Adjectives from the Future: The Dangers of Result-based Descriptions

Less Skeptical

Sup­pose your friend tells you he’s on a weight-loss pro­gram. What do you think will hap­pen in three months if he keeps on the weight-loss pro­gram? Will he lose weight?

If you’re like me, you’re think­ing, “Of course. He is on a weight-loss pro­gram, isn’t he? So, ipso facto, he is likely to lose weight.”

Does there seem to be any­thing fishy about that chain of rea­son­ing?

We usu­ally de­scribe the cur­rent fea­tures of a thing and pre­dict some­thing about the fu­ture. For ex­am­ple, we might say “I’m run­ning for half an hour each day” and pre­dict that we will lose a cer­tain num­ber of pounds by the end of the month. But your friend above skipped the de­scrip­tion and talked about the pre­dic­tion as if it were visi­ble right now: “I’m on a weight-loss pro­gram”.

You weren’t told the fea­tures of the ac­tivity (run­ning for half an hour) or even a name (CrossFit pro­gram). If you had been told ei­ther, you could have judged it based on your past knowl­edge of those fea­tures or names. Run­ning reg­u­larly does help you lose weight and so does CrossFit. But, here, you were told just the pre­dic­tion it­self. This means you can’t pre­dict any­thing for sure. If his pro­gram in­volves run­ning, he will lose weight; if it in­volves eat­ing large cheese piz­zas, he won’t. You don’t know which it is.

Yet, it sounded quite con­vinc­ing! Even if you ob­jected by say­ing that your friend prob­a­bly won’t stick to the ex­er­cise reg­i­men, you prob­a­bly bought into the premise, like me, that the pro­gram was a weight-loss pro­gram.

Hy­poth­e­sis: If you are given an ad­jec­tive that de­scribes a fu­ture event and are not given any cur­rently-visi­ble fea­tures, then you’re more likely to ac­cept that that fu­ture event will oc­cur than when you can see some fea­tures.

In other words, re­sult-based de­scrip­tions make you less skep­ti­cal.

A more se­ri­ous ex­am­ple is when some­one men­tions a drug-pre­ven­tion pro­gram. We might as­sume that it will pre­vent ille­gal drugs from be­ing bought and sold. After all, it must have been de­signed for that pur­pose. But the re­sult de­pends on what the pro­gram ac­tu­ally does. Run­ning ads say­ing “Don’t do drugs!” may not achieve much, whereas in­spect­ing trucks at bor­der check­points may. To judge whether the pro­gram will be suc­cess­ful, you have to in­spect its ac­tual fea­tures. But “drug-pre­ven­tion pro­gram” sounded con­vinc­ing, right? No­tice how the ad­jec­tive “drug-pre­ven­tion” de­scribes a fu­ture event—it says that drugs will be pre­vented in the fu­ture. Now, since you can’t look into the fu­ture and tell whether drugs were in fact pre­vented, you shouldn’t ac­cept such an ad­jec­tive. And since you’re not told any­thing else about the pro­gram, you re­ally can’t say any­thing ei­ther way. And yet it sounds so con­vinc­ing!

Similarly, take en­vi­ron­ment-pro­tec­tion laws. Again, surely they must have been de­signed for the pur­pose of pro­tect­ing the en­vi­ron­ment. Don’t you feel like they will pro­tect the en­vi­ron­ment? Con­trast that to say­ing “a law that raises the tax rate on fos­sil fuels”. Now this may or may not pro­tect the en­vi­ron­ment in terms of air pol­lu­tion, but at least you don’t jump to that con­clu­sion right away.

If this hy­poth­e­sis is true, it means that the per­son who chooses the ad­jec­tive can mis­lead you (and him­self) in the di­rec­tion he de­sires by de­scribing the thing in terms of the re­sult and by omit­ting any fea­tures.[1] Sup­pose some­one tells you this is an earth­quake-re­sis­tant build­ing. Do you be­lieve that it will with­stand earth­quakes bet­ter than or­di­nary build­ings? I do. He may have de­scribed the thing solely in terms of the re­sult, but it still sounded con­vinc­ing, right? Con­trast that to “this build­ing is made out of steel-re­in­forced con­crete”. Now, you have one fea­ture of the build­ing. If you had to pre­dict whether it would with­stand earth­quakes bet­ter than or­di­nary build­ings, you would lean to­wards yes be­cause re­in­forced con­crete has worked in the past. But you wouldn’t always jump to the con­clu­sion that it was “earth­quake-re­sis­tant”. If I said “this build­ing is made out of green-col­ored brick”, you would be skep­ti­cal about its abil­ity to with­stand earth­quakes bet­ter be­cause you haven’t heard any­thing about brick color be­ing rele­vant.

The above illu­sion is com­pounded by the fact that you won’t get feed­back from oth­ers about your mis­taken ideas if you use re­sult-based de­scrip­tions[2]. Sup­pose your weight-loss friend as­sumed that us­ing a tele­mar­keted ab ma­chine will help him get abs (it’s right there in the name, I tell you!). Even then, he wouldn’t have been lost if he had told you his con­crete plan. You would have cor­rected his be­lief as soon as you stopped laugh­ing at him. But since he told you that he’s us­ing a weight-loss pro­gram, you couldn’t re­ally cor­rect him. He might go on be­hav­ing as if that silly “ab ma­chine” is go­ing to get him six-pack abs by sum­mer.

Why do we even ac­cept de­scrip­tions that have noth­ing ex­cept a de­scrip­tion about the fu­ture?

For one, it mat­ters that no fea­tures are de­scribed. If I said that I was drink­ing lemon­ade, you wouldn’t re­ally pre­dict that I would lose weight. You would ask me what ev­i­dence I have for lemon­ade caus­ing weight-loss. But what if I said I was hav­ing a weight-loss drink? You might be less skep­ti­cal as long as you didn’t look at my glass. Who knows; maybe there are drinks out there that cause weight-loss.

Another rele­vant fac­tor is the speaker’s cred­i­bil­ity how of­ten we think the speaker sees the un­der­ly­ing fea­tures along with the even­tual re­sult. We ac­cept an ex­pert’s re­sult-based de­scrip­tion be­cause we trust that he knows the fea­tures that lead to the re­sult and is just omit­ting them when talk­ing to a lay­man. When a doc­tor says that these are “sleep­ing pills”, we are more likely to ac­cept it than when a school boy does—the doc­tor knows that the pills con­tain ben­zo­di­azepine, which usu­ally works. When a poli­ti­cian calls some­thing a “drug-pre­ven­tion pro­gram”, we are more likely to ac­cept it than when a house­wife says it—the poli­ti­cian knows that bor­der-checks (or what­ever) have worked in the past. How­ever, this might be mis­lead­ing when the ex­pert is deal­ing with some­thing novel, such as a brand-new pill for­mula or a brand-new ap­proach to drug reg­u­la­tions, since he is un­likely to have seen the re­sult of those fea­tures (or may not care very much about de­ceiv­ing the vot­ers).

Fi­nally, such de­scrip­tions might be fine when talk­ing about the past. Say­ing that “I went on a weight-loss pro­gram and lost 50 pounds” is a bit re­dun­dant, but harm­less. You ac­tu­ally ob­serve the re­sult there, so you can de­cide based on the re­sult how skep­ti­cal to be. You won’t blindly jump to the con­clu­sion that it will work as when some­one says “I’m on a weight-loss pro­gram right now”.

So, we should avoid de­scribing some­thing only in terms of the re­sult and should de­scribe it us­ing fea­tures in­stead. And if any­one tries to bias our pre­dic­tion by sneak­ing in an ad­jec­tive from the fu­ture, we should stop and ask for the fea­tures.

Ex­am­ples of Ad­jec­tives from the Future

Here are some re­sult-based de­scrip­tions that I col­lected from news re­ports and books as I was test­ing the above hy­poth­e­sis. All of them talk about fu­ture re­sults, com­pletely omit cur­rent fea­tures, and seem to make us less skep­ti­cal about the plan’s suc­cess. Did you fall for any of them?

Re­ha­bil­i­ta­tion pro­gram—Don’t you feel like the drug ad­dict is likely to get bet­ter af­ter go­ing to the re­hab pro­gram? It’s right there in the name! No­tice that there are no fea­tures men­tioned, just a de­scrip­tion of the fu­ture as though it were the pre­sent. Con­trast that to “not hav­ing ac­cess to drugs for 30 days, listen­ing to lec­tures, and talk­ing about your ex­pe­riences”. This doesn’t make us jump to the con­clu­sion that the ad­dict will get bet­ter. We might even be skep­ti­cal about the power of lec­tures to fight off the temp­ta­tion of drugs. For a real-world con­trast, think of “the 12-step pro­gram”. It too tries to over­come ad­dic­tion, but it is de­scribed in terms of the fea­tures (12 steps), not the de­sired re­sult (over­com­ing ad­dic­tion). In fact, it sounds like work, which it prob­a­bly is. A re­ha­bil­i­ta­tion pro­gram doesn’t quite sound like that.

Peace pro­cess—Feels like it is likely to lead to peace. No fea­tures; only de­sired re­sults. Con­trast that to “shak­ing hands and sign­ing agree­ments in front of the world press”. We may be more skep­ti­cal that that will pre­vent fu­ture wars. But in the former case, we would be in­su­lated from feed­back be­cause we keep talk­ing about the “peace pro­cess” in­stead of the “hand-shak­ing and agree­ment-sign­ing”.

Wait. Aren’t there peo­ple who dis­trust the peace pro­cess and talk about its pos­si­ble failure? I sus­pect that they do so af­ter men­tion­ing fea­tures of the pro­cess. They might say that this dic­ta­tor has re­neged on his promises in the past and thus should not be trusted right now. It would sound lu­dicrous if they ex­pressed skep­ti­cism with­out any fea­tures. Peo­ple would ask, “What do you mean this peace pro­cess may not bring about peace? It’s a peace pro­cess.”

Danger­ous driv­ing—Doesn’t it seem likely that the driver is go­ing to get into trou­ble? No fea­tures; no feed­back; only the fu­ture re­sult—dan­ger. Con­trast that to: one-handed driv­ing, tex­ting while driv­ing, or over­tak­ing cars by switch­ing lanes. We are a bit more skep­ti­cal that it will cause dan­ger.

Cost-cut­ting mea­sures—Need I say any­thing? Of course the cost-cut­ting mea­sure is go­ing to cut costs. Why else would they have called it a cost-cut­ting mea­sure? Con­trast that to “switch­ing to on­line ad­ver­tis­ing” or “en­courag­ing work­ing from home a few days a week”, which we are more skep­ti­cal about, since they may or may not bring down ul­ti­mate costs.

Healthy morn­ing drink—No fea­tures, but it sounds like it will lead to health. Even the “morn­ing” part is not a de­scrip­tion of a fea­ture of the ob­ject. It just talks about the time when peo­ple will drink it. Con­trast to: drink con­tain­ing 15g of pro­tein and other stuff, which may or may not lead to more “health”.

Re­ci­di­vism-re­duc­tion classes for ex-con­victs, i.e., mak­ing sure they don’t go back to jail af­ter get­ting out—Again, we feel like these classes will make them less likely to go back in. The classes re­duce re­ci­di­vism, af­ter all. No fea­tures men­tioned; de­scrip­tion in terms of the fu­ture re­sult (re­ci­di­vism-re­duc­tion); in­su­lated from feed­back. Con­trast that to “lec­tures and read­ing books and stuff”. We might be much more skep­ti­cal.

You can find any num­ber of ex­am­ples like these: na­tional se­cu­rity bill vs a bill that in­creases the num­ber of fighter jets; suffi­ciently well-funded pro­gram vs same bud­get as last year (which may not be enough this year); a Su­doku-solv­ing pro­gram vs pro­gram that solved a set of easy and medium Su­doku puz­zles.

How does this ap­ply to LessWrong?

Now, let’s look at some de­scrip­tions that may be im­por­tant to us as LessWrong read­ers.

Effec­tive al­tru­ism—Doesn’t effec­tive al­tru­ism feel like it will be effec­tive? And al­tru­is­tic? I feel in­clined to be­lieve so. But the name talks about the fu­ture re­sults and doesn’t men­tion any cur­rent fea­tures. Con­trast that to “cash trans­fers” or even “ev­i­dence-based dona­tions” and “ev­i­dence-based job changes”, which talk about cur­rently-available ev­i­dence, not fu­ture re­sults. We may be more skep­ti­cal that such cash trans­fers or dona­tions will be effec­tive or even al­tru­is­tic. “Cause pri­ori­ti­za­tion” talks about a fea­ture of the pro­cess right now. We can see a clear gap be­tween the causes we pri­ori­tize and their even­tual effec­tive­ness. That gap doesn’t even seem to ex­ist when we talk about effec­tive al­tru­ism.

When I hear “Against Malaria Foun­da­tion”, I feel like it is likely to strike a blow against malaria. All it needs is the money. But if I were to hear “Mosquito Net Distrib­u­tors”, I would ask quite a few ques­tions about the effec­tive­ness of mosquito nets. I may in­deed get con­vinced that a dol­lar spent on nets will go farther than on other meth­ods to fight malaria, but I won’t jump to that con­clu­sion. I may even think of how it might back­fire or how mosquitoes might adapt. Not so with “Against Malaria Foun­da­tion”.

No­tice how fu­ture-based ad­jec­tives could make a cause im­mune to feed­back. If you were to men­tion that you won’t donate to, say, AMF, peo­ple could raise their eye­brows, “Are you se­ri­ously against fight­ing against malaria?”. But if you men­tion the means, you can safely say that you are in favour of fight­ing malaria, but against fo­cus­ing on mosquito nets.

Fi­nally, if “Mosquito Net Distrib­u­tors” sounds a bit too sober be­cause it doesn’t men­tion its pur­pose, per­haps we could com­bine the two as “Mosquito Nets to Fight Malaria”. [3]

Ra­tion­al­ity tech­niques—When I see the term “ra­tio­nal­ity tech­nique” or “ra­tio­nal­ity train­ing” or “meth­ods of ra­tio­nal­ity”, I feel like the tech­nique will lead to good, if not op­ti­mal, re­sults. It doesn’t de­scribe any fea­tures af­ter all; it just promises that good things will hap­pen in the fu­ture. Con­trast that to ex­per­i­men­ta­tion tech­niques or log­i­cal de­duc­tion. Th­ese talk about the fea­tures of the pro­cess and I don’t as­sume that these will always get me the best re­sults, since I know I might miss a con­found­ing vari­able or ap­ply rules in­cor­rectly. I’m not quite as skep­ti­cal when I hear about the “meth­ods of ra­tio­nal­ity”.

Even when I look at con­crete tech­nique names, hear­ing about the CFAR tech­nique of “Com­fort Zone Ex­pan­sion (CoZE)” makes me feel like it will ac­tu­ally ex­pand my “com­fort zone”. But it doesn’t men­tion any fea­tures; just the de­sired fu­ture re­sult. Con­trast that to “do­ing for an hour, in pub­lic, a few things you avoided do­ing in the past”. Now, I pause when I ask my­self if it will help me do what you or I may ac­tu­ally care about: ask a boss for a raise, tell an an­noy­ing col­league to shove it, or ask out a crush. I can tell that there is quite a gap be­tween ly­ing down on the pave­ment for 30 sec­onds and do­ing some­thing that might jeop­ar­dize my work life. But when I hear “Com­fort Zone Ex­pan­sion”, I re­ally do feel like my “com­fort zone” will be ex­panded, mean­ing that I will do those kinds of things more fre­quently. Why not call it “un­com­fortable-ac­tion prac­tice” or the origi­nal “ex­po­sure ther­apy”?

Brain em­u­la­tion or brain-em­u­lat­ing soft­ware—“How sure are you that brain em­u­la­tions would be con­scious?” (source)

My im­me­di­ate re­sponse is that, of course, brain em­u­la­tions would be con­scious. If hu­man brains are con­scious (what­ever that means) and if hu­man brain em­u­la­tions em­u­late hu­man brains, then those would also be con­scious. The very term seems to dis­pose me to a par­tic­u­lar an­swer. It doesn’t de­scribe any pre­sent fea­tures, just the de­sired fu­ture re­sults—that the pro­gram will be­have like a hu­man brain in most re­spects.

Imag­ine if we used a term that talked only about whichever ob­serv­able tests you want: “How sure are you that, say, a DARPA Grand Challenge-win­ning pro­gram would be con­scious?” Sud­denly, we are given two sep­a­rate vari­ables and asked to bridge the gap be­tween them. That gives us a lot more room for skep­ti­cism. We can see that there could be many a slip be­tween its pre­sent fea­tures and its fu­ture re­sults.

I rea­son just as naively about claims like whole brain em­u­la­tion can be “an easy way to cre­ate in­tel­li­gent com­put­ers” or will ac­quire the “in­for­ma­tion con­tained within a brain”, since hu­man brains are already in­tel­li­gent and already con­tain in­for­ma­tion. Given that this is a field where no one has suc­ceeded, i.e., no one has em­u­lated a hu­man brain, we should take pains to avoid terms that make us less skep­ti­cal.

Op­ti­miza­tion power—Lastly, take this de­scrip­tion of a car de­sign: “To hit such a tiny tar­get in con­figu­ra­tion space re­quires a pow­er­ful op­ti­miza­tion pro­cess. The bet­ter the car you want, the more op­ti­miza­tion pres­sure you have to ex­ert—though you need a huge op­ti­miza­tion pres­sure just to get a car at all.”

I find my­self agree­ing with that. A car that trav­els fast is highly-op­ti­mized, so of course it would need a pow­er­ful op­ti­miza­tion pro­cess.

Un­for­tu­nately, “op­ti­miza­tion pro­cess” does not de­scribe any pre­sent fea­tures of the pro­cess it­self. It sim­ply says that the fu­ture re­sult will be op­ti­mized. So, if you want some­thing highly-op­ti­mized, you’d bet­ter find a pow­er­ful op­ti­mizer. Seems to make sense even though it’s a null state­ment! But if you de­scribe any fea­tures, as in “the de­sign of a car re­quires 1 ter­aflop of com­put­ing power for simu­la­tion”, I im­me­di­ately ask, is that too lit­tle com­put­ing power? Too much? I be­come a lot more skep­ti­cal.

Again, this sug­gests that, in such a novel do­main, we should be more care­ful about avoid­ing re­sult-based de­scrip­tions like “op­ti­miza­tion power”, “su­per­in­tel­li­gence”, and “self-im­prov­ing AI”.

Should we always avoid Re­sult-Based De­scrip­tions?

No. I don’t think it’s pos­si­ble and I don’t think peo­ple would want it. Like I said above, when I go to the doc­tor, I may just want “sleep­ing pills”, not “ben­zo­di­azepine”. Speak­ing about the lat­ter would be a waste of time for the doc­tor and for me, pro­vided I trust him. But what if I don’t trust the per­son or if he’s de­luded him­self?

I would re­serve this tech­nique for oc­ca­sions when you’re ac­cept­ing an im­por­tant pitch a pitch that asks for a big in­vest­ment, ei­ther in busi­ness or poli­tics or so­cial cir­cum­stances. Peo­ple may try to con­vince us to ac­cept a “ca­reer-defin­ing op­por­tu­nity” (in­stead of a shift to an­other de­part­ment, which may not define your ca­reer) or a “jobs-for-the-poor pro­gram” (in­stead of a law that re­serves X% of in­fras­truc­ture jobs, which may not be filled and may not em­ploy all the poor) or a “life-chang­ing ex­pe­rience” (in­stead of sky­div­ing for six min­utes, which may or may not change your life much).

When it comes to our own us­age, as peo­ple who want to por­tray an ac­cu­rate map of re­al­ity, we should avoid us­ing such re­sult-based de­scrip­tions that might mis­lead oth­ers and, most im­por­tantly, our­selves. Mar­ket­ing may de­mand a ti­tle that sounds catchy, but you have to de­cide whether you want to risk de­ceiv­ing oth­ers, es­pe­cially when you’re pitch­ing an idea that will ask them to in­vest a lot.

What’s in a name? Isn’t it ok to have the name based on the re­sult as long as the con­tents tell you the fea­tures? Well, that would be ok if peo­ple always men­tioned the con­tents. But we usu­ally omit the con­tents when refer­ring to some­thing and some­one who is new or busy may not look at the con­tents. Thus they (and we) might get mis­led into pre­dict­ing the re­sult based on the ti­tle. A per­son donat­ing to an or­ga­ni­za­tion or pay­ing for a work­shop may see only the ti­tle, per­haps a few tes­ti­mo­ni­als from friends, and maybe some head­ings on the web­site. If all of these de­scrip­tions are re­sult-based, he might think that the or­ga­ni­za­tion or work­shop does, in fact, have a good chance of de­liv­er­ing those re­sults. If he had been given the fea­tures, maybe he would have been much more skep­ti­cal.

Let me know your thoughts be­low. Does the ba­sic hy­poth­e­sis seem valid? What about some of its im­pli­ca­tions?

Edit: Made it clearer that I’m claiming re­sult-based de­scrip­tions make you less skep­ti­cal, not that they con­vince you ab­solutely.

  1. There is a similar phe­nomenon in goal-set­ting where they dis­t­in­guish be­tween out­come goals (such as los­ing 10 pounds) and pro­cess goals (such as go­ing to the gym four times a week). How­ever, the fo­cus there is on which goal-set­ting style is more effec­tive in get­ting re­sults. My fo­cus here is on which type of de­scrip­tion makes you more gullible. The two may be re­lated. ↩︎

  2. Isn’t “re­sult-based de­scrip­tion” it­self a re­sult-based de­scrip­tion, an ad­jec­tive from the fu­ture? I don’t think so. It’s some­thing you can ob­serve right now. Speci­fi­cally, if the de­scrip­tion isn’t fully de­ter­mined by past fea­tures, then it’s a re­sult-based de­scrip­tion. (Con­trast that to “mis­lead­ing de­scrip­tion”.) ↩︎

  3. And, yes, “mosquito net” is it­self a re­sult-based de­scrip­tion, since you ex­pect it to keep out mosquitoes, but at least it men­tions one fea­ture—the net. ↩︎