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.