is not a universal stance

Deep honesty is not a property of a person that you need to adopt wholesale. It’s something you can do more or less of, at different times, in different domains.

It is not independent of the listener

The words which will help a young child to understand what’s going on will be different from the words which will help an expert. Deep honesty is attuned to the situation, and the audience.

It is not telling people everything

Deep honesty doesn’t mean you have to share every detail that might be relevant. Deep honesty is in touch with what the listener cares about, and is in touch with your and their rights to choose where to spend time communicating. If the cashier at the grocery store asks how you’re doing, it’s not deeply honest to give the same answer you’d give to a therapist — it’s just inappropriate.

It does not relieve you of a responsibility to be kind

Deep honesty means you don’t take responsibility for how others respond to your words. Your responsibility is to make your words good — speaking with truth, relevance, and kindness. Their responsibility is to act well given that. But blunt truths can be hurtful. It is often compatible with deep honesty to refrain from sharing things where it seems kinder to do so (although be honest with yourself about whether it would be a deeper kindness to share). And it’s of course important, if sharing something that might be difficult to hear, to think about how it can be delivered in a gentle way.

It is not incompatible with consequentialism

A pure act-consequentialist, choosing the words that they predict will have the best outcomes, might sometimes lie. Many consequentialists would reject that as naïve and demand at least shallow honesty as a side constraint.

However, this may still be too naïve. The winner’s curse is that whoever wins an auction is liable to have overestimated the value of the object. The same dynamic applies when you’re optimizing for what to say. You have noisy estimates of how good each option will be, and it’s likely that the one that looks best to you will be an overestimate — perhaps because it interacts with some kind of blindspot you have. If you’re optimizing under shallow honesty, you’re stuck with this problem. With deep honesty, you can hope that you may reveal useful information to people who don’t share your blindspots (even if you don’t know what that information is). And especially when you’re interacting with very competent people, you may not be so good at telling how they will receive any particular message.

So deep honesty as a heuristic for action for boundedly rational actors looks pretty good on consequentialist grounds. It’s very compatible with taking the low-hanging fruit of consequentialism — thinking through possible bad effects of communication, and taking steps to mitigate those. (Deep honesty also looks generally very good from non-consequentialist perspectives on ethics.)

Challenging cases for deep honesty

It’s not always the wise or moral choice to be deeply honest. Deep honesty is a risk, and it’s a bigger risk in some cases than others.

Even when you’re not being deeply honest about everything, it’s often worth remaining deeply honest at the meta level^[3]. Warn people that you’re biased and may argue for one side. Tell them that you’re simplifying things, or steering around a topic you don’t want to get into.

Large inferential gaps

Sometimes you have a very different worldview from your audience.

If you have a good understanding of their perspective (e.g. imagine explaining something to your own small child), you may be able to predict that they might draw inferences you’d consider inaccurate from things you share. It isn’t deeply honest to knowingly let them draw important false inferences, at least without warning them about this issue. But when bandwidth is limited, you may well not be able to bottom out all of the differences in perspective. In this case, deep honesty means improving their understanding of relevant topics in ways they’d endorse with moderately more context (you don’t get to assume they come to endorse your whole worldview). Sometimes this means (transparently) steering around a topic that’s more likely to cause inaccurate inferences; sometimes it means going out of your way to cancel possible implications.

Sometimes you don’t even know what inferences they might draw. Then it’s especially easy for attempts at communication to go wrong, and you might want to be correspondingly cautious about it. Deep honesty may take you into a zone of sharing things you might not otherwise share and are vulnerable. On the other hand, it’s hard to optimize in a shallowly honest way when you don’t understand the audience, so the unforeseen benefits of deep honesty can be especially helpful in these cases.

Audiences you don’t want to cooperate with

Sometimes people will want information so that they can cause harm, and it is reasonable to not help them. Sometimes you will meet people who actively want to twist your words, and it is reasonable to not give them ammunition.^[4]

Multiple audiences

It can be harder to be deeply honest when delivering a single message to multiple audiences that have different contexts and background assumptions. What’s most useful to one audience may not be most useful to another.

We can distinguish between active deep honesty, where you are trying to share whatever information the listener would most want (to reach an informed independent view), and passive deep honesty, where you’re at least not aiming to persuade the listener of something. With multiple audiences, you may only get to choose one to be actively deeply honest with in any moment^[5], but you can always be passively deeply honest with all of them.

Sometimes you have some audiences you’d like to be deeply honest with, and others you wouldn’t. Now you have to make a judgment call about how much you value deep trust with the first group, versus how worried you are about the risks of deep honesty with the second.

Here’s a very rough sketch of the concepts we’re using

What being deeply honest might look like

Deep honesty is about empowering your listeners. In principle this could involve conscious optimization for what seems like it might be most useful for them. But as a practical matter, the best guide is often asking yourself, as you say something, “did it feel honest to say that?”^[6]. Anecdotally, it seems like this can lead to a qualitatively different mode of expression — where you don’t allow your communication to be steered by ulterior motives — and that some people are pretty good at intuiting when people are or aren’t in this mode^[7]. This is a very helpful skill to develop.

Deep honesty is often a bit scary, because you don’t know how others will react to it. This is why engaging in it can require something like faith, that striving after virtue will lead to good things, even if you’re not in a position to be able to say what those are.

Fortunately, although deep honesty has been described here as some kind of intuitive act of faith, it is still just an action you can take with consequences you can observe. So rather than diving in wholesale, you can just try to pay a bit more attention to where you’re already doing it or not doing it, whether it seems like others are doing it, and experiment with doing it a bit more in cases where that seems like it might work out.

Seriously, skipping to 100% deep honesty all the time would be a mistake, and also probably impossible. But it seems like maybe deep honesty is underrated right now.

So ask yourself more often, when thinking about how to communicate, “what is kind, true, and useful?” and “what is the heart of the matter?” rather than “what will have good effects?”. Take a moment to appreciate the people who seem to actually consistently say what they really believe, even if it means revealing that they’re wrong or ignorant or have silly reasoning about something, and especially if it’s not politically expedient.

Give it a go when you get the chance, and see where that gets you.

^
Of course even this is not always necessary — for example, if someone asks how you’re doing, many people think it’s fine to say “good”, even if your cat just died.
^
As regrettably more or less happened to one poor author of this very post.
^
As a case in point, we want to discuss the fact that in making this (three-authored) post anonymous, we’re holding back from deep honesty. We don’t regard our identities as a great secret, but it seems to us that publishing under our names could be seen as having an ulterior motive, of trying to persuade people to deeply trust us (in general). This (it seems to us) could actually make it harder for people to feel a certain kind of trust in our purposes in writing this post, and could get in the way of people engaging directly with the ideas. By removing ourselves from the equation, we hope to keep things clean. However, we are conscious that there might be some benefits we cannot foresee to posting non-anonymously, so it seems possible this is the wrong call. We at least wanted to be open about our thinking on this point.
^
Obviously, this determination is hazardous. At least one of the authors believes they have gotten this importantly wrong in the past sometimes, and that more deep honesty would have served their own goals and the world better.
^
Although not-infrequently the same information will be desired by many audiences. And even when that isn’t the case, you can often talk to multiple audiences in sequence, flagging who each part is meant for.
^
Although it seems like the ability to accurately judge this can be inhibited by stress. This predicts that people who are under stress will typically be less honest, even without having chosen to be dishonest — and is an extra reason to worry about patterns of overwork and/or burnout in sectors of EA. This is also one argument for sometimes taking a deep enough rest that you are not stressed, and for asking yourself then how deeply honest your previous actions were.
^
One author of this post would like to thank the various people who consistently and gracefully called him out on his occasional insincerity, for prompting him to actually notice it and sharpen the relevant intuitions, and heartily recommends that everyone find such friends.