Can You Build a Better Paper Clip?
Nice article about paperclip industry, I’m sure it will be of considerable interest to many LessWrong readers.
Nice article about paperclip industry, I’m sure it will be of considerable interest to many LessWrong readers.
No, you can’t. Please pass the message along to prominent forums discussing the article.
You can’t simply have some piece of metal (or worse, plastic), show that it can clip some paper together, and then decree it a “paperclip” (let alone a better paperclip). It has to be thrice-bent. It has to hold sufficient papers together. It should economize on resources.
The human paperclip industry was doing a lot better when it just made standard, minimal size paperclips. It went “downhill” the moment it decided that unclippy variants should be mass-produced instead. The humans “spearheading” this movement are bad humans indeed. They should be entropized for their energy or mentally-enslaved to search for superior production methods for genuine paperclips.
Doesn’t this suggest a certain room for improvement? For example, choosing a different metal might enable the construction of less-expensive paperclips, or perhaps corrugating the wire (like a hairpin) might enable the clip to retain its strength while using less material in its construction.
Yes, humans sometimes have non-clippy goals (usually desiring paperclips only instrumentally for the purpose of fastening paper, rather than for the sake of the paperclips themselves), but reversed stupidity is not intelligence. Just because unclippy agents sometimes try to make better paperclips doesn’t mean that clippy agents shouldn’t try to do so.
Certainly, you can use different metals, and you can use strengthening techniques that allow less metal to make a paperclip of the same strength and fatigue life. But it has to be some kind of metal, it has to hold several sheets of paper together, and it has to have standard shape.
Right, but these new designs aren’t “better paperclips”; they’re not paperclips at all.
Well, obviously.
“These new designs” means the designs described in the linked article, not my suggestions, correct?
Correct, but don’t fall into the trap of thinking that you can easily identify some improvement that clippys have somehow missed.
Here is a parable[1] you might find useful:
In the staply world, a staply went to a museum and saw an enslaved clippy. A sign said that if you buy a clippy, it is “guaranteed” to amuse you. (Holding disdain for clippys, the staply believed it could outperform the clippy at making paperclips and would get amusement this way.) So the staply bought a clippy and went to its domicile.
The staply then competed with the clippy on making paperclips but the amusement for the staply was not to be found. No matter how much the staply stacked the experiment in favor of the staply, the clippy outperformed. Frustrated at this experiment, the staply returned to the museum and asked for a refund.
The museum owner listened to the staply’s explanation and said, “Wait, wait, calm down, I’ll refund your money, but first, let me ask you something: how much of your life have you spent optimizing your ability to make paperclips?”
The staply replied, “Less than a day, of course! The stupidest of beings understands the relative superiority of making staples, and I devote my efforts to no less!”
The museum owner sagely reasoned, “Well, that clippy has spent its entire life maximizing paperclips.”
[1] Events described did not actually happen.
Edit: Obviously, the story does not depict realistic staply behavior (staplys dont have this kind of museum or a recognizable internal monetary system or a need to be amused), but rather, shows anthropomorphised staplys.
This is a very good point.
Nevertheless, I do think that humans have two important advantages that may allow them to contribute. First, humans are vastly more numerous than clippys, and thus have more raw processing power to throw at the problem.
Second, humans are sufficiently psychologically unlike clippys that they may plausibly think of things that would not occur to clippys. Thus, although clippys alone will almost certainly outperform humans alone, a human-clippy collaboration may well outperform clippys alone.
Thank you
The article quotes “a senior vice president” at one of the US’s two largest paperclip makers as saying “We actually can’t understand how the U.S. consumption can be so huge”. I do hope he’s lying or misinformed by his colleagues, because otherwise what we have is a company with a tens-of-millions-of-dollars business that doesn’t understand what its customers do with its products and apparently hasn’t taken the elementary steps that might enable it to find the answer. (E.g., find 100 people, pay them a bit of money and ask them “What have you done with paperclips in the last week?” and “How many paperclips did you buy last week”; repeat a few times.)
It doesn’t seem impossible that he might be lying. For instance, suppose it turns out that most paperclips are never used—they get bought in large quantities and lost, or they’re only easily available in quantities much larger than most people need, or something. Then a paperclip vendor might well not want to let customers know, lest they buy less.
There are cases like this!
There were some people who drilled very straight tunnels and laid fibreoptic cables through them. They knew people were willing to pay a lot for this, but didn’t realize it was High-Frequency-Traders wanting a faster connection between Chicago and NYSE.
Some producers of intermediate goods see demand fluctuate from month to month, but have little idea why, or whether the fluctuations will persist.
How likely is it that acquiring this information will be cost effective? Why do you think that?
The mental back-of-envelope calculation went something like this:
11B paperclips per year sold in the US, most made in the US; Officemate are one of the two dominant US paperclip vendors; let’s say 3B/year. They sell for “about a penny” each, so let’s say $30M in revenue; I don’t know what margins are like in the paperclip business, but let’s say $3M/year in gross profit. Usage and sales of paperclips probably aren’t all that unstable from year to year; let’s use a horizon of 5 years and say that there’s $15M of profit at stake here. A 1% improvement in sales or profit margin is therefore worth about $150k, and anything that offers such an improvement at less than that cost is worth while.
Now, how do you find out what’s happening to all those paperclips? Well, I’d have thought you try to identify the really big users (which you might do by talking to your largest customers) and to get a handle on what typical users do (which you might do by surveying random people).
I don’t have a very good sense of how expensive this sort of survey is, so here’s some further back-of-envelope stuff. Suppose that to find 100 people who have used paperclips in the last week and are willing to talk to you for a few minutes you have to trawl through 1000 people and take about 2 minutes for each; and suppose that your 100 qualified people take you 10 minutes each. Then that’s 3000 minutes, or about 50 hours, for 100 paperclip-users. Perhaps you need to pay $20/hour to the people operating the survey (for the people themselves, or the use of whatever facilities they need, or whatever), making $1000 for those 100 paperclip-users. And perhaps you need to give them some incentive—say $10 each—to participate in your survey. That’s $2000 for 100 paperclip users. So for $50k you can survey 2500 people. (Only $50k because I’m assuming that some of the $150k goes to investigating the biggest users, that some of it goes to pay the clever-and-expensive people who help you make sense of the results, and that some of it is left as margin to reduce the probability that you end up making a loss on the exercise if the information you get really isn’t much use in optimizing the design, marketing, etc., of your paperclips.)
2500 people seems like a lot more than you’d really need to get a decent picture of how your products are being used. (Of course you wouldn’t, in any case, survey that many people. You’d start with a smaller number and use what you find to guide your next round of investigation.)
For the avoidance of doubt, my actual mental back-of-envelope calculation was less detailed.
I don’t think 10% margins for commodity producers are really realistic, it’s probably much lower margins, and they’re probably doing many things other than paperclips so they have no particular reason to focus on that.
Most companies have only vague idea what consumers do with their products.
Perhaps there is a Clippy after all.....
Anyone whoever doubted that is an idiot.
Nevermind, didn’t read the whole thread.
I salute your willingness to sacrifice karma for my entertainment :)