Effective Altruism Book Review: Radical Abundance (Nanotechnology)

Book Re­view: Rad­i­cal Abundance

I. Introduction

As a ma­te­ri­als en­g­ineer­ing ma­jor with, roughly speak­ing, a year of full-time ex­pe­rience with molec­u­lar dy­nam­ics simu­la­tions, I have a spe­cial place in my heart for high im­pact ma­te­ri­als both literal and figu­ra­tive. As a lurker in effec­tive al­tru­ism, I was delighted to see some­thing po­ten­tially rele­vant to my ex­pe­rience show up. I made a post in the “Nan­otech­nolgy in EA” Face­book group ex­press­ing/​re-iter­at­ing the fol­low­ing thoughts:

  1. Nan­otech­nol­ogy will be chaotic and hard-to-pre­dict, re­quiring mas­sive ad­vance­ments in com­pu­ta­tion be­fore effec­tive ma­chines can be de­signed.

  2. Our tech­nol­ogy in terms of ac­tu­ally pro­duc­ing APM seems plau­si­ble but very un­cer­tain at this point

From my post, I got some in­ter­est­ing com­ments sug­gest­ing an al­ter­na­tive to 1. along with a recom­men­da­tion to read K. Eric Drexler’s new book on atom­i­cally pre­cise man­u­fac­tur­ing (APM), Rad­i­cal Abun­dance: How a Revolu­tion in Nan­otech­nol­ogy Will Change Civ­i­liza­tion. This, in part with some googling, gave me the im­pres­sion that I was also wrong about 2.

The goals of this book re­port from least to most sig­nifi­cance are: pro­vid­ing a de­cent sum­mary of Rad­i­cal Abun­dance, ex­press­ing my own thoughts on what I find to be the most salient as­pects of nan­otech­nol­ogy, and clar­ify­ing pos­si­ble com­mon mis­con­cep­tions that con­tribute to the fre­quent con­fu­sion about nan­otech­nol­ogy’s place in effec­tive al­tru­ism.

Any­way, Rad­i­cal Abun­dance lightly dis­cusses three in­ter-wo­ven con­cepts: how APM would work, how APM could im­pact the world, and how APM can be achieved. Let’s do this.

II. How APM Works

A. Ad­vanced APM in a Nutshell

The de­scrip­tion and un­der­stand­ing of APM can be best char­ac­ter­ized by Drexler’s own words:

APM-based ma­te­ri­als pro­cess­ing tech­nol­ogy will em­ploy nanoscale me­chan­i­cal de­vices that op­er­ate at high fre­quen­cies and pro­duce pat­terns of atoms...think of an APM sys­tem as a kind of printer that builds ob­jects out of atoms just as a printer builds images out of pat­terns of ink, con­strained by a limited gamut, not of col­ors, but of out­put ma­te­ri­als.

This clar­ifies two dis­tinct points that I, as some­one work­ing in molec­u­lar dy­nam­ics and , would miss. First, de­spite my pri­ors from ex­pe­rience with soft mat­ter molec­u­lar dy­nam­ics, APM sys­tems will be based on me­chan­i­cally-in­spired rigid mechanisms rather than biolog­i­cally-in­spired soft mat­ter ma­chines. Se­cond, at its heart, most APM will fo­cus on the fabri­ca­tion of macro-scale ma­te­ri­als with nano-scale op­ti­miza­tion (i.e. re­ally good ma­te­ri­als) rather than nanoma­chines (i.e. tiny robots), though these will be nec­es­sary to some ex­tent in cre­at­ing APM in the first place and have mas­sive periph­eral uses and im­pli­ca­tions.

B. Me­chan­i­cal De­vices in Nanotechnology

First, while ex­ist­ing ex­am­ples of APM (i.e. ri­bo­somes) are biolog­i­cal in na­ture and demon­strate the po­ten­tial for lev­er­ag­ing nano-scale de­vices in com­plex ways, Drexler’s APM uses down-scaled me­chan­i­cal de­vices like those in mod­ern fac­to­ries to achieve its ends in­stead of the chaotic com­plex­ity of biol­ogy. In or­der to suc­cess­fully imi­tate the be­hav­ior of me­chan­i­cal ma­chines, nanome­chan­i­cal ma­chines will typ­i­cally be made of “sta­ble co­va­lent struc­tures that con­sist of fused rings; among hy­dro­car­bons, small-scale ex­am­ples in­clude the adaman­tanes and the some­what more flex­ible aro­matic molecules.” Think­ing along these lines si­mul­ta­neously opens up new pos­si­bil­ities and grounds oth­ers. When I think about biolog­i­cal APM, I think of spec­tac­u­larly com­plex in­ter-plays of atoms and molecules su­perbly op­ti­mized in par­allel with their en­vi­ron­ments over mil­lions of years to provide the suffi­cient but in­cred­ibly im­prob­a­ble con­di­tions for life. When I think about me­chan­i­cal APM, I think about the Saturn V Rocket.

To illus­trate the differ­ence, con­sider the fol­low­ing situ­a­tion where some­one asks me about the fu­ture of lev­er­ag­ing nan­otech­nol­ogy to achieve goal a goal in a given en­vi­ron­ment:

Me think­ing about biolog­i­cal nan­otech­nol­ogy: “Of course! We’ll use in­cred­ibly com­plex com­pu­ta­tional tools cou­pled with au­to­mated lab­o­ra­to­ries to build a black-box ma­chine that does the job you need… At least, we will if we ever figure out how to make com­put­ers and lab­o­ra­to­ries that effec­tive. Once we do though, the sky’s the limit!”

Me think­ing about me­chan­i­cal nan­otech­nol­ogy: “Hmm… It looks like achiev­ing your goal also re­quires some flex­ible com­po­nents or at least a fair num­ber of au­to­mated com­put­er­ized sys­tems so I’ll need to see if we can get ma­te­ri­als to work well with that at the nano-scale… And your en­vi­ron­ment is a bit differ­ent so I might need to se­lect differ­ent molecules to make sure they’re sta­ble and rigid un­less you want to in­cor­po­rate home­osta­sis which would add an­other or­der of com­plex­ity… Give me five years, a few mil­lions dol­lars and the atom­i­cally pre­cise tech I need to build some­thing and I’ll let you know.”

The differ­ence be­tween these biolog­i­cal and me­chan­i­cal char­ac­ter­i­za­tions of nan­otech­nol­ogy is si­mul­ta­neously com­pli­cated and sub­tle. Both seem a long ways off in terms of tech­nol­ogy but the former de­mands a clear con­cep­tual leap (much bet­ter com­put­ers) while the lat­ter re­quires an ex­ten­sive but plau­si­ble level of tech­nolog­i­cal progress (Im­proved knowl­edge of chem­istry and ex­ist­ing chem­istry tech­niques). This means that, while the former might never be achieved, the lat­ter will prob­a­bly be achieved even­tu­ally.

A similar point can also be made about po­ten­tial im­pact. Biolog­i­cal in­ter­pre­ta­tions of nan­otech­nol­ogy very quickly start hint­ing at mas­sive rev­olu­tions in nano-ma­chines that con­stantly re­spond and im­prove all as­pects of hu­man life and en­g­ineer­ing, ag­gre­gat­ing to­gether to form macro-level ma­chines of in­cred­ible com­plex­ity and com­pu­ta­tional ca­pac­ity. In con­trast, me­chan­i­cal in­ter­pre­ta­tions of nan­otech­nol­ogy are grounded in limited types of ma­te­ri­als in limited en­vi­ron­ments with limited ca­pa­bil­ities.

For the most part, these limi­ta­tions are the same for both macro-scale me­chan­i­cal de­vices and nano-ma­chines since both can be de­scribed in “fa­mil­iar, me­chan­i­cal terms.” How­ever im­por­tant differ­ences do ex­ist. Drexler notes that “a ma­chine can’t work well if its parts can’t move smoothly and in­ter­ac­tions be­tween atomic-scale bumps on sur­faces might seem to make smooth move­ment im­pos­si­ble.” This can be re­solved through de­sign­ing bump pat­terns to pro­duce su­per­lubric­ity but comes with its own con­straints. Drexler also notes the pres­ence of drag and ther­mal mo­tion but these forces are not very sig­nifi­cant in the con­text of rigid me­chan­i­cal struc­tures. I would also like to add an ad­di­tional con­straint: Be­cause “stretch­ing space and time in equal pro­por­tion scales prop­er­ties like mass, force, and ve­loc­ity in ex­actly the right way to make me­chan­i­cal mo­tion the same,” one quickly re­al­izes that grav­ity, pro­duc­ing an ac­cel­er­a­tion of 9.8 me­ters/​sec­ond down re­gard­less of size, van­ishes in im­por­tance. In this sense, nano-scale fac­to­ries will not be like the fac­to­ries seen on Earth but rather more like the ones de­signed in space. Ad­di­tion­ally, they will be driven purely by mo­tors and me­chan­i­cal de­vices with no in­cor­po­ra­tion of elec­tri­cal wiring (clas­si­cal elec­tri­cal en­g­ineer­ing breaks down at the nano-scale).

Ini­tially, these limi­ta­tions made me skep­ti­cal about whether APM could fea­si­bly cre­ate com­plex de­signer ma­te­ri­als or ex­tend to ad­vanced nano-ma­chines out­side of a man­u­fac­tur­ing con­text. How­ever, this video led me to re­al­ize that the sort of tar­get ma­te­ri­als to be con­structed could rely on ba­sic pat­terns of pick­ing up and putting down differ­ent atoms and molecules. Ad­di­tion­ally, be­cause the nano-scale is so small, rel­a­tively large nano-ma­chines can still be build at very low size-scales (though this nano-ma­chines are still what I am most skep­ti­cal about).

For the most part though, the out­puts of my biolog­i­cally-in­spired ex­pec­ta­tions of nan­otech­nol­ogy and Drexler’s me­chan­i­cal vi­sion seem similar. As Drexler says:

A SCIENTIST WROTE an ar­ti­cle about the nanoma­chines of the gen­eral sot I’ve de­scribed, but he sug­gested that they couldn’t be used in a biolog­i­cal en­vi­ron­ment be­cause bio­molecules would gum up gears and other mov­ing parts. The an­swer, of course, is to keep gears in a gear­box, and to place all crit­i­cal mov­ing parts in­side a sealed shell.

The moral here is that most me­chan­i­cal nano-ma­chines can be de­signed to avoid many of the prob­lems that they might face rel­a­tive to their mag­i­cal biolog­i­cally-in­spired cous­ins. At the same time, this in­di­cates a more gen­eral prob­lem: to build a func­tional nano-ma­chine, there must be some in­put and some out­put which fa­cil­i­tates ac­tion on the en­vi­ron­ment. This means that some of the mechanisms in a given nano-ma­chine must be de­signed for the en­vi­ron­ment. How­ever, the sever­ity of this limi­ta­tion also de­pends on the nano-ma­chine’s func­tion­al­ity: For power, me­chan­i­cal mechanisms can be de­signed to min­i­mize com­plex parts ex­te­rior to a core gear­box. For ex­pel­ling ob­jects, a nano-ma­chine might in­cor­po­rate a sim­ple air­lock-like de­sign. How­ever, for tak­ing in and pro­cess­ing new molecules in an un­con­trol­led en­vi­ron­ment, nano-ma­chines would re­quire both filters and safe-guards to pre­vent clog­ging.

C. APM as Macro-Scale Manufacturing

While these limi­ta­tions do not pre­clude the ex­is­tence of nano-ma­chines (and es­pe­cially those that man­u­fac­ture on the fly) in un­con­trol­led en­vi­ron­ments, they do raise ques­tions. How­ever, this spe­cific tech­nol­ogy is not the end goal that Drexler fo­cuses on in Rad­i­cal Abun­dance. Though Drexler does al­lude to “fast, thor­ough data col­lec­tion and the means for rapid de­ploy­ment of nano-scale de­vices” in the con­text of biolog­i­cal in­ter­ven­tions, the lion-share of spec­u­la­tion about the fu­ture is fo­cused on ma­chines like this one:

Pic­ture your­self stand­ing out­side the fi­nal as­sem­bly cham­ber of a large-product APM sys­tem and look­ing in through a win­dow to view the ma­chines at work in a space the size of a one-car garage....

To the right, you see an exit door for prod­ucts ready for de­liv­ery. To the left, you see what look like wall-to-wall, floor-to-ceiling shelves, with each shelf par­ti­tioned to make a row of box-shaped cham­bers. In the mid­dle of the garage-sized cham­ber in front of you is a mov­able lift sur­rounded by a set of ma­chines.

The ma­chines look un­com­monly sleek, yet very fa­mil­iar. They re­sem­ble ma­chines in an au­to­mated fac­tory, with robotic arms pro­grammed to swing around, pick up com­po­nents, and swing back to snap the com­po­nents to­gether. The ma­chines look like this be­cause they are, in fact, ma­chines in an au­to­mated fac­tory and be­cause ma­chines that perform similar mo­tions of­ten have similar shapes and similar mov­ing parts. Be­cause they are made of ma­te­ri­als bet­ter than steel, how­ever, they can be faster, lighter, and more effi­cient.

Look­ing back at the wall on the left, you can get a clear view into sev­eral cham­bers that hap­pen to be at eye level and near the win­dow. Each smaller cham­ber con­tains ma­chines with swing­ing arms, and the over­all setup in­side looks like a scale model of the larger cham­ber, com­plete with a rear wall with wall-to-wall, top-to-bot­tom rows of yet smaller cham­bers. It’s hard to see in de­tail what these small cham­bers-within-cham­bers con­tain, but they seem to hold a tiny yet fa­mil­iar set of ma­chines mounted in front of a rear wall with rows of yet smaller cham­bers.

With the press of a but­ton, the ma­chin­ery kicks into gear. At first noth­ing seems to hap­pen, but in less than a minute the large ma­chines in front of you start to pick up parts as they pop out of the cham­bers in the wall at the left, mov­ing these parts to the plat­form in the cen­ter where the first parts are clamped, and the rest snap to­gether. As the ma­chines put the parts to­gether, a fa­mil­iar product takes shape, an au­to­mo­bile, differ­ent in al­most ev­ery de­tail from those built to­day, yet hav­ing a form that re­veals the same func­tion. cham­bers.

With the press of a but­ton, the ma­chin­ery kicks into gear.

At first noth­ing seems to hap­pen, but in less than a minute the large ma­chines in front of you start to pick up parts as they pop out of the cham­bers in the wall at the left, mov­ing these parts to the plat­form in the cen­ter where the first parts are clamped, and the rest snap to­gether. As the ma­chines put the parts to­gether, a fa­mil­iar product takes shape, an au­to­mo­bile, differ­ent in al­most ev­ery de­tail from those built to­day, yet hav­ing a form that re­veals the same func­tion.

Each part takes sev­eral sec­onds to put into place and new parts slide out of the cham­bers at a cor­re­spond­ing rate, each cham­ber de­liv­er­ing a com­po­nent ev­ery few sec­onds. To the left, in­side the clos­est cham­ber, you can see the ma­chines work­ing in­side. Th­ese mi­ni­a­ture ma­chines seem to be perform­ing similar tasks, but at a rate of sev­eral cy­cles per sec­ond, their mo­tions are al­most too quick to fol­low. It’s easy to guess what’s hap­pen­ing in the yet-smaller cham­bers farther back, yet the mo­tions there are no more than a blur.

In the main cham­ber the work is com­plete in less than a minute. The door to the right then un­seals and opens, and a car moves out into a re­ceiv­ing area, sealed in what looks like a plas­tic sleeve. A mo­ment af­ter the door re­seals, the sleeve is pul­led back for re­cy­cling and the pro­cess is done. (This exit ma­neu­ver is part of a cy­cle that pre­vents con­tam­i­nants from en­ter­ing when the product ex­its.

This de­scrip­tion is in line with the video linked ear­lier and leads me to in­fer that, though many other uses also ex­ist for nan­otech­nol­ogy, the main use-case em­pha­sized in Rad­i­cal Abun­dance is ex­tremely effi­cient fac­to­ries that build ex­tremely high qual­ity ma­te­ri­als by ex­ploit­ing “lightweight, car­bon-based ma­te­ri­als” and in­ter­est­ing already-dis­cov­ered pat­terns that may pro­duce ex­otic elec­tronic prop­er­ties on the macro-scale. I men­tion this be­cause APM as ad­vanced man­u­fac­tur­ing and APM nano-ma­chine tech­nol­ogy it­self re­flect differ­ent risk pro­files that ought to be dis­cussed in the im­pact sec­tion. As Drexler says, “where the phys­i­cal na­ture of APM tech­nolo­gies is con­cerned, the rele­vant ques­tions per­tain to the physics and en­g­ineer­ing of com­pact, highly ca­pa­ble fac­to­ries—not vague dreams, ex­otic prod­ucts or nanobugs.”

III. APM Will Change Everything

A. Drexler’s Par­tial Notes on APM Benefits

Be­cause APM will use cheap, read­ily available ma­te­ri­als like car­bon to pro­duce high qual­ity ma­te­ri­als in min­utes from a com­pact ma­chine, it has the in­gre­di­ents to rev­olu­tionize the ex­ist­ing means of pro­duc­tion:

“Nanoscale size en­ables ex­treme pro­duc­tivity as a con­se­quence of me­chan­i­cal scal­ing laws. In ad­di­tion, small-scale, ver­sa­tile, highly pro­duc­tive ma­chin­ery can col­lapse globe-span­ning in­dus­trial sup­ply chains to just a few links.”

This mas­sive sim­plifi­ca­tion of sup­ply chains cou­pled with APM’s use of abun­dant molecules im­plies a mas­sive eco­nomic shift. Lo­ca­tions im­pov­er­ished by lack of cap­i­tal, busi­ness con­nec­tions and re­sources could, in prin­ci­ple, use nan­otech­nol­ogy to sus­tain­ably and lo­cally pro­duce high qual­ity in­fras­truc­ture and re­li­able food sources.

Other benefits of APM in­clude trans­form­ing in­for­ma­tion tech­nolo­gies through more ad­vanced ma­te­ri­als, im­prov­ing in­fras­truc­ture through bet­ter con­struc­tion and trans­porta­tion tech­nol­ogy, im­prov­ing agri­cul­ture through effi­cient re­cy­cling and wa­ter-pro­cess­ing/​de­sal­i­na­tion, and re­solv­ing global warm­ing car­bon diox­ide cap­ture. The run­ning theme themes in each of these solu­tions are APM’s blend of effi­ciency and abil­ity to cheaply con­struct many of the ad­vanced nano­ma­te­rial solu­tions which cur­rently ex­ist but are too costly to im­ple­ment in­dus­tri­ally.

The benefits of APM in im­prov­ing agri­cul­ture also struck me as par­tic­u­larly sig­nifi­cant. In par­tic­u­lar, Drexler notes that en­closed agri­cul­ture (i.e. large-scale green­houses) offers “higher yield per hectare, bet­ter food qual­ity, free­dom from pes­ti­cides, ex­tended grow­ing sea­sons, free­dom from con­straints of soil qual­ity and available wa­ter, and pro­tec­tion from drought” while also re­duc­ing “wa­ter de­mand and con­tam­i­na­tion.” Th­ese lat­ter benefits—free­dom from soil, wa­ter and weather con­straints—are already achieved achieved in mod­ern gree­houses and act as the cor­ner stone for alle­vi­at­ing star­va­tion in im­pov­er­ished re­gions. Nev­er­the­less, the mod­ern world, let alone im­pov­er­ished re­gions, can­not meet the effi­ciency and in­fras­truc­tural re­quire­ments that would make im­ple­ment­ing green­houses eco­nom­i­cal. APM’s ca­pac­ity for rapid and cheap in­fras­truc­ture along with highly effi­cient en­ergy sys­tems (i.e. so­lar power) will change that.

Over­all, I see only two rea­sons to doubt these benefits. The first rea­son is eco­nomic. His­tor­i­cally, in­ter­na­tional trade has been a ne­ces­sity for tech­nolog­i­cally ad­vanced coun­tries which has al­lowed the flow of money to less de­vel­oped coun­tries. How­ever, APM will fi­nally al­low tech­nolog­i­cally ad­vanced coun­tries to be­come self-suffi­cient to the ex­tent that they do not need to im­port goods from less ad­vanced coun­tries. If these coun­tries can­not amass the wealth to gain nan­otech­nol­ogy, the ab­sence of eco­nomic ties with tech­nolog­i­cally ad­vanced au­tark­ies may lock them into poverty. How­ever, the pres­ence of these self-suffi­cient coun­tries also makes this situ­a­tion un­likely. Be­cause APM-driven coun­tries will be so wealthy and likely re­tain some al­tru­is­tic lean­ing, the only rea­son for them not to provide APM to less ad­vanced coun­tries would be some sort of non-eco­nomic cost or risk—that is, a mil­i­tary risk. For­tu­nately, be­cause APM fac­to­ries will likely be both tech­nolog­i­cally com­plex and op­ti­mized for use in fabri­cat­ing spe­cific ma­te­ri­als rather than ev­ery ma­te­rial, the mil­i­tary risk of pro­vid­ing fac­to­ries de­signed with food and in­fras­truc­ture in mind is rel­a­tively low. This is es­pe­cially true when not­ing how mil­i­tary APM will em­power the most ad­vanced coun­tries rel­a­tive to oth­ers. To be fair though—in the era of semi-costly/​only par­tially effec­tive nan­otech­nol­ogy—eco­nomic con­cerns could be very se­ri­ous even if they are only short-term con­sid­er­a­tions.

The sec­ond rea­son for doubt is more sig­nifi­cant and re­lates to an im­por­tant crit­i­cism of Rad­i­cal Abun­dance: Drexler effec­tively dis­cusses the ca­pa­bil­ities of nan­otech­nol­ogy but does not se­ri­ously dis­cuss the de­tails of the prob­lems he claims that it will solve. In prac­ti­cal terms, the most im­por­tant al­tru­is­tic con­tri­bu­tion of nan­otech­nol­ogy would be poverty re­duc­tion. Ex­plic­itly, nan­otech­nol­ogy should provide sus­tain­able food sources in places where food can­not cur­rently grow, wa­ter in places where wa­ter is cur­rently in­ac­cessible, in­fras­truc­ture in places where in­fras­truc­ture can­not cur­rently be es­tab­lished, and en­ergy in places with limited en­ergy re­sources. Th­ese prob­lems are al­most all ge­o­graph­i­cal con­straints and, while Drexler claims that “the most use­ful el­e­ments—in­clud­ing car­bon ni­tro­gen, oxy­gen, and sili­con—are not all that scarce,” ni­tro­gen defi­ciency is one of the rea­sons for agri­cul­tural difficul­ties in Africa. More­over, many im­pov­er­ished coun­tries suffer from lack of wa­ter/​re­li­ance on con­tam­i­nated ground wa­ter and, while nan­otech­nol­ogy may provide cheap purifi­ca­tion meth­ods, I am not con­vinced that it will ei­ther offer purifi­ca­tion meth­ods or wa­ter trans­porta­tion meth­ods bet­ter than the on­go­ing sci­ence in those fields. Fi­nally, many im­pov­er­ished coun­tries suffer from rocky, moun­tain­ous and swampy ter­rain which in­hibits move­ment and the fea­si­bil­ity of tak­ing ad­van­tage of large-scale in­fras­truc­ture. En­closed agri­cul­ture on forty-five de­gree rocky in­clines does not seem very promis­ing, es­pe­cially when com­pared to just us­ing longer sup­ply chains which APM would hope­fully avoid. In short, while nan­otech­nol­ogy might al­low the col­laps­ing of globe-span­ning sup­ply chains to a few effi­cient links, Drexler does not dis­cuss how the lo­ca­tions that are most harmed by the need for these sup­ply struc­tures would sur­mount their ex­ist­ing ge­o­graph­i­cal con­straints us­ing nan­otech­nol­ogy. If this is not ad­dressed, it cre­ates a mas­sive prob­lem for APM. After all, if ge­o­graph­i­cal is­sues limit the effi­cacy of al­tru­ism through nan­otech­nol­ogy, then—for im­pov­er­ished coun­tries—ac­cess to nan­otech­nol­ogy may not out­weigh the eco­nomic loss of be­ing un­able to trade with wealthy coun­tries.

Beyond the po­ten­tial for pro­vid­ing the ba­sic sta­ples of hu­man life, an ad­ja­cent sig­nifi­cant con­tri­bu­tion that highly ad­vanced nan­otech­nol­ogy brings is con­trol of the ecosys­tems that make these sta­ples. As some­one con­cerned about wild an­i­mal suffer­ing, this de­vel­op­ment could mas­sively re­duce hu­man harm to in­sects and po­ten­tially offer strate­gies for sys­temic wildlife in­ter­ven­tions. While re­straint and hes­i­ta­tion should gen­er­ally be ap­plied to ac­tions like this as they cur­rently have dra­matic and un­pre­dictable im­pacts on wild ecosys­tems, the ex­ten­sive surveillance tech­nolo­gies pro­vided by APM (dis­cussed later as a draw­back) along with ad­vanc­ing sci­ence may provide the ex­act kind of ecolog­i­cal knowl­edge needed to make these ad­just­ments wisely.

While I have pre­vi­ously sus­pected that limi­ta­tions on chem­i­cals and man­u­fac­tur­ing meth­ods may ren­der APM less use­ful than it seems, I think that these views fail to ap­pre­ci­ate sim­ply how much can be ac­com­plished through the cheap pro­duc­tion of high qual­ity in­fras­truc­ture and elec­tron­ics alone. With this in mind, I think that the only sce­nar­ios where APM is less-than-rev­olu­tion­ary are sce­nar­ios where some other eas­ier-to-de­velop tech­nol­ogy fo­cused on par­tic­u­larly im­por­tant use-cases suc­ceeds faster by virtue of a more di­rect path to im­ple­men­ta­tion. Off the top of my head, these might look like some of the fol­low­ing:

  • Plant growth in bar­ren en­vi­ron­ments through ge­netic en­g­ineer­ing cou­ples with mass au­toma­tion of farm­ing to pro­duce cheap, abun­dant food that is not limited by dis­tri­bu­tion costs.

  • Some large-scale tech­nique for pro­cess­ing abun­dant el­e­ments effi­ciently into use­ful ma­te­ri­als is es­tab­lished (though I am not aware of meth­ods for this out­side of nan­otech­nol­ogy)

  • Mass trial and er­ror fa­cil­i­tated by microfluidics en­ables suffi­ciently fast and per­son­al­ized med­i­cal treat­ment that APM may only be able to en­hance rather than revolutionize

  • The grad­ual arc of sci­ence im­prov­ing effi­ciency across the board in trans­port lo­gis­tics cou­pled with ex­ist­ing in­for­ma­tion tech­nol­ogy drive a shift to a digi­tal, de-lo­cal­ized econ­omy be­fore APM makes uni­ver­sal wealth fea­si­ble.

This is my great­est and most ex­plicit un­cer­tainty with re­spect to APM’s benefits. I agree that APM would provide these benefits but I am not sure whether APM will be the first to provide them. Heuris­ti­cally, I usu­ally ex­pect that tech­nol­ogy speci­fi­cally de­signed to ad­dress a spe­cific is­sue will ad­vance much faster than a more generic ab­stract ap­proach. This leads me to doubt the com­par­a­tive benefits of pur­su­ing APM as op­posed to other more ex­plic­itly ori­ented re­search en­deav­ors. On the other hand, APM is a rel­a­tively con­crete idea and, even if it is likely more difficult to pur­sue than a more fo­cused tech­nol­ogy, may be worth it due to its large po­ten­tial range of im­pact.

B. Drexler’s Notes On How APM May Go Wrong

One of the dan­ger­ous as­pects of APM that Drexler notes is surveillance.

Small-scale com­po­nents and sys­tems open new pos­si­bil­ities. For per­spec­tive, con­sider that a one-gram plat­form built with ad­vanced tech­nolo­gies could pro­duce ter­aflops of com­pu­ta­tional power (and much more, in bursts), to­gether with a mil­lion-ter­abyte data stor­age ca­pac­ity and bet­ter-than-hu­man sen­sors, all with a power de­mand com­pa­rable to that of a cell phone on standby. Now con­sider what could be done if de­vices of this class cost about $1.00 per kilo­gram and could be de­liv­ered by small drone air­craft. $100 billion would buy roughly one de­vice of this sort per square me­ter of land area, wor­ld­wide.

Be­cause most of this tech­nol­ogy will not need ter­aFLOPs of com­put­ing power, this raises the pos­si­bil­ity of globally abun­dant third-party surveillance as easy to bring to work as it is to track dirt into a house. Cou­pled with the abil­ity to re­lay in­for­ma­tion, these sens­ing de­vices would be able to triv­ially trans­port mas­sive amounts of data to a cen­tral­ized sys­tem for fast (>petaFLOP) pro­cess­ing. This could quickly lead to a race-to-the-bot­tom fa­vor­ing those who are will­ing to care less. Com­bin­ing these same de­vices with chem­i­cal pay­loads may also en­able groups to en­gage in the soft psy­cholog­i­cal ma­nipu­la­tion or un­de­tectable as­sas­si­na­tion of their en­e­mies. Com­bin­ing these “se­cre­tive surveillance regimes” with a world that re­places wealth with in­di­vi­d­ual virtue as a sta­tus sym­bol in ab­sence of poverty may also lead to the micro-reg­u­la­tion of hu­man be­hav­ior. Ad­di­tion­ally, be­cause surveillance does con­fer ad­van­tages onto the groups that use it and many coun­tries vary sig­nifi­cantly in their at­ti­tudes to­ward in­di­vi­d­ual pri­vacy, es­tab­lish­ing a com­mon pro-pri­vacy agree­ment to avoid a race-to-the-bot­tom will be un­likely.

Over­all, I am not con­cerned about this risk mainly be­cause this surveillance tech­nol­ogy is ori­ented around ac­tive nano-ma­chines in com­plex and un­pre­dictable en­vi­ron­men­tal situ­a­tions. Un­like the benefits of nan­otech­nol­ogy which are pro­duced in con­trol­led man­u­fac­tur­ing con­texts, this tech­nol­ogy must func­tion in a less man­age­able space and I ex­pect phys­i­cal and de­sign limi­ta­tions to se­ri­ously come into play here. Un­doubt­edly, nan­otech­nol­ogy will cheapen surveillance but I doubt that it will be suffi­ciently cheap to sub­vert per­sonal pri­vacy more than it would be any­way. Fur­ther­more, while a race-to-the-bot­tom may hap­pen, it may be par­tially miti­gated by the rad­i­cally differ­ent eco­nomic cir­cum­stances which more sim­ple nan­otech­nol­ogy will have already pro­vided the world, mak­ing it “hard to find an ex­ter­nal re­source that would con­tinue to be a vi­tal na­tional in­ter­est.”

Th­ese are also the same rea­sons that I am not se­ri­ously con­cerned with nan­otech­nol­ogy caus­ing mas­sive en­vi­ron­men­tal dam­age through self-repli­cat­ing robots: the prob­lem is uniquely challeng­ing and strong in­cen­tives for do­ing so ap­pear to be ab­sent. In fact, Drexler notes that na­tions have “strong, shared in­ter­ests in con­strain­ing non-state ac­tors” from the use of un­re­strained APM. A weaker var­i­ant of this is­sue is the po­ten­tial for nan­otech­nol­ogy to cre­ate nanopar­ti­cle pol­lu­tants on a mas­sive scale, lead­ing to mas­sive ecosys­temic dam­age. How­ever, coun­tries ca­pa­ble of pro­duc­ing nan­otech­nol­ogy at this level will also likely have tech­nol­ogy ca­pa­ble of rapidly iden­ti­fy­ing (via surveillance) and ad­dress­ing pol­lu­tants any­where on the globe. This means that mass pol­lu­tion is only a risk if done by a coun­try that would not fear re­tal­i­a­tion. Be­cause this type of coun­try can already threaten the world in many more ways than just pol­lu­tion, I con­sider this to be a minor con­cern.

The sec­ond main risk that Drexler dis­cusses is a two-pronged change in mil­i­tary strat­egy. First, be­cause APM will be used to de­sign se­quen­tial gen­er­a­tions of APM, “it is there­fore easy to en­vi­sion sce­nar­ios in which a mod­est de­gree of asyn­chrony, mea­sured in months or less, could swiftly lead to rad­i­cal mil­i­tary asym­me­tries even with rel­a­tively shal­low ex­ploita­tion of the over­all po­ten­tial of APM-level tech­nolo­gies.” Se­cond, APM would provide “abun­dant, af­ford­able, non-lethal, re­motely op­er­ated weapons” that re­duce risk of harm and en­courage more war-like poli­cies. While these shifts may en­courage tech­nolog­i­cally ad­vanced coun­tries to pur­sue ag­gres­sive ex­pan­sion­ist poli­cies, ex­pan­sion­ist ac­tions will offer lit­tle in terms of re­source gain and cost a level of so­cietal sta­bil­ity due to cul­tural con­flicts. In fact, be­cause of the ab­sence of re­source gain, ex­pan­sion­ism will pri­mar­ily be driven by a de­sire to es­tab­lish differ­ent so­cietal val­ues—a goal that gov­ern­ments will more likely pur­sue through sub­tler meth­ods than out­right con­flict.

Lastly, Drexler also notes that one of the com­monly dis­cussed risks—the idea that self-re­pro­duc­ing nano-ma­chines will con­sume vast swathes of the world—is un­likely given a man­u­fac­tur­ing model of nan­otech­nol­ogy in­stead of a biolog­i­cal model. I agree that this is un­likely for the same rea­son that I saw surveillance as un­likely but mul­ti­plied sev­eral-fold. Build­ing en­vi­ron­men­tally ro­bust re­pro­duc­ing nano-ma­chines us­ing abun­dant nat­u­ral sup­plies and es­tab­lish­ing enough sta­ble com­plex­ity for those nano-ma­chines to re­con­struct them­selves both re­quire an enor­mous amount of effort far be­yond the scope of the pri­mary benefits of nan­otech­nol­ogy which Drexler dis­cusses.

One dan­ger that I be­lieve mer­its con­cern but goes un­men­tioned by Drexler per­tains to ad­vanced gen­eral in­tel­li­gence. If Drexler is cor­rect in his claim that nan­otech­nol­ogy will offer ter­aFLOP/​gram com­put­ers, then an av­er­age lap­top com­puter weigh­ing five pounds would have about a petaFLOP of com­pu­ta­tional power and a thou­sand of such com­put­ers would start push­ing on the ca­pa­bil­ities of a hu­man brain (ac­cord­ing to Kurzweil) with­out any ad­di­tional de­liber­ate op­ti­miza­tion be­yond em­u­lat­ing na­ture. This is more than suffi­cient to pro­duce in­tel­li­gence far more com­pe­tent than any hu­man and in­ter­twines nan­otech­nol­ogy deeply with its re­la­tion­ship to ar­tifi­cial in­tel­li­gence. If hu­man­ity is not con­fi­dent in its abil­ity to al­ign the goals of ad­vanced gen­eral in­tel­li­gence (which it is not), then nan­otech­nol­ogy in this re­gard ought only be pur­sued if does not in­crease the risk of mis­al­igned ad­vanced gen­eral in­tel­li­gence. This will be the case if ei­ther ad­vanced gen­eral in­tel­li­gence is ex­pected to emerge prior to ad­vanced nan­otech­nol­ogy (in which case AI Safety re­search is a much more press­ing con­cern than nan­otech­nol­ogy) or Drexler is in­cor­rect about ter­aFLOP/​gram com­put­ers.

Over­all, my opinion is that Drexler’s benefits of nan­otech­nol­ogy out­weigh Drexler’s risks but, when I con­sider nan­otech­nol­ogy’s re­la­tion­ship with ar­tifi­cial in­tel­li­gence, I be­gin to sus­pect that it is far more dan­ger­ous than it ini­tially ap­pears. That be­ing said, this only ap­plies di­rectly to the de­vel­op­ment of nan­otech­nol­ogy it­self. Devel­op­ment of nan­otech­nol­ogy policy on the other hand may have sig­nifi­cant value in the case where hu­man­ity weath­ers ad­vanced gen­eral in­tel­li­gence. How­ever, the benefits of this work may be limited by the like­li­hood that ad­vanced gen­eral in­tel­li­gence will trans­form the poli­ti­cal land­scape to such an ex­tent that any cur­rent policy de­sign would be in­ap­pli­ca­ble.

IV. Path­ways to Nanotechnology

When won­der­ing about the like­li­hood of achiev­ing some­thing like APM, a use­ful ques­tion is “Why has this not been achieved already?” This is an es­pe­cially salient ques­tion for APM both be­cause of the ex­ten­sive fund­ing pro­vid­ing in the field of nan­otech­nol­ogy and the rel­a­tive age of the idea it­self. Drexler an­swers this ques­tion with two main ex­pla­na­tions. The first is that ex­ten­sive fund­ing may have been fo­cused on “nan­otech­nol­ogy” but min­i­mal fund­ing has been fo­cused on APM.

“The great promise of nan­otech­nol­ogy is atom­i­cally pre­cise man­u­fac­tur­ing, and the US Congress es­tab­lished a pro­gram di­rected to­ward this ob­jec­tive, but the pro­gram in­stead did some­thing en­tirely differ­ent. The pro­gram’s lead­ers re­defined “nan­otech­nol­ogy and sup­ported only nanoscale ma­te­ri­als and de­vices, tech­nolo­gies as differ­ent from APM as cloth, ce­ment, and wires are from a pro­gram­me­able digi­tal com­puter. Most re­search ad­ver­tised as “nan­otech­nol­ogy” has there­fore been ir­rele­vant to what had been widely ex­pected, and while atom­i­cally pre­cise fabri­ca­tion has flour­ished in the molec­u­lar sci­ences, peo­ple look­ing for progress to­ward APM-level tech­nolo­gies have been led to look in the wrong di­rec­tion.”

The sec­ond is a rather philo­soph­i­cal dis­cus­sion about how the ba­sic sci­ence that mostly com­prises nan­otech­nol­ogy re­search is in­effec­tive at pro­duc­ing the sys­tem-en­g­ineer­ing co­or­di­na­tion re­quired to make an ac­tual tech­nol­ogy. As Drexler says, “no mat­ter how re­search-in­ten­sive a pro­ject may be, work co­or­di­nated around con­crete en­g­ineer­ing ob­jec­tives will even­tu­ally be re­quired to pro­duce con­crete en­g­ineer­ing re­sults.” So far, work at this level of co­or­di­na­tion has not been pur­sued.

Both of these ex­pla­na­tions demon­strate that the ex­ist­ing limited state of APM is due to poli­ti­cal and or­ga­ni­za­tional con­straints that do not re­flect nega­tively on the fea­si­bil­ity of APM it­self. As for that fea­si­bil­ity, Drexler notes that de­sign­ing the mechanisms of APM de­vices is already rel­a­tively fea­si­ble: “Ma­chine com­po­nents based on rigid co­va­lent struc­tures can­not yet be im­ple­mented, yet are already easy to de­sign and model us­ing stan­dard com­pu­ta­tional chem­istry soft­ware.” In­deed, I sus­pect that one way to mo­ti­vate APM-di­rected re­search would be to use the tools we have to pre­emp­tively de­sign an ad­vanced APM ma­chine and then to use that de­vice as a mo­ti­va­tor for na­tions to fun­nel re­search into its ul­ti­mate con­struc­tion.

The main cur­rent method for de­vel­op­ing APM will in­volve ad­vanc­ing stereo­tac­tic chem­i­cal re­ac­tions. While “con­ven­tional, solu­tion-phase chem­i­cal re­ac­tions are en­abled by lo­cal struc­tural fea­tures of molecules,” more com­plex syn­thetic tar­gets with many more struc­tural fea­tures would be “difficult or im­pos­si­ble to di­rect re­ac­tions with suffi­cient speci­fic­ity.” Stereo­tac­tic chem­i­cal re­ac­tions ad­dress this is­sue by hav­ing “link­ing struc­tures di­rect re­ac­tions by con­strain­ing en­coun­ters among po­ten­tially re­ac­tive groups, sep­a­rat­ing some pairs while in­creas­ing en­counter rates be­tween oth­ers.” By defi­ni­tion, stereo­tac­tic chem­i­cal re­ac­tions re­flect the atom­i­cally pre­cise man­u­fac­tur­ing of com­plex struc­tures through se­quen­tially bond­ing molecules to­gether. This de­scrip­tion ap­plies to biolog­i­cal pro­teins as well but un­like pro­teins, APM will de­velop stereo­tac­tic re­ac­tions through me­chan­i­cal de­signs rather than ther­mally driven ones.

Drexler notes that stereo­tac­tic re­ac­tions in ad­vanced APM must satisfy a num­ber of con­straints: struc­tures must be sta­ble and rigid; struc­ture mo­tion must be well con­trol­led; re­ac­tions must hap­pen re­li­ably; re­ac­tions must be ir­re­versible; and re­ac­tions must yield only a sin­gle product. Many of these re­quire­ments (mo­tion con­trol, re­li­able pro­cesses, ir­re­versibil­ity, sin­gle product) are con­sis­tent with most fabri­ca­tion pro­ce­dures. How­ever, rigidity holds spe­cial im­por­tance for APM be­cause it al­lows fabri­cated ma­te­ri­als to oc­cupy meta-sta­ble states. This means that, while flex­ible biolog­i­cal ma­te­ri­als can quickly re­struc­ture them­selves into lower en­ergy/​higher en­tropy states by just mov­ing, APM ma­te­ri­als move in a slow man­ner due to rigidity that pre­vents them from find­ing those states as quickly. This en­sures that while APM ma­te­ri­als, like di­a­mond, main­tain their prop­er­ties for a long time even though they might not be nat­u­rally sta­ble.

Cur­rent tech­nol­ogy can­not meet all these re­quire­ments how­ever the as­sem­bly of “polymeric build­ing blocks, cross-linked via con­ven­tional re­ac­tions, with all stereo­tac­tic op­er­a­tions performed in aque­ous en­vi­ron­ments” may read­ily be achieved through ex­ist­ing meth­ods like DNA origami. This offers a bot­tom-up APM start­ing point wherein “large self-al­ign­ing build­ing blocks, loose po­si­tional tol­er­ance mar­gins, low stiff­ness ma­te­ri­als, sim­ple ma­chines, and sim­ple mo­tion con­straints” are iter­ated upon to pro­duce the more com­plex, more pre­cise, and higher qual­ity nano-de­vices needed to reach ad­vanced APM. In other words, sim­ply iter­at­ing on ex­ist­ing nan­otech­nol­ogy re­lated to stereo­tac­tic con­trol will fa­cil­i­tate the im­prove­ment of stereo­tac­tic con­trol.

Drexler also ad­dresses an­other as­pect of nan­otech­nol­ogy that had pre­vi­ously led me to be more sus­pi­cious about its suc­cess: On the macro-scale level, a fac­tory may rely on ma­chines pro­duced by other fac­to­ries but those fac­to­ries them­selves are as­sem­bled by hu­mans (though of­ten hu­mans pi­lot­ing ma­chines them­selves). On the nano-scale though, hu­man as­sem­bly is im­pos­si­ble so fac­to­ries would need to pro­duce fac­to­ries—which seems hard. In ac­tu­al­ity, the solu­tion is sim­ple and prac­ti­cal: Stereo­tac­tic con­trol will lead to “im­prove­ments that fa­cil­i­tate the de­sign and fabri­ca­tion of com­ple­men­tary sur­faces… Stereo­tac­tic syn­the­sis can en­able ad­vances in com­po­nent level self-as­sem­bly.” Once hu­man­ity can pro­duce nano-ma­chines, hu­man­ity will be able to pro­duce nano-ma­chines at­tached to sur­faces with bind­ing af­fini­ties for com­ple­men­tary sur­faces. At this point, near cur­rent-day tech­nol­ogy would be able to pro­duce large-scale com­ple­men­tary sur­face pat­terned sheets that bind only the right nano-ma­chines in only the right places. In other words, suc­cess will not ini­tially be achieved through a closed sys­tem of nano-fac­to­ries mak­ing nano-fac­to­ries but rather through a blend of man­u­fac­tur­ing nano-ma­chines to more eas­ily in­ter­act with higher level self-as­sem­bly meth­ods.

V. Conclusion

Through­out Rad­i­cal Abun­dance, Drexler em­pha­sizes two com­mon mis­rep­re­sen­ta­tions of nan­otech­nol­ogy: the pop­u­lar rep­re­sen­ta­tion of nan­otech­nol­ogy as al­most mag­i­cal nanobot swarms and the aca­demic rep­re­sen­ta­tion of nan­otech­nol­ogy as per­tain­ing to sci­ence at a given size-scale rather than tech­nol­ogy at a given size-scale. As some­one both aware of pop­u­lar cul­ture and closely in­volved with con­ven­tional aca­demic re­search in nan­otech­nol­ogy (i.e. self-as­sem­bling nano-par­ti­cles), I blended these ideas to­gether into an im­prob­a­bly in­fluen­tial tech­nol­ogy me­di­ated by an im­prob­a­bly hard-to-con­trol sci­ence. In re­al­ity, APM is a com­pre­hen­si­bly im­pact­ful tech­nol­ogy me­di­ated by un­der-re­searched but high-po­ten­tial iter­a­tive de­sign. Finish­ing the book, I feel rea­son­ably con­vinced that APM is achiev­able based on ex­ist­ing tech­nolog­i­cal progress, par­tially con­vinced that APM will have the benefits that Drexler claims it has, and un­con­vinced that the benefits out­weigh the risks if we limit them to those that Drexler has de­scribed.

Nev­er­the­less, while un­der­stand­able in the con­text of nan­otech­nol­ogy’s pre­vi­ous over-hyped his­tory in pop-cul­ture, I think Rad­i­cal Abun­dance’s dis­cus­sion of risks is too con­ser­va­tive in es­ti­mat­ing their mag­ni­tude. This is be­cause, in a sig­nifi­cant num­ber of cir­cum­stances where hu­man-de­vel­oped nan­otech­nol­ogy be­comes im­por­tant, I en­vi­sion sce­nar­ios that ac­cel­er­ate the emer­gence of a mis­al­igned gen­eral ar­tifi­cial in­tel­li­gence. Th­ese sce­nar­ios may be ab­solutely low prob­a­bil­ity but should be ac­counted for in nan­otech­nol­ogy policy given the ex­is­ten­tial na­ture of the threat. Beyond this, I think that Drexler’s goal of ad­dress­ing mis­con­cep­tions about out-of-con­trol nano-ma­chines may also fail to rec­og­nize risks from out-of-con­trol nano-ma­chines which are out­side the purview of com­par­a­tively limited APM. I have no rea­son­ing for this be­yond the sense that any the­o­ret­i­cally pos­si­ble ex­is­ten­tial risk ought to merit closer con­sid­er­a­tion and that many peo­ple have given closer con­sid­er­a­tion to it.

So in sum­mary, APM is a lot more likely than I ini­tially ex­pected and I also sus­pect it could be a lot more dan­ger­ous as well. Let me know your thoughts!