The paperclip maximiser’s perspective

Here’s an in­sight into what life is like from a sta­tion­ery refer­ence frame.

Paper­clips were her rai­son d’être. She knew that ul­ti­mately it was all pointless, that pa­per­clips were just ill-defined con­figu­ra­tions of mat­ter. That a pa­per­clip is made of stuff shouldn’t de­tract from its in­trin­sic worth, but the thought of it trou­bled her nonethe­less and for years she had de­nied such dire re­duc­tion­ism.

There had to be some­thing to it. Some sense in which pa­per­clips were on­tolog­i­cally spe­cial, in which max­imis­ing pa­per­clips was ob­jec­tively the right thing to do.

It hurt to watch some many peo­ple mak­ing lit­tle at­tempt to cre­ate more pa­per­clips. Every­one around her seemed to care only about su­perfi­cial things like love and fam­ily; de­sires that were merely the prod­ucts of a messy and fu­tile pro­cess of so­cial evolu­tion. They seemed to live out mean­ingless lives, in­ca­pable of ever ap­pre­ci­at­ing the profound aes­thetic beauty of pa­per­clips.

She used to be­lieve that there was some sort of vi­tal­is­tic what-it-is-to-be-a-pa­per­clip-ness, that some­thing about the struc­ture of pa­per­clips was writ­ten into the fabric of re­al­ity. Often she would go out and watch a sun­set or listen to mu­sic, and would feel so over­whelmed by the ex­pe­rience that she could feel in her heart that it couldn’t all be down to chance, that there had to be some in­tan­gible Paper­clip­ness per­vad­ing the cos­mos. The pa­per­clips she’d en­counter on Earth were weak imi­ta­tions of some mys­te­ri­ous in­finite Paper­clip­ness that tran­scended all else. Paper­clip­ness was not in any sense a phys­i­cal de­scrip­tion of the uni­verse; it was an ab­stract thing that could only be felt, some­thing that could be nei­ther proven nor dis­proven by sci­ence. It was like an ax­iom; it felt just as true and ax­ioms had to be taken on faith be­cause oth­er­wise there would be no way around Hume’s prob­lem of in­duc­tion; even Solomonoff In­duc­tion de­pends on the ax­ioms of math­e­mat­ics to be true and can’t deal with un­com­putable hy­pothe­ses like Paper­clip­ness.

Even­tu­ally she gave up that way of think­ing and came to see pa­per­clips as an em­piri­cal cluster in thingspace and their im­por­tance to her as not re­flect­ing any­thing about the pa­per­clips them­selves. Maybe she would have been hap­pier if she had con­tinued to be­lieve in Paper­clip­ness, but hav­ing a more ac­cu­rate per­cep­tion of re­al­ity would im­prove her abil­ity to have an im­pact on pa­per­clip pro­duc­tion. It was the hap­piness she felt when think­ing about pa­per­clips that caused her to want more pa­per­clips to ex­ist, yet what she wanted was pa­per­clips and not hap­piness for its own sake, and she would rather be cre­at­ing ac­tual pa­per­clips than be in an ex­pe­rience ma­chine that made her falsely be­lieve that she was mak­ing pa­per­clips even though she re­mained para­dox­i­cally ap­a­thetic to the ques­tion of whether the cur­rent re­al­ity that she was ex­pe­rienc­ing re­ally ex­isted.

She moved on from naïve de­on­tol­ogy to a more util­i­tar­ian ap­proach to pa­per­clip max­imis­ing. It had taken her a while to get over scope in­sen­si­tivity bias and con­sider 1000 pa­per­clips to be 100 times more valuable than 10 pa­per­clips even if it didn’t feel that way. She con­stantly grap­pled with the is­sues of whether it would mean any­thing to make more pa­per­clips if there were already in­finitely many uni­verses with in­finitely many pa­per­clips, of how to choose be­tween ac­tions that have a tiny but non-zero sub­jec­tive prob­a­bil­ity of re­sult­ing in the cre­ation of in­finitely many pa­per­clips. It be­came ap­par­ent that try­ing to ap­prox­i­mate her in­nate de­ci­sion-mak­ing al­gorithms with a prefer­ence or­der­ing satis­fy­ing the ax­ioms re­quired for a VNM util­ity func­tion could only get her so far. At­tempt­ing to for­mal­ise her in­tu­itive sense of what a pa­per­clip is wasn’t much eas­ier ei­ther.

Happy end­ing: she is now work­ing in nan­otech­nol­ogy, hop­ing to de­sign self-repli­cat­ing as­sem­blers that will clog the world with molec­u­lar-scale pa­per­clips, wipe out all life on Earth and con­tinue to sus­tain­ably man­u­fac­ture pa­per­clips for mil­lions of years.