Noise on the Channel

Articulation of these ideas in their present form owes a debt to interactions with Tsvi Benson-Tilsen, and Erin Tatum.

Almost everyone will be familiar with the concept of signal vs noise. Literally, it’s a signal processing concept which differentiates useful information which we’re trying to communicate from useless distractor information which can corrupt our signal. Those who grew up in a pre-digital age will be familiar with “static” on the telephone line or TV. Digital information can be transmitted almost error-free through a noisy channel via redundant encodings which allow error-correction, at a rate determined by Shannon’s noisy-channel coding theorem. This is a likely reason for the level of redundancy in natural language, as well: it aids communication in a (literally) noisy environment.

Metaphorically, we use the concept of signal vs noise to talk about everything from inboxes and newsfeeds to writing styles. To this end, people talk about the signal-to-noise-ratio: the proportion of useful/​desirable information to total information in a given information source. This is useful in part because it helps manage attention: the total amount of useful information on (say) Twitter might be very large, but because of a very low signal-to-noise ratio, it may not be an efficient way to get information. In contrast to the technical signal-processing model, where the sender and receiver share a concept of which information is useful, this metaphorical generalization admits that the sender’s “signal” might be the receiver’s “noise”.

I’m here to talk about a further metaphorical extension of the signal/​noise concept. I don’t know whether this concept is especially useful, but it’s very strongly a part of my personal experience—this is one of the most salient aspects of a conversation for me, and one of the biggest factors in determining how enjoyable or productive a conversation is. I call it “fuzz” or “static” or “noise on the channel”.

EDIT: Jimmy rightly points out that the concept I’m pointing at is more like “the opposite of bandwidth” (ie, I’m describing the way a low-bandwidth channel warps conversations). Noise makes for low effective bandwidth due to the above-mentioned noisy-channel coding theorem. So it makes sense that the two have very similar effects on conversations.

How much static is in this conversation?

I’m pointing to a set of conditions which all have a similar way of making conversations more difficult and less fruitful.

Here are some examples of what I’m talking about.

  1. Literally, a noisy room. A bar on a busy night; everyone is shouting in an effort to be heard over the loud music and the other people shouting. (Literal unironic object-level question: why do so many people think this is a good social setting? Maybe the noise serves an important social function I’m not seeing?) 2 One or both people are hard of hearing. This is practically the same as a noisy room.

  2. One or both of the participants are repeatedly distracted. Threads of inquiry keep getting interrupted, and sometimes forgotten.

  3. You are talking to someone who has to leave in a minute. You both know you don’t have time to get into any complicated topics.

  4. One or both participants lack fluency in their common language. Otherwise simple things may take minutes to get across, much like a game of charades or person-do-thing. Complex subjects cannot be discussed, unless the conversation is very low-noise in other relevant aspects (IE, the participants are committed and have a lot of time).

  5. One or both people lack interest in the discussion. Like the example where someone needs to leave soon, it’s likely that you don’t have a lot of time, because a disinterested person may break off the conversation early. Like the example where there are constant distractions, it’s likely that you don’t have full attention, and points may get cut off or dropped.

  6. There is a high inferential distance. The conversation participants have very different ways of thinking about the subject at hand, which have been developed over long time periods and have a lot of details. Even when the language appears to be shared, there may be hidden differences which are actually critical (see the double illusion of transparency). Like the case of lacking fluency, this means both speakers need to spend a lot of time carefully conveying concepts and checking whether they’re understood.

  7. There are a lot of conversational land-mines. Secrets which need to be kept, or touchy subjects which can’t be brought up. You need to tread very carefully to avoid blowing up.

In all of these situations, I experience a very similar stressful feeling. I’m trying to squeeze my ideas through a tiny straw. Often the ideas stay bottled up, because it’s impossible to communicate complex thoughts. One of the main things I want to get across in this post is my model of why communication is so terrible in these situations.

Why Noise Sucks So Much

All of the object-level difficulties I listed in the previous conversation are different. However, I think the main source of difficulty in such conversations is often the Nth-order effects the “noise” has on the conversation, which are very similar. Many different obstacles to good conversation cause each other and compound on each other to make for a sucky conversation.

In a noisy room,

  • I have to shout to be heard.

  • Shouting takes effort, which makes me a little more reluctant to speak.

  • I’m not sure if I will be heard, which makes the expected value of speaking lower.

  • I’m not sure whether I was heard, which means I’m not sure I can build on my previous statements.

  • It’s difficult to hear the other person, which means I have to fill in the gaps, making assumptions about what they probably said.

  • The same is true for them, meaning I have to worry about whether I was really understood.

  • The need to make additional statements to check whether I’ve understood what they said multiplies with the extra effort of shouting.

  • Even if we largely are being understood, the constant worry that we aren’t still makes it more difficult to build on previous points in the discussion.

  • All of the above combines to lower the expected value of the conversation.

  • Because both of us know these things lower the expected value of conversation, we both have less faith in each other’s commitment to the conversation.

  • Even if we are both fairly committed to the conversation, our lack of faith in the other person’s commitment means we have to treat them like a possibly distracted/​disinterested person. This lowers the expectations for the conversation even further, recursively compounding the effect.

  • This worry that the other person isn’t going to be very committed to a good conversation means we can’t even expect lengthy error-checking procedures to enable us to get complex points across, because we don’t know whether the other person will be motivated enough to participate in correcting errors or verifying that points were understood.

  • All of the above means that we are restricted to things which (1) can be communicated fairly quickly, and (2) are commonplace enough that the other party is likely to guess our meaning correctly despite all the communication difficulties. Basically, small talk. This restriction in feasible subject matter further drops the expected value of the conversation, further compounding other effects.

  • Since both people probably realize that the feasible subject matter of conversation is restricted, this knowledge plays into the guesswork we do when trying to figure out what the other person meant /​ check whether we heard them correctly. This fact itself further reinforces the restriction of subject matter, since it means we’ll be even more likely to be misunderstood if we say something complicated.

I could go on. The point is that the bad effects compound each other. A noisy conversation involves a heavy game-theoretic component. Each participant’s expectations of the value of the conversation is heavily dependent on (their estimate of) each other’s expectations. There’s a stag hunt for a good conversation, but the cost of hunting stag is being driven up, without driving up the reward. This means people are even more likely to hunt rabbit than usual, even if hunting stag would still be the overall better option. (And the perception that people are more likely to hunt rabbit makes it even more likely, which feeds back in… well you get the idea.)

You might think you’re not doing all the metacognition which I describe above; or, that “normal people” don’t do that much metacognition. And maybe not. But I don’t think you actually have to do the metacognition in order to feel the consequences. A simpler reinforcement-learning like algorithm will still teach you, via conditioning, that you can’t expect deep conversations in certain contexts. As people learn that, they’ll try less, and teach each other even more that it’s not going to work. So without even thinking about all the recursive implications of the noisy environment, you might have a general sense of doom about difficult conversations in noisy environments. If you’re like me, that sense of doom will also pervade a wide variety of similar situations which aren’t literally noisy, but share critical features in common with noise.

The Wonderful Magic of Noise-Free Conversations

I still expect some readers to not really know what I’m talking about. Those readers may not even know that they don’t know what I’m talking about. Noise is pervasive. A truly low-noise conversation is a rare and precious thing. It’s like falling in love. It’s like an old friend who understands you. It’s Deep Work. It’s the joy of being seen and being understood. You don’t know what you’re missing until you’ve experienced it.

Of course, this is all a matter of degree. There’s the simple everyday variation in “noise” which comes from distracted vs undistracted time, close friends vs acquaintances, et cetera. Then there’s the rare, really deep conversations which happen when two people are really very interested in understanding each other, repeatedly make time for each other, and work together to eliminate distractions and other barriers. And then there are the as-yet-undreamt-of heights of noise-free conversations which can only be attained by black-belt rationalists who have first internalized and then later transcended all kinds of cognitive skills related to good conversation, after ingesting all the right nootropics and heading to an extended wilderness retreat.

Let’s reverse some of the previous points I made, to clarify what a really low-noise conversation looks like:

  • Low literal noise. Everyone’s literal words are understood easily. Everyone knows this without hesitation, so it fades into the background and doesn’t take any attention.

  • No distractions. Everyone has a clear mind to focus entirely on the discussion. Again, everyone knows this and doesn’t have to think about it.

  • High level of interest. It’s common knowledge that everyone in the conversation wants to continue engaging in the conversation, and is interested in understanding what others have to say. There is a high expectation of follow-through on lines of thinking, even if those lines of thinking are very tricky and subtle and will take a lot of time to follow through.

  • Relatedly, large time commitment. The conversation has all the time it needs. If the conversation eventually has to end on this particular day, there is a high degree of trust that you’ll get together again soon to continue it, and do so repeatedly for as long as the subject requires it. There is no end in sight.

  • Points are never dropped unless everyone thinks they’re finished. In the ideal, there is perfect memory of the conversation, everyone readily knows what the open points are, and those points get returned to in an expedient manner. (Of course in reality, different points have to compete for time.) Conclusions of the conversation are fully internalized by all participants, and applied in any relevant contexts which come up later (in this conversation or beyond). One example of a helpful tool is a shared space for notes like a whiteboard, on which important points get written.

  • There is a large shared context of understanding. Complicated concepts, feelings, and intuitions which would normally be obscure are easily conveyed and understood, due to special shared language which the participants have developed for their needs in this conversation.

  • You can say anything that’s on your mind. There are no conversational landmines, no secrets, no taboos. Nor is anything considered off-topic; since there is a strong shared interest in the subject matter and a high degree of trust in that mutual interest, there is no need to police the conversation to avoid distractions. Nor would there be any need even if not for that, due to the large amount of time available, and the infallible memory everyone has for the active points of discussion. All of this means that when you start on a seemingly irrelevant branch of discussion, no one tries to reel you in; nor will they blame you if it ultimately turns out to be irrelevant. Nonetheless, everyone does largely stay on-topic.

Despite my praise for low-noise conversations, it bears mentioning that this isn’t the optimal kind of conversation to have for all purposes. Relaxed, distracted conversations can be great for getting to know someone—e.g., a highly distracted conversation over a board game. Some subjects demand fast, time-limited conversations. Not all subjects of conversation merit a high level of interest; boredom is sometimes the correct response. And so on.

It’s also sometimes possible to get really good conversations by dramatically lowering some kinds of “noise” despite other types being very high. For example, a conversation with high inferential distance is likely to have a lot of really valuable information, if you can give it the time and attention to bridge the gap. Another example: email conversations are likely to be slower and lower-commitment, but this can be compensated for by the fact that all points are remembered (everything is in a text record) and participants can take a lot of time to compose their thoughts. (Keep in mind that the probability you’ll write a thoughtful reply influences the amount of effort the other party will put into their email.)

Dealing with Noise

Sometimes you just have to make due with a noisy conversation. In that case, it pays to have some coping strategies.

Lower your epistemic standards. Sad to say, you may be faced with the choice between communicating something poorly and not communicating it at all. In some cases, communicating it poorly will be preferable. I wouldn’t recommend practicing this as a skill so much as trying to notice that you already do it—better, at least, to explicitly flag for yourself that you’re less than totally accurate. Some examples:

  • Guess at what the other person means, rather than seeking clarification. You don’t have time/​energy/​etc to get clarification. Fly by the seat of your pants in this conversation. Just make a guess and go with it.

  • Settle for communicating something in the right cluster. Maybe there isn’t bandwidth in the conversation to tell them what you were really up to yesterday, even though they asked. Maybe “working” is a lie for subtle reasons. You weren’t really working. But it gives them approximately the right idea.

Pick the most important point, and drop the rest. The conversation doesn’t have the attention for everything right now; you just have to make a choice.

Accept being unheard or misunderstood. Maybe you were feeling kind of off about something that happened yesterday and you wanted a sympathetic ear to talk it out with. Oh well. This conversation isn’t the one where that’s going to happen. Let’s talk about the weather or something instead.

Am I the Noisy One?

On the other hand, you could be doing any of the above things unnecessarily, creating a “noisy” conversation despite the lack of a noisy environment. Like I said, a good conversation is a stag hunt. Are you hunting rabbit unnecessarily? Are you ignoring your conversation partner’s attempts to hunt stag? Are you not giving them the opportunity to try?

I suspect this can be easy to miss if you don’t have a lot of experience with the deeper sort of conversation which (unknown to you) your conversation partner is trying to have. Imagine an angsty teenager who assumes any genuine conversation about feelings is a setup for making fun of them. Or imagine someone just starting as a graduate student, who doesn’t have any experience with pre-rigorous research concepts turning into rigorous concepts later, so blocks themself off from engaging with ideas that don’t sound rigorous (because they’re trying to be a serious researcher).

If you notice yourself engaging in some of the “dealing with noise” strategies from the previous section: are you hunting rabbit when others were trying to hunt stag?

Credibly Committing to Continuing Conversation

If approaching this as a problem to be solved, rather than just a phenomenon to be aware of, one approach is to visibly set time aside, set aside distractions, and give a conversation your full attention. Remove distractions: set aside phone, laptop, etc. Find a private room or a semi-isolated outdoor location. Perhaps take the conversation on a long walk without a cell phone, which provides a visible commitment to keep talking for some amount of time. If you want to make sure there are follow-up conversations, maybe mention that early on, to establish common knowledge that this is only the first part of a continuing conversation.

Again, this isn’t a guide to how every conversation should ideally go. Not every conversation deserves your maximal attention. And the Schelling choice is rabbit, not stag.

Maybe it’s possible to 8020 this. Perhaps it’s possible to be someone who has deep conversations even if they’re brief and have no certainty of being continued later. Maybe you can get a lot of the benefit by merely giving off the feeling that you might, if only you had more time, listen and participate deeply in the conversation. Maybe you can find a way to get away with reversing some or all of the advice I gave in “Dealing with Noise”—raise your epistemic expectations, remember all the points, don’t accept being unheard or misunderstood. Just give off an aura of reasonableness except instead of making people avoid dramatic expressions of emotion, it makes them feel that you’re willing to hunt stag in the conversation.

If so, let me know what the trick is.