Reducing long-term risks from malevolent actors

Summary

  • Dic­ta­tors who ex­hibited highly nar­cis­sis­tic, psy­cho­pathic, or sadis­tic traits were in­volved in some of the great­est catas­tro­phes in hu­man his­tory. (More)

  • Malev­olent in­di­vi­d­u­als in po­si­tions of power could nega­tively af­fect hu­man­ity’s long-term tra­jec­tory by, for ex­am­ple, ex­ac­er­bat­ing in­ter­na­tional con­flict or other broad risk fac­tors. (More)

  • Malev­olent hu­mans with ac­cess to ad­vanced tech­nol­ogy—such as whole brain em­u­la­tion or other forms of trans­for­ma­tive AI—could cause se­ri­ous ex­is­ten­tial risks and suffer­ing risks. (More)

  • We there­fore con­sider in­ter­ven­tions to re­duce the ex­pected in­fluence of malev­olent hu­mans on the long-term fu­ture.

    • The de­vel­op­ment of ma­nipu­la­tion-proof mea­sures of malev­olence seems valuable, since they could be used to screen for malev­olent hu­mans in high-im­pact set­tings, such as heads of gov­ern­ment or CEOs. (More)

    • We also ex­plore pos­si­ble fu­ture tech­nolo­gies that may offer un­prece­dented lev­er­age to miti­gate against malev­olent traits. (More)

    • Select­ing against psy­cho­pathic and sadis­tic ten­den­cies in ge­net­i­cally en­hanced, highly in­tel­li­gent hu­mans might be par­tic­u­larly im­por­tant. How­ever, risks of un­in­tended nega­tive con­se­quences must be han­dled with ex­treme cau­tion. (More)

  • We ar­gue that fur­ther work on re­duc­ing malev­olence would be valuable from many moral per­spec­tives and con­sti­tutes a promis­ing fo­cus area for longter­mist EAs. (More)

What do we mean by malev­olence?

Be­fore we make any claims about the causal effects of malev­olence, we first need to ex­plain what we mean by the term. To this end, con­sider some of the ar­guably most evil hu­mans in his­tory—Hitler, Mao, and Stalin—and the dis­tinct per­son­al­ity traits they seem to have shared.[1]

Stalin re­peat­edly turned against former com­rades and friends (Her­sh­man & Lieb, 1994, ch. 15, ch. 18), gave de­tailed in­struc­tions on how to tor­ture his vic­tims, or­dered their loved ones to watch (Glad, 2002, p. 13), and de­liber­ately kil­led mil­lions through var­i­ous atroc­i­ties. Like­wise, mil­lions of peo­ple were tor­tured and mur­dered un­der Mao’s rule, of­ten ac­cord­ing to his de­tailed in­struc­tions (Diköt­ter, 2011; 2016; Chang & Hal­li­day, ch. 8, ch. 23, 2007). He also took plea­sure in watch­ing acts of tor­ture and imi­tat­ing in what his vic­tims went through (Chang & Hal­li­day, ch. 48, 2007). Hitler was not only re­spon­si­ble for the death of mil­lions, he also en­gaged in per­sonal sadism. On his spe­cific in­struc­tions, the plot­ters of the 1944 as­sas­si­na­tion at­tempt were hung by pi­ano wires and their ag­o­niz­ing deaths were filmed (Glad, 2002). Ac­cord­ing to Albert Speer, “Hitler loved the film and had it shown over and over again” (Toland, 1976, p. 818). Hitler, Mao, and Stalin—and most other dic­ta­tors—also poured enor­mous re­sources into the cre­ation of per­son­al­ity cults, man­i­fest­ing their colos­sal nar­cis­sism (Diköt­ter, 2019). (The sec­tion Malev­olent traits of Hitler, Mao, Stalin, and other dic­ta­tors in Ap­pendix B pro­vides more ev­i­dence.)

Many sci­en­tific con­structs of hu­man malev­olence could be used to sum­ma­rize the rele­vant psy­cholog­i­cal traits shared by Hitler, Mao, Stalin, and other malev­olent in­di­vi­d­u­als in po­si­tions of power. We fo­cus on the Dark Te­trad traits (Paulhus, 2014) be­cause they seem es­pe­cially rele­vant and have been stud­ied ex­ten­sively by psy­chol­o­gists. The Dark Te­trad com­prises the fol­low­ing four traits—the more well-known Dark Triad (Paulhus & Willi­ams, 2002) refers to the first three traits:

  • Machi­avel­li­anism is char­ac­ter­ized by ma­nipu­lat­ing and de­ceiv­ing oth­ers to fur­ther one’s own in­ter­ests, in­differ­ence to moral­ity, and ob­ses­sion with achiev­ing power or wealth.

  • Nar­cis­sism in­volves an in­flated sense of one’s im­por­tance and abil­ities, an ex­ces­sive need for ad­mira­tion, a lack of em­pa­thy, and ob­ses­sion with achiev­ing fame or power.

  • Psy­chopa­thy is char­ac­ter­ized by bold­ness, cal­lous­ness, im­pul­sive­ness, a lack of em­pa­thy and guilt, and an­ti­so­cial be­hav­ior, in­clud­ing vi­o­lence and crime.

  • Sadism in­volves de­riv­ing plea­sure from in­flict­ing suffer­ing and pain on oth­ers.

There is con­sid­er­able over­lap be­tween the Dark Te­trad traits. In gen­eral, al­most all plau­si­ble op­er­a­tional­iza­tions of malev­olence tend to pos­i­tively cor­re­late with each other and nega­tively with “benev­olent” traits such as al­tru­ism, hu­mil­ity or hon­esty. (See the sec­tion Cor­re­la­tions be­tween dark traits and other traits in Ap­pendix B for more de­tails.)

This sug­gests the ex­is­tence of a gen­eral fac­tor of hu­man malev­olence[2]: the Dark Fac­tor of Per­son­al­ity (Mosha­gen et al., 2018)—analo­gous to g, the gen­eral fac­tor of in­tel­li­gence—char­ac­ter­ized by ego­ism, lack of em­pa­thy[3] and guilt, Machi­avel­li­anism, moral dis­en­gage­ment, nar­cis­sism, psy­chopa­thy, sadism, and spite­ful­ness. Like most per­son­al­ity traits (John­son et al., 2008), malev­olent traits seem rel­a­tively sta­ble over the lifes­pan (Obradović et al., 2007) and in­fluenced by ge­netic fac­tors (Ver­non et al., 2008), but more on this be­low.[4]

Through­out this ar­ti­cle, we will as­sume a di­men­sional—rather than cat­e­gor­i­cal, “black-or-white”—con­cep­tion of malev­olence. That is, we be­lieve that malev­olent traits ex­ist on a con­tinuum—just like most other hu­man traits such as ex­traver­sion or in­tel­li­gence (cf. Haslam et al., 2012; Plomin, 2019, ch. 5). Slight Machi­avel­lian or sadis­tic ten­den­cies, for ex­am­ple, are com­mon. Many hu­mans seem to flat­ter their su­pe­ri­ors and en­joy see­ing (non-tragic) mishaps of their poli­ti­cal op­po­nents. But only a few in­di­vi­d­u­als will de­rive plea­sure from wit­ness­ing hu­man tor­ture or will kill their former friends just to con­soli­date their power.

It is this lat­ter type of hu­man—show­ing clear signs of at least some highly ele­vated Dark Te­trad traits—who we have in mind when we use the term “malev­olent”.

Malev­olent hu­mans in power pose se­ri­ous long-term risks

In this sec­tion we dis­cuss why and how malev­olent in­di­vi­d­u­als in highly in­fluen­tial po­si­tions—such as poli­ti­cal lead­ers or CEOs of no­table com­pa­nies—could nega­tively af­fect hu­man­ity’s long-term tra­jec­tory, ul­ti­mately in­creas­ing ex­is­ten­tial risks (in­clud­ing ex­tinc­tion risks) and risks of as­tro­nom­i­cal suffer­ing (s-risks).

Malev­olent hu­mans of­ten rise to power

Malev­olent hu­mans are un­likely to sub­stan­tially af­fect the long-term fu­ture if they can­not rise to power. But alas, they of­ten do. The most salient ex­am­ples are dic­ta­tors who clearly ex­hibited ele­vated malev­olent traits: not only Hitler, Mao, and Stalin, but also Sad­dam Hus­sein, Mus­solini, Kim Il-sung, Kim Jong-il, Du­va­lier, Ceaușescu, and Pol Pot, among many oth­ers.

In fact, peo­ple with in­creased malev­olent traits might even be over­rep­re­sented among busi­ness (Babiak et al., 2010; Boddy et al., 2010; Lilien­feld, 2014), mil­i­tary, and poli­ti­cal lead­ers (Post, 2003; Lilien­feld et al., 2012), per­haps be­cause malev­olent traits—es­pe­cially Machi­avel­li­anism and nar­cis­sism—of­ten en­tail an ob­ses­sion with gain­ing power and fame (Ka­jo­nius et al., 2016; Lee et al., 2013; Southard & Zei­gler-Hill, 2016) and could even be ad­van­ta­geous in gain­ing power (Del­uga, 2011; Tay­lor, 2019). Again, Ap­pendix B pro­vides more de­tails.

His­tory sug­gests that malev­olent lead­ers have caused enor­mous harm

One rea­son for ex­pect­ing malev­olent hu­mans in power to pose risks to the fu­ture is that they seem to have caused great harm in the past.

Hitler, Mao, and Stalin were di­rectly in­volved in sev­eral of the great­est atroc­i­ties in his­tory, such as World War II, the Holo­caust, the Great Leap For­ward, the Cul­tural Revolu­tion, and the Great Ter­ror. There thus seems to be a cor­re­la­tion be­tween the malev­olence of (au­to­cratic) poli­ti­cal lead­ers and the amount of harm that oc­curred un­der their rule; at least ac­cord­ing to our un­der­stand­ing of his­tory.[5] If the past is any guide to the fu­ture, in­di­vi­d­u­als with highly ele­vated dark traits could again man­age to rise to po­si­tions of ex­treme power and cause ex­traor­di­nary harm.

How­ever, cor­re­la­tion does not im­ply cau­sa­tion. Even if we grant that this cor­re­la­tion in­deed ex­ists[6], one could ar­gue that there are bet­ter ex­pla­na­tions for how these atroc­i­ties came about. In par­tic­u­lar, it seems plau­si­ble that other fac­tors—such as poli­ti­cal in­sta­bil­ity or ex­trem­ist ide­olo­gies—mat­ter most. We dis­cuss these is­sues in more de­tail in this sec­tion of Ap­pendix A.

It’s also worth men­tion­ing that in­di­vi­d­u­als with malev­olent per­son­al­ities are more likely to adopt dan­ger­ous ide­olo­gies. Dark Te­trad traits are as­so­ci­ated with poli­ti­cal ex­trem­ism gen­er­ally, in­clud­ing sup­port­ing the use of vi­o­lence to achieve poli­ti­cal and other ide­olog­i­cal goals (Dus­para & Gre­it­e­meyer, 2017; Međe­dović & Kneže­vić, 2018; Gøtzsche-Astrup, 2019; Jones, 2013).[7]

Thus, while we agree that his­tory is largely shaped by eco­nomic, poli­ti­cal, cul­tural, in­sti­tu­tional, ide­olog­i­cal and other sys­temic forces, we be­lieve that the per­son­al­ity traits of in­di­vi­d­ual lead­ers—at the very least in au­to­cratic regimes—can plau­si­bly make a sub­stan­tial differ­ence as well (see also Ber­toli et al., 2019; By­man & Pol­lack, 2001; es­pe­cially p. 115-121; Gal­lagher & Allen, 2014; Jones & Olken, 2005). After all, there were hu­mans who rose to power within rather au­to­cratic regimes but who nev­er­the­less en­acted rel­a­tively benefi­cial poli­cies. Ex­am­ples in­clude Juan Car­los I of Spain, Mustafa Ke­mal Atatürk, Mikhail Gor­bachev, Lee Kuan Yew, and Mar­cus Aure­lius, who seemed to ex­hibit more benev­olent per­son­al­ity traits than the likes of Hitler, Mao, and Stalin.

Malev­olent lead­ers have the po­ten­tial to cor­rupt hu­man­ity’s long-term future

One could ques­tion whether malev­olent in­di­vi­d­u­als can sub­stan­tially in­fluence the long-term tra­jec­tory of hu­man­ity for the worse, even from po­si­tions of ex­treme power. It is pos­si­ble that they only cause short-term harm, in which case re­duc­ing malev­olence may not be a pri­or­ity from a longter­mist per­spec­tive.

How­ever, we be­lieve malev­olent lead­ers plau­si­bly have a sig­nifi­cant detri­men­tal effect on the long-term fu­ture. Hitler, Stalin, and Mao, for in­stance, seemed to have had a profoundly nega­tive in­fluence on global af­fairs and in­ter­na­tional co­op­er­a­tion, some of which can ar­guably still be felt to­day, more than half a cen­tury af­ter the atroc­i­ties they per­pe­trated.[8] That said, it is difficult—if not im­pos­si­ble—to as­sess long-term im­pacts, as we do not know what would have hap­pened coun­ter­fac­tu­ally.

Broad risk fac­tors due to malev­olent leaders

Beck­stead (2013) asks whether there is “a com­mon set of broad fac­tors which, if we push on them, sys­tem­at­i­cally lead to bet­ter fu­tures”. It seems plau­si­ble that malev­olent hu­mans in power would push such fac­tors in the wrong di­rec­tion.[9]

Speci­fi­cally, we con­jec­ture that malev­olent hu­mans in power would af­fect the risk fac­tors be­low in the fol­low­ing ways:

  • In­crease the spread of poli­ti­cal ex­trem­ism and other dan­ger­ous ide­olo­gies (see again Ap­pendix A).

  • Ex­ac­er­bate the risk of great power wars and in­ter­na­tional con­flict (By­man & Pol­lack, 2001, par­tic­u­larly p. 112, 134, 137-138; Gal­lagher & Allen, 2014)[10], in­clud­ing the risk of nu­clear war and arms races in­volv­ing trans­for­ma­tive AI.

  • In­crease the like­li­hood of the for­ma­tion of a global to­tal­i­tar­ian regime, po­ten­tially re­sult­ing in a per­ma­nent lock-in of harm­ful val­ues and power struc­tures.[11]

  • In­crease the like­li­hood of reck­less be­havi­our, rather than care­ful re­flec­tion, in high-stakes situ­a­tions (for ex­am­ple, those re­sem­bling the Cuban Mis­sile Cri­sis).

    • Dark Triad traits, psy­chopa­thy in par­tic­u­lar, are as­so­ci­ated with ex­treme risk-tak­ing (Hosker-Field et al., 2016; Visser et al., 2014).

  • In­crease in­tra­na­tional con­flict and un­der­mine pub­lic in­sti­tu­tions, so­cial co­or­di­na­tion, col­lec­tive de­ci­sion mak­ing and gen­eral dis­course[12], par­tic­u­larly by:

    • Ex­ac­er­bat­ing eco­nomic and so­cial in­equal­ity.[13]

    • In­creas­ing cor­rup­tion (Ben­da­han et al., 2015, table 5), rent-seek­ing, and the risk of fi­nan­cial crises (Boddy, 2011).

    • Re­duc­ing ac­cess to in­for­ma­tion, e.g., through cen­sor­ship and pro­pa­ganda.[14]

    • Re­duc­ing trust in gov­ern­ment and in­sti­tu­tions (Bowler & Karp, 2004).[15]

Such trends would plau­si­bly lead to worse fu­tures in ex­pec­ta­tion. They also plau­si­bly in­crease ex­is­ten­tial risks (in­clud­ing ex­tinc­tion risks) and suffer­ing risks (see the next sec­tion). How­ever, the ev­i­dence link­ing these risk fac­tors to malev­olent hu­mans in power is fairly weak, for var­i­ous rea­sons. We are there­fore only some­what con­fi­dent in these con­nec­tions.

Ex­is­ten­tial and suffer­ing risks due to malev­olent leaders

In terms of more con­crete sce­nar­ios, the most ex­treme risks to the long-term fu­ture would ar­guably re­sult from malev­olent hu­mans with ac­cess to highly ad­vanced tech­nol­ogy, par­tic­u­larly trans­for­ma­tive AI.

The fol­low­ing list out­lines some (non-ex­haus­tive) ex­am­ples of how malev­olent in­di­vi­d­u­als could in­crease ex­is­ten­tial and suffer­ing risks:

  • As noted above, malev­olent in­di­vi­d­u­als tend to ex­hibit more risk-tak­ing be­havi­our. In the con­text of a pro­ject to de­velop and de­ploy trans­for­ma­tive AI, they are there­fore more likely to ig­nore po­ten­tial warn­ing signs and omit pre­cau­tion­ary mea­sures. This in­creases the risk of mis­al­igned trans­for­ma­tive AI.

  • Malev­olent hu­mans are likely less op­posed to mak­ing threats than the av­er­age hu­man (Jona­son et al., 2012; Ullrich et al., 2001)[16] and plau­si­bly less mo­ti­vated to pur­sue peace­ful bar­gain­ing strate­gies. Con­flicts in­volv­ing malev­olent hu­mans are there­fore sig­nifi­cantly more likely to es­ca­late and re­sult in catas­trophic out­comes. Also, it could be dan­ger­ous if AI sys­tems in­herit some of their val­ues or heuris­tics, such as an in­creased will­ing­ness to make and carry out threats and/​or a re­duced will­ing­ness to com­pro­mise.

  • Ad­vanced tech­nol­ogy might en­able sadis­tic in­di­vi­d­u­als in power to cre­ate suffer­ing on an un­prece­dented scale.

  • A malev­olent in­di­vi­d­ual, or a small group of such in­di­vi­d­u­als—e.g., the in­ner cir­cle of an au­to­cratic state—might man­age to ob­tain con­trol of Earth (cf. MacAskill, 2020)[17], and even­tu­ally the ob­serv­able uni­verse. For ex­am­ple, imag­ine Hitler or Stalin had ac­cess to ad­vanced tech­nol­ogy—in­clud­ing al­igned AGI and mind up­load­ing, en­abling im­mor­tal­ity. Such a lock-in of per­ma­nent rule by a (global) malev­olent dic­ta­tor would clearly qual­ify as an ex­is­ten­tial risk, as it would thwart any prospect of a more valuable fu­ture. It also con­sti­tutes a sig­nifi­cant s-risk as there would be no­body left to keep any sadis­tic ten­den­cies of the dic­ta­tor in check.

While spe­cific sce­nar­ios are nec­es­sar­ily spec­u­la­tive, it seems clear that malev­olent lead­ers pose a se­ri­ous threat to hu­man­ity’s long-term fu­ture. Of course, malev­olent lead­ers are not the root of all evil, and many con­flicts, wars and atroc­i­ties would hap­pen with­out them. Nev­er­the­less, we be­lieve that pre­vent­ing malev­olent in­di­vi­d­u­als from ris­ing to power is likely valuable and ro­bustly pos­i­tive, ac­cord­ing to al­most all moral per­spec­tives (com­pare also Beck­stead, 2013; To­masik, 2013a, 2013b).

In­ter­ven­tions to re­duce the in­fluence of malev­olent actors

Ad­vanc­ing the sci­ence of malevolence

Fur­ther re­search into the con­struct of malev­olence and its con­se­quences would al­low us to make more rigor­ous state­ments about the links be­tween malev­olent lead­ers and bad out­comes.

A more es­tab­lished sci­ence of malev­olence would also help raise aware­ness of malev­olent per­son­al­ity traits and how to de­tect them among the gen­eral pub­lic, in­fluencers, poli­ti­ci­ans, re­searchers, and aca­demics. Gen­er­ally, the more we know about malev­olence, the eas­ier it is to ac­com­plish many of the in­ter­ven­tions dis­cussed be­low.

Devel­op­ing bet­ter con­structs and mea­sures of malevolence

It seems worth­while to de­velop con­structs cap­tur­ing more pre­cisely the con­stel­la­tion of traits most wor­ri­some from a longter­mist per­spec­tive, as ex­ist­ing con­structs will not always do so.

For ex­am­ple, one of the most com­monly used scales to mea­sure psy­chopa­thy, the Psy­chopa­thy Check­list-Re­vised by Hare et al. (1990), con­sists of 20 items, grouped into two fac­tors. Fac­tor 1—char­ac­ter­ized by cru­elty, grandios­ity, ma­nipu­la­tive­ness, and a lack of guilt—ar­guably rep­re­sents the core per­son­al­ity traits of psy­chopa­thy. How­ever, scor­ing highly on fac­tor 2—char­ac­ter­ized by im­pul­sivity, re­ac­tive anger, and lack of re­al­is­tic goals—is less prob­le­matic from our per­spec­tive. In fact, hu­mans scor­ing high on fac­tor 1 but low on fac­tor 2 are prob­a­bly more dan­ger­ous than hu­mans scor­ing high on both fac­tors (more on this be­low). Gen­er­ally, most mea­sures of psy­chopa­thy in­clude items re­lated to in­creased im­pul­sivity (e.g., Cooke & Michie, 2001; Leven­son et al., 1995; Lilien­feld & An­drews, 1996).

The most dan­ger­ous in­di­vi­d­u­als tend to go undiagnosed

In­di­vi­d­u­als offi­cially di­ag­nosed as malev­olent—e.g. those di­ag­nosed with psy­chopa­thy, an­ti­so­cial or nar­cis­sis­tic per­son­al­ity di­s­or­der—are prob­a­bly un­rep­re­sen­ta­tive of the most dan­ger­ous in­di­vi­d­u­als. This is be­cause an offi­cial di­ag­no­sis is only made when some­body suffers from im­me­di­ate and se­vere prob­lems (re­lat­ing to their malev­olence) or was forced to seek ther­apy, e.g., be­cause they com­mit­ted a crime.

In con­trast, malev­olent hu­mans with good im­pulse-con­trol and oth­er­wise de­cent men­tal health have no rea­son to seek out a ther­a­pist and will gen­er­ally not be con­victed of crimes. The most dan­ger­ous malev­olent hu­mans will re­al­ize that not be­ing un­masked as malev­olent is of the high­est im­por­tance, and will have suffi­cient mo­ti­va­tion, cun­ning, self-aware­ness, charisma, so­cial skills, in­tel­li­gence and im­pulse-con­trol to avoid de­tec­tion (Pe­rina et al., 2020).[18] Such in­di­vi­d­u­als might even de­liber­ately dis­play per­son­al­ity char­ac­ter­is­tics en­tirely at odds with their ac­tual per­son­al­ity. In fact, many dic­ta­tors did pre­cisely that and por­trayed them­selves—of­ten suc­cess­fully—as self­less vi­sion­ar­ies, tire­lessly work­ing for the greater good (e.g., Diköt­ter, 2019). It may there­fore be very valuable to con­duct more re­search on this hard-to-de­tect type of con­scien­tious, strate­gic malev­olence (cf., e.g., Gao & Raine, 2010; Lilien­feld et al., 2015; Mul­lins-Sweatt et al., 2010).

Ma­nipu­la­tion-proof mea­sures of malevolence

To pre­vent malev­olent hu­mans from reach­ing highly in­fluen­tial po­si­tions, we need to be able to re­li­ably de­tect those traits.

Cur­rently, most mea­sures of dark traits take the form of ei­ther in­ter­views or self-com­pleted ques­tion­naires. Smart malev­olent hu­mans can eas­ily ma­nipu­late these types of in­stru­ments and evade de­tec­tion by ly­ing. It is key, there­fore, that we de­velop ma­nipu­la­tion-proof mea­sures of malev­olence, i.e., mea­sures that can­not (eas­ily) be gamed.

One pos­si­bil­ity would be to ask peers and pre­vi­ous as­so­ci­ates to eval­u­ate the per­son­al­ity traits of the per­son in ques­tion.[19] Of course, this raises sev­eral prob­lems. Malev­olent hu­mans could have charmed and fooled many of their (former) friends and col­leagues. They could also bribe or ma­nipu­late oth­ers to lie. So, while other-re­port mea­sures (e.g., 360 de­gree as­sess­ments) may be harder to ma­nipu­late than self-re­ported ones and are there­fore valuable, they are un­likely to com­pletely solve the prob­lem.[20]

Phys­iolog­i­cal or neu­ro­biolog­i­cal mea­sures based on meth­ods like EEG or fMRI might be par­tic­u­larly difficult to ma­nipu­late—though this would prob­a­bly re­quire sub­stan­tial tech­nolog­i­cal and sci­en­tific progress. Neu­roimag­ing tech­niques might al­low us to iden­tify ab­nor­mal brain struc­tures or de­tect sus­pi­cious be­hav­ior, such as show­ing neu­rolog­i­cal signs of plea­sure and/​or no dis­tress when see­ing other hu­mans or an­i­mals in pain. There­fore, more neu­ro­biolog­i­cal re­search on the neu­rolog­i­cal sig­na­tures of plea­sure and dis­plea­sure (e.g., Ber­ridge & Kringelbach, 2013), and on the neu­ro­biol­ogy of sadism and psy­chopa­thy, might be very valuable.[21] (Note that we have not in­ves­ti­gated this in de­tail, so it is prob­a­bly best to start with a sys­tem­atic liter­a­ture re­view.) How­ever, such meth­ods also raise eth­i­cal ques­tions about judg­ing peo­ple by brain scans rather than their ac­tual be­hav­ior.

Po­ten­tial mi­suse and nega­tive consequences

Ma­nipu­la­tion-proof mea­sures of malev­olence could also be mi­sused—like all tech­nol­ogy. For in­stance, gov­ern­ments might falsely brand poli­ti­cal op­po­nents as psy­chopaths.

Another con­cern is that such tests may con­sti­tute an un­fair form of dis­crim­i­na­tion against hu­mans with cer­tain traits. This is be­cause they mea­sure in­nate char­ac­ter­is­tics that are im­pos­si­ble to change, rather than ex­clu­sively con­sid­er­ing the ac­tual be­havi­our of in­di­vi­d­u­als. Also, even if this is deemed ac­cept­able in the case of malev­olence, ad­vo­cat­ing for test­ing in this con­text might lead to the wide­spread adop­tion of per­son­al­ity test­ing in gen­eral, which some be­lieve could have nega­tive con­se­quences. (On the other hand, ex­ist­ing se­lec­tion pro­ce­dures also im­plic­itly or ex­plic­itly se­lect based on in­nate traits such as in­tel­li­gence, and also in­clude var­i­ous kinds of tests.)

Lastly, un­less tests of malev­olence have perfect val­idity and re­li­a­bil­ity, there will be mea­sure­ment er­rors: Some peo­ple will be di­ag­nosed as highly malev­olent even though they aren’t, and some truly malev­olent peo­ple will es­cape de­tec­tion.

How valuable would ma­nipu­la­tion-proof mea­sures of malev­olence be in prac­tice?

Given the po­ten­tially enor­mous benefits, why has there been so lit­tle in­ter­est in the de­vel­op­ment of ma­nipu­la­tion-proof tests of malev­olence? First, do­ing so is likely difficult and, es­pe­cially if it in­volves neu­ro­science re­search, ex­pen­sive[22] (as an ex­am­ple, MRI ma­chines cost be­tween $0.3M and $3M). Se­cond, malev­olent hu­mans might, in some cases, ac­tu­ally benefit in­di­vi­d­ual com­pa­nies or poli­ti­cal par­ties: high lev­els of psy­chopa­thy and nar­cis­sism could be use­ful for things like ne­go­ti­at­ing, mo­ti­vat­ing em­ploy­ees, or win­ning pub­lic ap­proval. Third, most peo­ple likely over­es­ti­mate their abil­ity to dis­cern malev­olent traits in oth­ers, mak­ing them less in­ter­ested in such tests. Fi­nally, it seems that tests in gen­eral are not used much in at least some con­texts; for ex­am­ple, most elected or ap­pointed po­si­tions in gov­ern­ment do not re­quire in­tel­li­gence, knowl­edge, or per­son­al­ity tests.[23]

One might ar­gue that it was ob­vi­ous to most peo­ple that dic­ta­tors such as Hitler, Mao, and Stalin were malev­olent even be­fore they gained power, and that ma­nipu­la­tion-proof mea­sures of malev­olence would there­fore have been use­less. How­ever, we are doubt­ful that peo­ple can eas­ily de­tect malev­olence, at least in the most dan­ger­ous types of in­di­vi­d­u­als, as men­tioned above. (See also the sec­tion How well can peo­ple de­tect malev­olent traits in Ap­pendix A for more de­tails.)

Of course, many did sus­pect that Hitler, Mao, and Stalin were malev­olent. How­ever, this was not com­mon knowl­edge—and with­out ob­jec­tive ev­i­dence, call­ing an in­di­vi­d­ual malev­olent can eas­ily be dis­missed as slan­der. Not to men­tion that any­one mak­ing such ac­cu­sa­tions risks se­ri­ous reprisals. So even if a ma­jor­ity had re­al­ized early on that Hitler, Mao, and Stalin are malev­olent, it might not have helped.

How­ever, if ma­nipu­la­tion-proof and valid mea­sures of malev­olence had ex­isted—alongside strong norms to use them to screen poli­ti­cal lead­ers and wide­spread trust in their ac­cu­racy—it could have been com­mon knowl­edge that these in­di­vi­d­u­als were malev­olent, which would have sig­nifi­cantly re­duced their chance of ris­ing to power. Essen­tially, ma­nipu­la­tion-proof and valid mea­sures of malev­olence could serve as an ob­jec­tive ar­biter of good in­ten­tions, analo­gous to the sci­en­tific use of ex­per­i­ments as ob­jec­tive ar­biters of truth. Their role in a hiring pro­cess could then be com­pared to se­cu­rity clear­ances, for in­stance.

It is not clear whether even perfectly di­ag­nos­tic mea­sures of malev­olence would ever be­come wide­spread—for ex­am­ple, be­cause of the above­men­tioned eth­i­cal con­cerns. How­ever, for highly in­fluen­tial po­si­tions, peo­ple are most will­ing to make use of all available ev­i­dence (and can­di­dates for such po­si­tions have an in­cen­tive to provide cred­ible sig­nals of trust­wor­thi­ness). For in­stance, re­ceiv­ing a top-se­cret se­cu­rity clear­ance in­volves ex­ten­sive in­ter­views with one’s (former) spouses, col­leagues, friends and neigh­bors, alongside re­views of med­i­cal and psy­chi­a­tric records, and some­times even poly­graph ex­am­i­na­tions. This elab­o­rate and ar­guably pri­vacy-vi­o­lat­ing pro­cess would be un­ac­cept­able for a rou­tine job, but is con­sid­ered ap­pro­pri­ate given the stakes at hand. Last, it could already be valuable if only a few com­pa­nies or gov­ern­ment de­part­ments started us­ing ma­nipu­la­tion-proof mea­sures of malev­olence; near-uni­ver­sal adop­tion of such mea­sures is by no means nec­es­sary.

De­spite these caveats, we be­lieve that work on ma­nipu­la­tion-proof mea­sures of malev­olence is promis­ing. Sub­ject to per­sonal fit, it may be worth­while for some effec­tive al­tru­ists to con­sider ca­reers in psy­chol­ogy or neu­ro­science. This would al­low them to ad­vance the sci­ence of malev­olence, con­tribute to the de­vel­op­ment of ma­nipu­la­tion-proof mea­sures of malev­olence, and im­prove their chances to con­vince de­ci­sion-mak­ers to take such mea­sures se­ri­ously.

Poli­ti­cal interventions

Many fac­tors de­ter­mine whether an in­di­vi­d­ual can rise to a po­si­tion of power, and it is im­por­tant to in­clude (non-)malev­olence as a crite­rion when se­lect­ing lead­ers. Ideally, we should es­tab­lish strong norms against al­low­ing highly malev­olent lead­ers to rise to power—even in cases where ele­vated Dark Te­trad traits may be in­stru­men­tal in ad­vanc­ing the in­ter­ests of a com­pany or na­tion.

While the no­tion of Dark Te­trad traits is not fore­most in most peo­ple’s minds, one could ar­gue that much poli­ti­cal de­bate is about re­lated con­cepts like the trust­wor­thi­ness or hon­esty of can­di­dates, and vot­ers do value those at­tributes.[24]

Per­haps the key is­sue, then, is not a lack of aware­ness; rather the non-availa­bil­ity of re­li­able ob­jec­tive mea­sures and the over­es­ti­ma­tion of peo­ple’s abil­ity to de­tect malev­olence. In fact, hu­mans seem too ea­ger to view their poli­ti­cal op­po­nents as in­her­ently malev­olent and ill-in­ten­tioned. Con­versely how­ever, hu­mans also tend to view mem­bers of their own tribe as in­her­ently good and over­look their mis­deeds. (See again Ap­pendix A for more de­tails.)

The me­dia also tends to de­pict im­pul­sive psy­chopaths—say, ruth­less se­rial kil­lers with a long his­tory of vi­o­lence or crime. Th­ese are rel­a­tively easy to de­tect, po­ten­tially lead­ing to a false sense of se­cu­rity (com­pare also Babiak et al., 2010, p.174-175). As men­tioned above, it may there­fore be valuable to raise aware­ness that at least some types of malev­olent hu­mans are difficult to de­tect.

Alter­na­tively, we could in­fluence poli­ti­cal back­ground fac­tors that make malev­olent lead­ers more or less likely. It seems plau­si­ble that poli­ti­cal in­sta­bil­ity, es­pe­cially out­right rev­olu­tions, en­able malev­olent hu­mans to rise to power (Col­gan, 2013, p. 662-665). Gen­er­ally, democ­ra­cies plau­si­bly se­lect for more trust­wor­thy, pre­dictable and benev­olent lead­ers (By­man & Pol­lack, 2001, p.139-140). Thus, in­ter­ven­tions to pro­mote democ­racy and re­duce poli­ti­cal in­sta­bil­ity seem valuable—though this area seems rather crowded.

Even within es­tab­lished democ­ra­cies, we could try to iden­tify mea­sures that avoid ex­ces­sive po­lariza­tion and in­stead re­ward cross-party co­op­er­a­tion and com­pro­mise. Miti­gat­ing the of­ten highly com­bat­ive na­ture of poli­tics would plau­si­bly make it harder for malev­olent hu­mans to rise to power.[25] (For ex­am­ple, effec­tive al­tru­ists have dis­cussed elec­toral re­form as a pos­si­ble lever that could help achieve this.)

Since ele­vated Dark Te­trad traits are sig­nifi­cantly more com­mon among men (Paulhus & Willi­ams, 2002; Plouffe et al., 2017), it also seems benefi­cial to ad­vance gen­der equal­ity and in­crease the pro­por­tion of fe­male lead­ers.

Other po­ten­tial fac­tors that might fa­cil­i­tate the rise of malev­olent in­di­vi­d­u­als in­clude so­cial and eco­nomic in­equal­ity, poverty, eth­nic, mil­i­tary or re­li­gious con­flicts, and a “wide­spread sense of grievance or re­sent­ment” (Glad, 2002, p. 4). Thus, iden­ti­fy­ing cost-effec­tive in­ter­ven­tions to im­prove these fac­tors (as well as iden­ti­fy­ing fac­tors we haven’t thought of) could be promis­ing. A more thor­ough study of the his­tory of malev­olent hu­mans ris­ing to power would also be valuable to bet­ter un­der­stand which fac­tors are most pre­dic­tive.

Over­all, it seems plau­si­ble that many promis­ing poli­ti­cal in­ter­ven­tions to pre­vent malev­olent hu­mans from ris­ing to power have already been iden­ti­fied and im­ple­mented—such as, e.g., checks and bal­ances, the sep­a­ra­tion of pow­ers, and democ­racy it­self. After all, much of poli­ti­cal sci­ence and poli­ti­cal philos­o­phy is about pre­vent­ing the con­cen­tra­tion of power in the wrong hands.[26] We nev­er­the­less en­courage in­ter­ested read­ers to fur­ther ex­plore these top­ics.

Fu­ture tech­nolo­gies and malevolence

In this sec­tion, we ex­plore how pos­si­ble fu­ture tech­nolo­gies could be used to re­duce the in­fluence of malev­olent ac­tors.

Whole brain emulation

Whole brain em­u­la­tion is the hy­po­thet­i­cal pro­cess of scan­ning the struc­ture of a brain and repli­cat­ing it on a com­puter. Han­son (2016) ex­plores the pos­si­ble im­pli­ca­tions of this tech­nol­ogy. In his sce­nario, brain em­u­la­tions (“ems”) will shape fu­ture eco­nomic, tech­nolog­i­cal and poli­ti­cal pro­cesses due to their com­pet­i­tive ad­van­tage over biolog­i­cal minds.

One key ques­tion is: which hu­man brains will be up­loaded? We be­lieve that it would be cru­cial to screen po­ten­tial ems for malev­olence—par­tic­u­larly the first in­di­vi­d­u­als to be up­loaded. Con­sid­er­ing the power that the first ems would likely have to shape this new “Age of Em”, it could be dis­as­trous for hu­man­ity’s long-term fu­ture if a malev­olent in­di­vi­d­ual forms the ba­sis for (some of) the first ems (cf. Bostrom, 2002, p. 12). Con­versely, by screen­ing for malev­olence, us­ing ma­nipu­la­tion-proof mea­sures, we could effec­tively re­duce malev­olence among ems. This offers an un­prece­dented op­por­tu­nity to en­sure that malev­olent forces have sig­nifi­cantly less in­fluence over the long-term fu­ture.[27]

Trans­for­ma­tive AI

Many longter­mist effec­tive al­tru­ists think that shap­ing trans­for­ma­tive ar­tifi­cial in­tel­li­gence, and in par­tic­u­lar solv­ing the al­ign­ment prob­lem, is a par­tic­u­larly good lever to im­prove the long-term fu­ture. Some con­crete pro­pos­als for al­ign­ment—such as Iter­ated dis­til­la­tion and am­plifi­ca­tion—in­volve a “hu­man-in-the-loop” whose feed­back is used to al­ign in­creas­ingly ca­pa­ble AI.

In these sce­nar­ios, the “hu­man-in-the-loop” plau­si­bly has enor­mous re­spon­si­bil­ity and lev­er­age over the long-term fu­ture. It is there­fore ex­tremely valuable to en­sure that the rele­vant in­di­vi­d­ual or in­di­vi­d­u­als—if e.g. a jury or par­li­a­ment fulfills the role of “hu­man-in-the-loop”—do not ex­hibit malev­olent traits. (Again, this re­quires or is at least fa­cil­i­tated by the availa­bil­ity of ma­nipu­la­tion-proof mea­sures.)

Even with­out hu­man in­volve­ment, ar­tifi­cial agents may ex­hibit be­havi­our that re­sem­bles malev­olence (to the ex­tent that this no­tion makes sense in non-hu­man con­texts) if such heuris­tics prove use­ful in its train­ing pro­cess. After all, the fact that malev­olent traits such as psy­chopa­thy or sadism evolved in some hu­mans sug­gests that those traits pro­vided fit­ness ad­van­tages, at least in cer­tain con­texts (Book et al. 2015; McDon­ald et al., 2012; Nell, 2006; Jona­son et al., 2015).

In par­tic­u­lar, it is pos­si­ble that do­main-gen­eral ca­pa­bil­ities will emerge via in­creas­ingly com­plex multi-agent in­ter­ac­tions (Baker et al., 2019). In this case, it is cru­cial that the train­ing en­vi­ron­ment is set up in a way that pre­vents the evolu­tion of un­de­sir­able traits like malev­olence, and in­stead re­wards co­op­er­a­tive and trust­wor­thy be­havi­our.

To the ex­tent that ar­tifi­cial in­tel­li­gence de­signs are in­spired by the hu­man brain (“neu­ro­mor­phic AI”), it seems im­por­tant to un­der­stand the neu­ro­scien­tific ba­sis of malev­olence in hu­mans to re­duce the risk of neu­ro­mor­phic AIs also ex­hibit­ing malev­olent traits.

Ge­netic enhancement

A third class of rele­vant new tech­nolo­gies are those that make it pos­si­ble to change the ge­netic makeup of fu­ture hu­mans. This would offer un­prece­dented lev­er­age to change per­son­al­ity traits and “hu­man na­ture”, for bet­ter or for worse (cf. Ge­netic En­hance­ment as a Cause Area). In par­tic­u­lar, se­lec­tion against malev­olent traits could sig­nifi­cantly re­duce the in­fluence of malev­olent in­di­vi­d­u­als.

This is be­cause most var­i­ance in adult per­son­al­ity is due to ge­netic in­fluences (~30–50%) and non­shared en­vi­ron­ment effects (~35–55%), leav­ing com­par­a­tively lit­tle room for the shared en­vi­ron­ment (~5–25%)[28] (e.g., Knopik et al., 2018, ch. 16; John­son et al., 2008; Plomin, 2019; Vukaso­vić & Bratko, 2015). (See the sec­tions “Broad-sense her­i­ta­bil­ity es­ti­mates of dark traits” and “Is se­lect­ing for per­son­al­ity traits pos­si­ble?” in Ap­pendix B for more de­tails.) By con­trast, non­shared en­vi­ron­men­tal in­fluences—which in­clude mea­sure­ment er­ror, chance life events, and de novo mu­ta­tions—seem to be mostly un­sys­tem­atic, idiosyn­cratic, and un­sta­ble, and there­fore difficult to in­fluence (Plomin, 2019, ch. 7).

Ge­netic en­hance­ment tech­nolo­gies might also re­sult in the cre­ation of hu­mans with ex­traor­di­nary in­tel­li­gence (see, e.g., Shul­man & Bostrom, 2014, p.2-3). Such hu­mans, if cre­ated, will likely be over­rep­re­sented in po­si­tions of enor­mous in­fluence and would thus have an out­sized im­pact on the long-term fu­ture. Re­duc­ing malev­olence among those in­di­vi­d­u­als is there­fore es­pe­cially im­por­tant.

Overview of ge­netic en­hance­ment technologies

There are var­i­ous tech­nolo­gies that would make it pos­si­ble to mod­ify the ge­netic makeup of fu­ture hu­mans. We think the fol­low­ing four are most rele­vant:

  • In vitro fer­til­iza­tion (IVF) is a pro­cess of fer­til­i­sa­tion where an egg is com­bined with sperm out­side the body. In sev­eral Western coun­tries, 2-8% of new­borns are already con­ceived in this way. While this is pri­mar­ily used to ad­dress in­fer­til­ity, it is pos­si­ble to cre­ate sev­eral fer­til­ized eggs and se­lect among those.

  • Gene edit­ing (e.g., via CRISPR) is the in­ser­tion, dele­tion, mod­ifi­ca­tion or re­place­ment of DNA in the genome of an or­ganism.

  • Iter­ated em­bryo se­lec­tion (IES, Shul­man & Bostrom, 2014; Spar­row, 2013) takes a sam­ple of em­bryos and re­peats two steps: a) se­lect em­bryos that are higher in de­sired ge­netic char­ac­ter­is­tics; b) ex­tract stem cells from those em­bryos, con­vert them to sperm and ova, and cross those to pro­duce new em­bryos.

  • Genome syn­the­sis is the ar­tifi­cial man­u­fac­tur­ing of DNA, base pair by base pair.

Note that these meth­ods can in­ter­act with each other and should thus not be viewed as be­ing com­pletely sep­a­rate. For more de­tails, we highly recom­mend Gw­ern’s Em­bryo se­lec­tion for in­tel­li­gence.

How far away are these tech­nolo­gies? Gw­ern writes that “IES is still dis­tant and de­pends on a large num­ber of wet lab break­throughs and fine­tuned hu­man-cell pro­to­cols.” Nonethe­less, he states that: “[...] it seems clear, at least, that it will cer­tainly not hap­pen in the next decade, but af­ter that…?”. He con­cludes that “IES has been badly un­der-dis­cussed to date.”

Re­gard­ing genome syn­the­sis, Gw­ern writes that the “cost curve sug­gests that around 2035, whole hu­man genomes reach well-re­sourced re­search pro­ject ranges of $10-30m” and that it “is en­tirely pos­si­ble that IES will de­velop too slowly and will be ob­so­leted by genome syn­the­sis in 10-20 years.”

Gw­ern gives the fol­low­ing sum­mary:

“CRISPR & clon­ing are already available but will re­main unim­por­tant in­definitely for var­i­ous fun­da­men­tal rea­sons; [...] mas­sive mul­ti­ple em­bryo se­lec­tion is some ways off but in­creas­ingly in­evitable and the gains are large enough on both in­di­vi­d­ual & so­cietal lev­els to re­sult in a shock; IES will come some­time af­ter mas­sive mul­ti­ple em­bryo se­lec­tion but it’s im­pos­si­ble to say when, al­though the con­se­quences are po­ten­tially global; genome syn­the­sis is a similar level of se­ri­ous­ness, but is much more pre­dictable and can be looked for, very loosely, 2030-2040 (and pos­si­bly sooner).”

Dangers

Ge­netic en­hance­ment is widely crit­i­cized. Numer­ous atroc­i­ties have been com­mit­ted in the quest to forge a new, “bet­ter” kind of hu­man. We would like to em­pha­size that we do not ad­vo­cate for ge­netic en­hance­ment per se. We only ar­gue that if ge­netic en­hance­ment hap­pens, it seems prima fa­cie im­por­tant to se­lect against malev­olent traits—com­pa­rable to the ra­tio­nale be­hind differ­en­tial in­tel­lec­tual progress and differ­en­tial tech­nolog­i­cal progress.

Still, we are tread­ing dan­ger­ous wa­ters. Even just bring­ing up the pos­si­bil­ity of se­lec­tion for or against per­son­al­ity traits might in­spire mi­suse of such meth­ods. One par­tic­u­larly wor­ri­some sce­nario is se­lec­tion against all forms of re­bel­lion and in­de­pen­dence, branded as “an­ti­so­cial ten­den­cies”, which could en­able ex­treme to­tal­i­tar­i­anism. Gen­er­ally, the cur­rently dom­i­nant in­di­vi­d­u­als and classes could abuse these pow­er­ful tech­nolo­gies to ce­ment their power.

It is also worth not­ing that very high lev­els of usu­ally benefi­cial traits can be nega­tive: too much trust, for ex­am­ple, might re­sult in naïvety and an in­creased like­li­hood of be­ing ex­ploited. Similarly, com­pletely elimi­nat­ing usu­ally harm­ful traits could back­fire as well: for ex­am­ple, in cer­tain situ­a­tions, some de­gree of nar­cis­sism and Machi­avel­li­anism may benefit en­trepreneurs and poli­ti­ci­ans. Gen­er­ally, differ­ent per­son­al­ity traits are use­ful for differ­ent roles in so­ciety, so some di­ver­sity is benefi­cial.

For these and other rea­sons, it could be net nega­tive to shift per­son­al­ity traits by more than one or two stan­dard de­vi­a­tions.[29] How­ever, we are pri­mar­ily in­ter­ested in shrink­ing the right tail of the dis­tri­bu­tion—e.g., by se­lect­ing against em­bryos with poly­genic scores for Dark Te­trad traits above, say, the 99th per­centile. This could be done while only marginally de­creas­ing the mean. Gen­er­ally, if we ap­ply the (dou­ble) re­ver­sal test, cur­rent rates of dark traits—par­tic­u­larly highly ele­vated ones—ap­pear very far from op­ti­mal.

More­over, many dark traits ap­pear to be ge­net­i­cally cor­re­lated with each other and nega­tively ge­net­i­cally cor­re­lated with benev­olent per­son­al­ity traits (Ver­non et al., 2008). Thus, se­lect­ing against one dark trait will tend to de­crease other dark traits and in­crease benev­olent traits. This plau­si­bly makes se­lec­tion efforts more ro­bust, though this could also have some down­sides.[30]

Lastly, we should ar­guably be es­pe­cially cau­tious in sce­nar­ios that in­volve ge­net­i­cally en­hanced hu­mans of ex­traor­di­nary in­tel­li­gence. Ex­tremely in­tel­li­gent sadists and psy­chopaths would pose risks that out­weigh any plau­si­ble benefits.

Interventions

Fund­ing or oth­er­wise en­courag­ing more re­search on the ge­netic ba­sis of malev­olent traits would al­low us to bet­ter se­lect against these traits. Ideally, we would have a good un­der­stand­ing of the ge­netic ba­sis of malev­olent traits be­fore tech­nolo­gies such as genome syn­the­sis ar­rive. Thus, it is plau­si­bly time-sen­si­tive to do this re­search now, even if pow­er­ful ge­netic en­hance­ment tech­nolo­gies will not be de­vel­oped for the next sev­eral decades.

A par­tic­u­larly cost-effec­tive in­ter­ven­tion might be to con­vince per­sonal ge­nomics com­pa­nies, such as 23andMe, to offer tests of Dark Te­trad traits. 23andMe has over 10 mil­lion cus­tomers, so even if only a small frac­tion of cus­tomers took these tests (e.g., out of cu­ri­os­ity), we would already achieve sam­ple sizes sur­pass­ing those of large GWA stud­ies.[31] Im­proved psy­cholog­i­cal mea­sures of malev­olence with higher re­li­a­bil­ity and val­idity, as dis­cussed in pre­vi­ous sec­tions, would also en­able GWA stud­ies to bet­ter iden­tify ge­netic var­i­ants as­so­ci­ated with such traits.

In gen­eral, in­creas­ing the so­cial ac­cept­abil­ity of screen­ing for Dark Te­trad traits plau­si­bly in­creases the prob­a­bil­ity that fu­ture pro­jects in­volv­ing more pow­er­ful tech­nol­ogy will also do so. The more es­tab­lished and well-known Dark Te­trad traits are, and the less con­tro­ver­sial their her­i­ta­bil­ity, the eas­ier it would be to ac­com­plish many of the in­ter­ven­tions men­tioned above. It might be valuable, for in­stance, to per­suade sperm banks or other in­sti­tu­tions re­spon­si­ble for screen­ing sperm (or egg) donors to add mea­sures of Dark Te­trad traits to their screen­ing pro­cess and dis­play the re­sults promi­nently to women choos­ing sperm donors.

As with non-ge­netic in­ter­ven­tions, we could at­tempt to raise aware­ness of malev­olent traits, their her­i­ta­bil­ity, and their dan­gers. Rather than try­ing to make changes to the sup­ply side, it might be eas­ier to in­crease de­mand by pop­u­lariz­ing Dark Te­trad traits.[32] Most par­ents want their chil­dren to be re­spon­si­ble, em­pathic, and kind. If they are will­ing to pay for screen­ing for malev­olent traits, then sperm banks or oth­ers will offer such ser­vices.[33]

How­ever, con­sid­er­ing the sig­nifi­cant dan­gers out­lined above, we be­lieve that pub­lic ad­vo­cacy of the idea of ge­netic se­lec­tion against malev­olence would likely be pre­ma­ture. In­deed, more re­search on how to best avoid nega­tive con­se­quences—such as in­creased in­equal­ity or de­hu­man­iza­tion of (un)en­hanced hu­mans—of pos­si­ble in­ter­ven­tions in this area would be im­por­tant.

Sub­ject to per­sonal fit, it may also be worth­while for some effec­tive al­tru­ists to con­sider ca­reers in bioin­for­mat­ics, so­cial sci­ences re­lat­ing to GWA stud­ies, bioethics, or re­lated fields, to be in a good po­si­tion to later in­fluence key play­ers.

Con­clud­ing remarks

Many of the above in­ter­ven­tions face se­ri­ous tech­ni­cal challenges. It may be hard to de­velop ma­nipu­la­tion-proof mea­sures of malev­olence, and se­lec­tion on per­son­al­ity traits is prob­a­bly difficult due to low ad­di­tive her­i­ta­bil­ity. In ad­di­tion, many in­ter­ven­tions—es­pe­cially those re­lated to ge­netic en­hance­ment tech­nolo­gies—en­tail se­vere risks of mi­suse and un­in­tended nega­tive con­se­quences.

How­ever, some of the sug­gested in­ter­ven­tions in­volve nei­ther spec­u­la­tive fu­ture tech­nol­ogy nor con­tro­ver­sial ideas about ge­netic en­hance­ment. Over­all, we recom­mend a mix of differ­ent in­ter­ven­tions, as well as fur­ther work aiming to find new types of in­ter­ven­tions and check­ing the as­sump­tions that un­der­lie ex­ist­ing in­ter­ven­tions.

Most par­ents, cul­tures, and re­li­gions fea­ture some no­tion of “not be­ing evil”, so one could ar­gue that re­duc­ing malev­olence, broadly con­strued, is already quite crowded. How­ever, we be­lieve the in­ter­ven­tions we have ex­plored are more tar­geted, and are po­ten­tially more far-reach­ing and more ne­glected than, say, cul­tural norms or par­ent­ing.

Re­duc­ing the in­fluence of malev­olent ac­tors is not a panacea, of course. Many of the world’s biggest prob­lems are not (pri­mar­ily) due to malev­olent in­tent per se, and in­stead are mostly caused by in­com­pe­tence, ir­ra­tional­ity, in­differ­ence, and our in­abil­ity to co­or­di­nate the es­cape from un­de­sir­able equil­ibria.

That be­ing said, we be­lieve that re­duc­ing the chances of malev­olent in­di­vi­d­u­als ris­ing to power would have sub­stan­tially pos­i­tive effects un­der a broad range of sce­nar­ios and value sys­tems—whether they place pri­mary im­por­tance on avoid­ing ex­is­ten­tial risks, re­duc­ing suffer­ing, or im­prov­ing the qual­ity of the long-term fu­ture.

Ap­pendix A

How im­por­tant are situ­a­tional fac­tors and ide­olo­gies com­pared to per­son­al­ity traits?

In this sec­tion, we dis­cuss the ex­tent to which his­tor­i­cal atroc­i­ties can be at­tributed to the per­son­al­ity traits of in­di­vi­d­u­als ver­sus struc­tural fac­tors.

First, it seems plau­si­ble that back­ground con­di­tions that en­able dic­ta­tor­ships in the first place—such as poli­ti­cal in­sta­bil­ity and an ab­sent rule of law—also make it more likely that malev­olent hu­mans will rise to power. In­di­vi­d­u­als who are re­luc­tant to en­gage in mur­der and be­trayal, for ex­am­ple, will be at a con­sid­er­able dis­ad­van­tage un­der such con­di­tions (also see Col­gan 2013, es­pe­cially p. 662-665).

Similarly, power tends to cor­rupt (e.g., Ben­da­han et al., 2015; Cis­lak et al., 2018) so it could be ar­gued that most in­di­vi­d­u­als who rise to the top within au­to­cratic regimes, will be­come more malev­olent. Gen­er­ally, a wealth of so­cial psy­chol­ogy re­search at­tests to the im­por­tance of situ­a­tional fac­tors in ex­plain­ing hu­man be­hav­ior (Mil­gram, 1963; Burger, 2009), though the un­der­stand­ing of mod­ern psy­chol­ogy is that be­hav­ior de­pends on both situ­a­tional fac­tors and in­di­vi­d­ual per­son­al­ity traits (Bow­ers, 1973; Endler & Mag­nus­son, 1976).

One par­tic­u­larly rele­vant fac­tor is the spread of ex­trem­ist and fa­nat­i­cal ide­olo­gies such as fas­cism, vi­o­lent com­mu­nism, and fun­da­men­tal­ist re­li­gion, which have un­doubt­edly con­tributed to his­tor­i­cal atroc­i­ties. In fact, such ide­olo­gies have plau­si­bly had a much big­ger im­pact on his­tory than the per­son­al­ity traits of in­di­vi­d­u­als and could pose even greater risks to the long-term fu­ture. So why fo­cus on per­son­al­ity rather than ide­ol­ogy or struc­tural fac­tors?

For one, tens of mil­lions of peo­ple are already com­bat­ing the dan­ger­ous ide­olo­gies men­tioned above, or work on en­sur­ing poli­ti­cal sta­bil­ity and rule of law. Th­ese efforts are laud­able, but also seem very crowded, which sug­gests that many of the most cost-effec­tive in­ter­ven­tions have already been iden­ti­fied and car­ried out.

As men­tioned above, there is also am­ple ev­i­dence that in­di­vi­d­u­als with malev­olent per­son­al­ities are drawn to dan­ger­ous ide­olo­gies:[34] Dark Triad traits pre­dict in­creased in­ten­tion to en­gage in poli­ti­cal vi­o­lence (Gøtzsche-Astrup, 2019). Nar­cis­sism and psy­chopa­thy are as­so­ci­ated with poli­ti­cal ex­trem­ism (Dus­para & Gre­it­e­meyer, 2017). Sadis­tic and psy­cho­pathic traits pre­dict en­dors­ing a mil­i­tant ex­trem­ist mind-set, in par­tic­u­lar the use of vi­o­lence to achieve poli­ti­cal and other ide­olog­i­cal goals (Međe­dović & Kneže­vić, 2018). Machi­avel­li­anism and psy­chopa­thy pre­dict racist at­ti­tudes, in­clud­ing sup­port for Neo-Nazis and the KKK (Jones, 2013). Dark Triad traits cor­re­late with so­cial-dom­i­nance ori­en­ta­tion (Jones & Figueredo, 2013; Jones, 2013), a mea­sure of an in­di­vi­d­ual’s prefer­ence for eco­nomic and so­cial in­equal­ity within and be­tween groups (Pratto et al., 1994; Dal­lago et al. 2008).[35]

Most ide­olo­gies also seem open for in­ter­pre­ta­tion, leav­ing suffi­cient room for the idiosyn­cratic be­liefs and per­son­al­ity traits of lead­ers to make a differ­ence. Khrushchev and Gor­bachev, for ex­am­ple, while broadly shar­ing Stalin’s Marx­ist-Len­inist ide­ol­ogy, have caused much less harm than Stalin. Con­versely, as the ex­am­ples of Stal­inism, Mao­ism, and Juche show, malev­olent in­di­vi­d­u­als can de­velop an ex­ist­ing ide­ol­ogy fur­ther, mak­ing it even more harm­ful.

In the end, ide­olo­gies, be­lief sys­tems, and per­son­al­ity traits ap­pear in­evitably in­ter­twined. Nar­cis­sism, for ex­am­ple, en­tails in­flated be­liefs about one’s abil­ities and place in his­tory, by defi­ni­tion. Gen­er­ally, malev­olent in­di­vi­d­u­als tend to hold be­liefs that serve as (un)con­scious jus­tifi­ca­tions for their be­hav­ior, such as a sense of en­ti­tle­ment and grandios­ity, and seem more likely to en­dorse dan­ger­ous wor­ld­views and “ide­olo­gies that fa­vor dom­i­nance (of in­di­vi­d­u­als or groups)” (Mosha­gen, 2018, p. 659).

Fi­nally, it is in­struc­tive to com­pare large-scale atroc­i­ties to small-scale atroc­i­ties like mur­der or con­tract kil­ling. While rates of vi­o­lent crime surely de­pend on so­cial back­ground fac­tors and cul­turally trans­mit­ted norms, psy­chopa­thy is also con­sid­ered a strong pre­dic­tor for homi­cide, in­clud­ing in­stru­men­tal, calcu­lated mur­der (Fox & DeLisi, 2019). If we ac­cept that malev­olent per­son­al­ity traits like psy­chopa­thy play a causal role in vi­o­lent crime, it stands to rea­son that such traits also play at least some causal role in many large-scale atroc­i­ties.

How well can peo­ple de­tect malev­olent traits?

His­tor­i­cal ev­i­dence sug­gests that even many of their poli­ti­cal ad­ver­saries—at least for some time—did not re­al­ize that Hitler, Mao, and Stalin were malev­olent, even af­ter they were in power.

Cham­ber­lain fa­mously trusted Hitler’s sincer­ity for far too long. Churchill once re­marked that ‘‘Poor Cham­ber­lain be­lieved he could trust Hitler. He was wrong. But I don’t think that I am wrong about Stalin’’ (Yer­gin, 1977, p. 65). Similarly, Tru­man be­lieved that Stalin ‘‘could be de­pended upon…I got the im­pres­sion Stalin would stand by his agree­ments’’ (Lar­son, 1988, p. 246).[36] At least un­til the 1940s, many Western­ers and Chi­nese seemed to have been en­am­ored with Mao, po­ten­tially partly due to the in­fluen­tial book ‘Red Star over China’ (Snow, 1937) which painted him in an ex­tremely fa­vor­able light (Chang & Hal­li­day, ch. 18, 2007).

Countless fa­mous in­tel­lec­tu­als—in­clud­ing G.B. Shaw, H.G. Wells, Sartre, Si­mone de Beau­voir, Beatrice Webb, Sid­ney Webb, Su­san Son­tag, Oswald Spen­gler, Carl Jung, Kon­rad Lorenz, and Martin Hei­deg­ger—praised au­thor­i­tar­ian lead­ers like Mus­solini, Hitler, Stalin or Fidel Cas­tro (Hol­lan­der, 2016; 2017). Even to­day, most Rus­si­ans and Chi­nese think highly of Stalin and Mao, re­spec­tively.

In sum­mary, it seems that many hu­mans fail to de­tect malev­olent in­di­vi­d­u­als, par­tic­u­larly when ide­olog­i­cal, pa­tri­otic or other bi­ases af­fect their judg­ment. Gen­er­ally, Hitler, Mao, and Stalin—like many nar­cis­sists—seem to have been quite po­lariz­ing; some thought they were ob­vi­ously malev­olent, oth­ers viewed them as benev­olent, nearly mes­si­anic figures.

Ap­pendix B

See Ap­pendix B: Re­duc­ing long-term risks from malev­olent ac­tors for ad­di­tional de­tails.

Acknowledgements

Thanks to Jesse Clif­ton, Jonas Vol­lmer, Lukas Gloor, Ste­fan Torges, Chi Nguyen, Mo­jmír Stehlík, Richard Ngo, Pablo Staffor­ini, Cas­par Oester­held, Lu­cius Cavi­ola, Jo­hannes Treut­lein, and Ewelina Tur for their valuable com­ments and feed­back. Thanks to Sofia-Davis Fo­gel for copy edit­ing. All er­rors and views ex­pressed in this doc­u­ment are our own, not those of the com­menters.

David’s work on this post was funded by the Cen­ter on Long-Term Risk.

References

Az­i­zli, N., Atk­in­son, B. E., Baugh­man, H. M., Chin, K., Ver­non, P. A., Har­ris, E., & Ve­selka, L. (2016). Lies and crimes: Dark Triad, mis­con­duct, and high-stakes de­cep­tion. Per­son­al­ity and In­di­vi­d­ual Differ­ences, 89, 34-39.

Babiak, P., Neu­mann, C. S., & Hare, R. D. (2010). Cor­po­rate psy­chopa­thy: Talk­ing the walk. Be­hav­ioral sci­ences & the law, 28(2), 174-193.

Baker, B., Kan­itschei­der, I., Markov, T., Wu, Y., Pow­ell, G., McGrew, B., & Mor­datch, I. (2019). Emer­gent tool use from multi-agent au­tocur­ricula. arXiv preprint arXiv:1909.07528.

Baron-Co­hen, S. (2012). The sci­ence of evil: On em­pa­thy and the ori­gins of cru­elty. Ba­sic books.

Beck­stead (2013). How to com­pare broad and tar­geted at­tempts to shape the far fu­ture.

Ben­da­han, S., Zehn­der, C., Pralong, F. P., & An­ton­akis, J. (2015). Leader cor­rup­tion de­pends on power and testos­terone. The Lead­er­ship Quar­terly, 26(2), 101-122.

Ber­ridge, K. C., & Kringelbach, M. L. (2013). Neu­ro­science of af­fect: brain mechanisms of plea­sure and dis­plea­sure. Cur­rent opinion in neu­ro­biol­ogy, 23(3), 294-303.

Ber­toli, A., Dafoe, A., & Trager, R. (2019). Leader Age and In­ter­na­tional Con­flict.

Boddy, C. R., Ladyshewsky, R., & Galvin, P. (2010). Lead­ers with­out ethics in global busi­ness: Cor­po­rate psy­chopaths. Jour­nal of Public Af­fairs, 10(3), 121-138.

Boddy, C. R. (2011). The cor­po­rate psy­chopaths the­ory of the global fi­nan­cial crisis. Jour­nal of Busi­ness Ethics, 102(2), 255-259.

Book, A., Visser, B. A., & Volk, A. A. (2015). Un­pack­ing “evil”: Claiming the core of the Dark Triad. Per­son­al­ity and In­di­vi­d­ual Differ­ences, 73, 29-38.

Bostrom, N., & Ord, T. (2006). The re­ver­sal test: elimi­nat­ing sta­tus quo bias in ap­plied ethics. Ethics, 116(4), 656-679.

Bostrom, N. (2002). Ex­is­ten­tial risks: An­a­lyz­ing hu­man ex­tinc­tion sce­nar­ios and re­lated haz­ards. Jour­nal of Evolu­tion and Tech­nol­ogy, 9.

Bow­ers, K. S. (1973). Si­tu­a­tion­ism in psy­chol­ogy: An anal­y­sis and a cri­tique. Psy­cholog­i­cal re­view, 80(5), 307.

Bowler, S., & Karp, J. A. (2004). Poli­ti­ci­ans, scan­dals, and trust in gov­ern­ment. Poli­ti­cal Be­hav­ior, 26(3), 271-287.

Buck­els, E. E., Jones, D. N., & Paulhus, D. L. (2013). Be­hav­ioral con­fir­ma­tion of ev­ery­day sadism. Psy­cholog­i­cal sci­ence, 24(11), 2201-2209.

Burger, J. M. (2009). Repli­cat­ing Mil­gram: Would peo­ple still obey to­day?. Amer­i­can Psy­chol­o­gist, 64(1), 1.

Burkle, F. M. (2016). An­ti­so­cial per­son­al­ity di­s­or­der and patholog­i­cal nar­cis­sism in pro­longed con­flicts and wars of the 21st cen­tury. Disaster medicine and pub­lic health pre­pared­ness, 10(1), 118-128.

But­ton, K. S., Ioan­ni­dis, J. P., Mokrysz, C., Nosek, B. A., Flint, J., Robin­son, E. S., & Mu­nafò, M. R. (2013). Power failure: why small sam­ple size un­der­mines the re­li­a­bil­ity of neu­ro­science. Na­ture Re­views Neu­ro­science, 14(5), 365-376.

By­man, D. L., & Pol­lack, K. M. (2001). Let us now praise great men: Bring­ing the states­man back in. In­ter­na­tional Se­cu­rity, 25(4), 107-146.

Camp­bell, W. K., Goodie, A. S., & Foster, J. D. (2004). Nar­cis­sism, con­fi­dence, and risk at­ti­tude. Jour­nal of be­hav­ioral de­ci­sion mak­ing, 17(4), 297-311.

Camp­bell, W. K., Ru­dich, E. A., & Sedik­ides, C. (2002). Nar­cis­sism, self-es­teem, and the pos­i­tivity of self-views: Two por­traits of self-love. Per­son­al­ity and So­cial Psy­chol­ogy Bul­letin, 28(3), 358-368.

Chang, J., & Hal­li­day, J. (2007). Mao: The un­known story. Vin­tage.

Cis­lak, A., Ci­chocka, A., Wo­j­cik, A. D., & Frankowska, N. (2018). Power cor­rupts, but con­trol does not: What stands be­hind the effects of hold­ing high po­si­tions. Per­son­al­ity and So­cial Psy­chol­ogy Bul­letin, 44(6), 944-957.

Col­gan, J. D. (2013). Do­mes­tic rev­olu­tion­ary lead­ers and in­ter­na­tional con­flict. World Poli­tics, 65(4), 656-690.

Cooke, D. J., & Michie, C. (2001). Refin­ing the con­struct of psy­chopa­thy: Towards a hi­er­ar­chi­cal model. Psy­cholog­i­cal as­sess­ment, 13(2), 171.

Dal­lago, F., Cima, R., Roc­cato, M., Ri­colfi, L., & Miri­sola, A. (2008). The cor­re­la­tion be­tween right-wing au­thor­i­tar­i­anism and so­cial dom­i­nance ori­en­ta­tion: The mod­er­at­ing effects of poli­ti­cal and re­li­gious iden­tity. Ba­sic and ap­plied so­cial psy­chol­ogy, 30(4), 362-368.

Del­uga, R. J. (2001). Amer­i­can pres­i­den­tial Machi­avel­li­anism: Im­pli­ca­tions for charis­matic lead­er­ship and rated perfor­mance. The Lead­er­ship Quar­terly, 12(3), 339-363.

Dere­finko, K. J., & Ly­nam, D. R. (2006). Con­ver­gence and di­ver­gence among self-re­port psy­chopa­thy mea­sures: A per­son­al­ity-based ap­proach. Jour­nal of Per­son­al­ity Di­sor­ders, 20(3), 261-280.

Diköt­ter, F. (2011). Mao’s great famine: The his­tory of China’s most dev­as­tat­ing catas­tro­phe, 1958-1962. Blooms­bury Pub­lish­ing.

Diköt­ter, F. (2016). The Cul­tural Revolu­tion: A Peo­ple’s His­tory, 1962—1976. Blooms­bury Pub­lish­ing.

Diköt­ter, F. (2019). Dic­ta­tors: The Cult of Per­son­al­ity in the Twen­tieth Cen­tury. Blooms­bury Pub­lish­ing.

Dus­para, B., & Gre­it­e­meyer, T. (2017). The im­pact of dark tetrad traits on poli­ti­cal ori­en­ta­tion and ex­trem­ism: an anal­y­sis in the course of a pres­i­den­tial elec­tion. Heliyon, 3(10), e00425.

Endler, N. S., & Mag­nus­son, D. (1976). Toward an in­ter­ac­tional psy­chol­ogy of per­son­al­ity. Psy­cholog­i­cal bul­letin, 83(5), 956.

Fox, B., & DeLisi, M. (2019). Psy­cho­pathic kil­lers: a meta-an­a­lytic re­view of the psy­chopa­thy-homi­cide nexus. Ag­gres­sion and vi­o­lent be­hav­ior, 44, 67-79.

Gal­lagher, M. E., & Allen, S. H. (2014). Pres­i­den­tial per­son­al­ity: Not just a nui­sance. For­eign Policy Anal­y­sis, 10(1), 1-21.

Gao, Y., & Raine, A. (2010). Suc­cess­ful and un­suc­cess­ful psy­chopaths: A neu­ro­biolog­i­cal model. Be­hav­ioral sci­ences & the law, 28(2), 194-210.

Glad, B. (2002). Why tyrants go too far: Mal­ig­nant nar­cis­sism and ab­solute power. Poli­ti­cal Psy­chol­ogy, 23(1), 1-2.

Gluck, M., Hee­sacker, M., & Choi, H. D. (2020). How much of the dark triad is ac­counted for by sex­ism? Per­son­al­ity and In­di­vi­d­ual Differ­ences, 154, 109728.

Gøtzsche-Astrup, O. (2019). Par­ti­san­ship and vi­o­lent in­ten­tions in the United States.

Han­son, R. (2016). The Age of Em: Work, Love, and Life when Robots Rule the Earth. Oxford Univer­sity Press.

Hare, R. D. (2003). Man­ual for the Hare Psy­chopa­thy Check­list-Re­vised. Toronto, ON, Canada: Multi-Health Sys­tems.

Hare, R D., et al. (1990). “The re­vised psy­chopa­thy check­list: re­li­a­bil­ity and fac­tor struc­ture.” Psy­cholog­i­cal Assess­ment: A Jour­nal of Con­sult­ing and Clini­cal Psy­chol­ogy 2, no. 3: 338.

Har­ris, G. T., & Rice, M. E. (2006). Treat­ment of psy­chopa­thy. Hand­book of psy­chopa­thy, 555-572.

Haslam, N., Hol­land, E., & Kup­pens, P. (2012). Cat­e­gories ver­sus di­men­sions in per­son­al­ity and psy­chopathol­ogy: a quan­ti­ta­tive re­view of tax­o­met­ric re­search. Psy­cholog­i­cal medicine, 42(5), 903-920.

Her­sh­man, D. J., & Lieb, J. (1994). A broth­er­hood of tyrants: Manic de­pres­sion and ab­solute power. Amherst, NY: Prometheus

Hol­lan­der, P. (2016). From Ben­ito Mus­solini to Hugo Chavez. Cam­bridge Univer­sity Press.

Hol­lan­der, P. (2017). Poli­ti­cal pil­grims: Western in­tel­lec­tu­als in search of the good so­ciety. Rout­ledge.

Hosker-Field, A. M., Mol­nar, D. S., & Book, A. S. (2016). Psy­chopa­thy and risk tak­ing: Ex­am­in­ing the role of risk per­cep­tion. Per­son­al­ity and In­di­vi­d­ual Differ­ences, 91, 123-132.

Jain, K., & Bear­den, J. N. (2011). Machi­avel­li­anism and over­con­fi­dence.

John­son, A. M., Ver­non, P. A., & Feiler, A. R. (2008). Be­hav­ioral ge­netic stud­ies of per­son­al­ity: An in­tro­duc­tion and re­view of the re­sults of 50+ years of re­search. The SAGE hand­book of per­son­al­ity the­ory and as­sess­ment, 1, 145-173.

Jona­son, P. K., Duin­eveld, J. J., & Mid­dle­ton, J. P. (2015). Pathol­ogy, pseu­dopathol­ogy, and the Dark Triad of per­son­al­ity. Per­son­al­ity and In­di­vi­d­ual Differ­ences, 78, 43-47.

Jona­son, P. K., Slom­ski, S., & Par­tyka, J. (2012). The Dark Triad at work: How toxic em­ploy­ees get their way. Per­son­al­ity and in­di­vi­d­ual differ­ences, 52(3), 449-453.

Jones, B. F., & Olken, B. A. (2005). Do lead­ers mat­ter? Na­tional lead­er­ship and growth since World War II. The Quar­terly Jour­nal of Eco­nomics, 120(3), 835-864.

Jones, D. N., & Figueredo, A. J. (2013). The core of dark­ness: Un­cov­er­ing the heart of the Dark Triad. Euro­pean Jour­nal of Per­son­al­ity, 27(6), 521-531.

Jones, D. N. (2013). Psy­chopa­thy and machi­avel­li­anism pre­dict differ­ences in racially mo­ti­vated at­ti­tudes and their af­fili­a­tions. Jour­nal of Ap­plied So­cial Psy­chol­ogy, 43, E367–E378. doi:10.1111/​jasp.12035

Ka­jo­nius, P. J., Pers­son, B. N., & Jona­son, P. K. (2015). He­donism, achieve­ment, and power: Univer­sal val­ues that char­ac­ter­ize the dark triad. Per­son­al­ity and In­di­vi­d­ual Differ­ences, 77, 173-178.

Kline, P., & Cooper, C. (1984). A con­struct val­i­da­tion of the Ob­jec­tive-An­a­lytic Test Bat­tery (OATB). Per­son­al­ity and In­di­vi­d­ual Differ­ences, 5(3), 323-337.

Knopik, Nei­der­hiser, DeFries, Plomin (2018). Be­hav­ioral ge­net­ics. 7th edi­tion. Macmil­lan Learn­ing. New York.

Koehn, M. A., Jona­son, P. K., & Davis, M. D. (2019). A per­son-cen­tered view of prej­u­dice: The Big Five, Dark Triad, and prej­u­dice. Per­son­al­ity and In­di­vi­d­ual Differ­ences, 139, 313-316.

Lar­son, D. W. (1988) Prob­lems of Con­tent Anal­y­sis in For­eignPolicy Re­search. In­ter­na­tional Stud­ies Quar­terly 32 (2): 241–255.

Lee, K., Ash­ton, M. C., Wilt­shire, J., Bourdage, J. S., Visser, B. A., & Gal­lucci, A. (2013). Sex, power, and money: Pre­dic­tion from the Dark Triad and Hon­esty–Hu­mil­ity. Euro­pean Jour­nal of Per­son­al­ity, 27(2), 169-184.

Leven­son, M. R., Kiehl, K. A., & Fitz­patrick, C. M. (1995). Assess­ing psy­cho­pathic at­tributes in a non­in­sti­tu­tion­al­ized pop­u­la­tion. Jour­nal of per­son­al­ity and so­cial psy­chol­ogy, 68(1), 151.

Lilien­feld, S. O., & An­drews, B. P. (1996). Devel­op­ment and pre­limi­nary val­i­da­tion of a self-re­port mea­sure of psy­cho­pathic per­son­al­ity traits in a non­crim­i­nal pop­u­la­tion. Jour­nal of per­son­al­ity as­sess­ment, 66(3), 488-524.

Lilien­feld, S. O., Latz­man, R. D., Watts, A. L., Smith, S. F., & Dut­ton, K. (2014). Cor­re­lates of psy­cho­pathic per­son­al­ity traits in ev­ery­day life: Re­sults from a large com­mu­nity sur­vey. Fron­tiers in psy­chol­ogy, 5, 740.

Lilien­feld, S. O., Wald­man, I. D., Land­field, K., Watts, A. L., Ruben­zer, S., & Fasching­bauer, T. R. (2012). Fear­less dom­i­nance and the US pres­i­dency: Im­pli­ca­tions of psy­cho­pathic per­son­al­ity traits for suc­cess­ful and un­suc­cess­ful poli­ti­cal lead­er­ship. Jour­nal of per­son­al­ity and so­cial psy­chol­ogy, 103(3), 489.

Lilien­feld, S. O., Watts, A. L., & Smith, S. F. (2015). Suc­cess­ful psy­chopa­thy: A sci­en­tific sta­tus re­port. Cur­rent Direc­tions in Psy­cholog­i­cal Science, 24(4), 298-303.

MacAskill, W. (2020). Will MacAskill on the moral case against ever leav­ing the house, whether now is the hinge of his­tory, and the cul­ture of effec­tive al­tru­ism. 80000 Hours Pod­cast. Jan 24, 2020. https://​​80000hours.org/​​pod­cast/​​epi­sodes/​​will-macaskill-paral­y­sis-and-hinge-of-his­tory/​​

Ma­cenczak, L. A., Camp­bell, S., Hen­ley, A. B., & Camp­bell, W. K. (2016). Direct and in­ter­ac­tive effects of nar­cis­sism and power on over­con­fi­dence. Per­son­al­ity and In­di­vi­d­ual Differ­ences, 91, 113-122.

Mar­cus, D. K., Pres­zler, J., & Zei­gler-Hill, V. (2018). A net­work of dark per­son­al­ity traits: What lies at the heart of dark­ness?. Jour­nal of Re­search in Per­son­al­ity, 73, 56-62.

Matthews, G., Rein­er­man-Jones, L. E., Burke, C. S., Teo, G. W., & Scrib­ner, D. R. (2018). Na­tion­al­ism, per­son­al­ity, and de­ci­sion-mak­ing. Per­son­al­ity and In­di­vi­d­ual Differ­ences, 127, 89-100.

McDon­ald, M. M., Don­nel­lan, M. B., & Navar­rete, C. D. (2012). A life his­tory ap­proach to un­der­stand­ing the Dark Triad. Per­son­al­ity and In­di­vi­d­ual Differ­ences, 52(5), 601-605.

Međe­dović, J., & Kneže­vić, G. (2018). Dark and Pe­cu­liar. Jour­nal of In­di­vi­d­ual Differ­ences.

Mil­gram, S. (1963). Be­hav­ioral study of obe­di­ence. The Jour­nal of ab­nor­mal and so­cial psy­chol­ogy, 67(4), 371.

Mosha­gen, M., Hilbig, B. E., & Zet­tler, I. (2018). The dark core of per­son­al­ity. Psy­cholog­i­cal Re­view, 125(5), 656.

Mul­lins-Sweatt, S. N., Glover, N. G., Dere­finko, K. J., Miller, J. D., & Widi­ger, T. A. (2010). The search for the suc­cess­ful psy­chopath. Jour­nal of Re­search in Per­son­al­ity, 44(4), 554-558.

Nai, A., & Toros, E. (2020). The pe­cu­liar per­son­al­ity of strong­men: com­par­ing the Big Five and Dark Triad traits of au­to­crats and non-au­to­crats. Poli­ti­cal Re­search Ex­change, 2(1), 1-24.

Nai, A. (2019). Disagree­able nar­cis­sists, ex­tro­verted psy­chopaths, and elec­tions: A new dataset to mea­sure the per­son­al­ity of can­di­dates wor­ld­wide. Euro­pean Poli­ti­cal Science, 18(2), 309-334.

Nell, V. (2006). Cru­elty’s re­wards: The grat­ifi­ca­tions of per­pe­tra­tors and spec­ta­tors. Be­hav­ioral and Brain Sciences, 29(3), 211-224.

O’Con­nell, D., & Mar­cus, D. K. (2016). Psy­cho­pathic per­son­al­ity traits pre­dict pos­i­tive at­ti­tudes to­ward sex­u­ally preda­tory be­hav­iors in col­lege men and women. Per­son­al­ity and In­di­vi­d­ual Differ­ences, 94, 372-376.

Obradović, J., Par­dini, D. A., Long, J. D., & Loe­ber, R. (2007). Mea­sur­ing in­ter­per­sonal cal­lous­ness in boys from child­hood to ado­les­cence: An ex­am­i­na­tion of lon­gi­tu­di­nal in­var­i­ance and tem­po­ral sta­bil­ity. Jour­nal of Clini­cal Child and Ado­les­cent Psy­chol­ogy, 36(3), 276-292.

Paulhus, D. L., & Willi­ams, K. M. (2002). The dark triad of per­son­al­ity: Nar­cis­sism, Machi­avel­li­anism, and psy­chopa­thy. Jour­nal of re­search in per­son­al­ity, 36(6), 556-563.

Paulhus, D. L. (2014). Toward a tax­on­omy of dark per­son­al­ities. Cur­rent Direc­tions in Psy­cholog­i­cal Science, 23(6), 421-426.

Pe­rina, K., Hsu, S., Wash­ing­ton, C. (2020). Man­i­fold Pod­cast. https://​​man­i­foldlearn­ing.com/​​epi­sode-036/​​.

Plomin, R. (2019). Blueprint: How DNA makes us who we are. MIT Press.

Plouffe, R. A., Sak­lofske, D. H., & Smith, M. M. (2017). The as­sess­ment of sadis­tic per­son­al­ity: Pre­limi­nary psy­cho­me­t­ric ev­i­dence for a new mea­sure. Per­son­al­ity and in­di­vi­d­ual differ­ences, 104, 166-171.

Post, J. M. (2003). Assess­ing lead­ers at a dis­tance: The poli­ti­cal per­son­al­ity pro­file. The psy­cholog­i­cal as­sess­ment of poli­ti­cal lead­ers: With pro­files of Sad­dam Hus­sein and Bill Clin­ton, 69-104.

Pratto, F., Si­da­nius, J., Stal­l­worth, L. M., & Malle, B. F. (1994). So­cial dom­i­nance ori­en­ta­tion: A per­son­al­ity vari­able pre­dict­ing so­cial and poli­ti­cal at­ti­tudes. Jour­nal of per­son­al­ity and so­cial psy­chol­ogy, 67(4), 741.

Pre­ston, O. C., & Anestis, J. C. (2018). Psy­cho­pathic traits and poli­tics: Ex­am­in­ing af­fili­a­tion, sup­port of poli­ti­cal is­sues, and the role of em­pa­thy. Per­son­al­ity and in­di­vi­d­ual differ­ences, 131, 142-148.

Ron­son, J. (2012). The psy­chopath test: A jour­ney through the mad­ness in­dus­try. River­head Books.

Shul­man, C., & Bostrom, N. (2014). _Em­bryo Selec­tion for Cog­ni­tive En­hance­ment: Cu­ri­os­ity or Game-changer? Global Policy, 5(1), 85–92._doi:10.1111/​1758-5899.12123.

Southard, A. C., & Zei­gler-Hill, V. (2016). The dark triad traits and fame in­ter­est: Do dark per­son­al­ities de­sire star­dom?. Cur­rent Psy­chol­ogy, 35(2), 255-267.

Spar­row, R. (2013). In vitro eu­gen­ics. Jour­nal of Med­i­cal Ethics, 40(11), 725-731.

Tay­lor, S. (2019). Pathoc­racy. When peo­ple with per­son­al­ity di­s­or­ders gain power. Psy­chol­ogy To­day.

Toland, J. (1976). Adolf Hitler. New York: Dou­ble­day.

To­masik, B. (2013a). Char­ity Cost-Effec­tive­ness in an Uncer­tain World. Foun­da­tional Re­search In­sti­tute.

To­masik, B. (2013b). Differ­en­tial In­tel­lec­tual Progress as a Pos­i­tive-Sum Pro­ject. Foun­da­tional Re­search In­sti­tute.

Ullrich, S., Borke­nau, P., & Marneros, A. (2001). Per­son­al­ity di­s­or­ders in offen­ders: cat­e­gor­i­cal ver­sus di­men­sional ap­proaches. Jour­nal of Per­son­al­ity Di­sor­ders, 15(5), 442-449.

Ver­non, P. A., Villani, V. C., Vick­ers, L. C., & Har­ris, J. A. (2008). A be­hav­ioral ge­netic in­ves­ti­ga­tion of the Dark Triad and the Big 5. Per­son­al­ity and in­di­vi­d­ual Differ­ences, 44(2), 445-452.

Visser, B. A., Pozze­bon, J. A., & Reina-Ta­mayo, A. M. (2014). Sta­tus-driven risk tak­ing: Another “dark” per­son­al­ity?. Cana­dian Jour­nal of Be­havi­oural Science, 46(4), 485.

Vukaso­vić, T., & Bratko, D. (2015). Her­i­ta­bil­ity of per­son­al­ity: a meta-anal­y­sis of be­hav­ior ge­netic stud­ies. Psy­cholog­i­cal bul­letin, 141(4), 769.

Yer­gin, D. (1977). Shat­tered peace. Bos­ton: Houghton Mifflin Com­pany.


  1. Of course, as­sess­ing other peo­ple’s per­son­al­ity is always fraught with un­cer­tainty, es­pe­cially if they are long dead. ↩︎

  2. Other re­searchers have sug­gested similar con­structs aimed to rep­re­sent the com­mon core of “evil” (e.g., Book et al., 2015; Jones & Figueredo, 2013; Mar­cus et al., 2018). ↩︎

  3. Baron-Co­hen (2012) ar­gues that the defin­ing fea­ture of hu­man evil is “zero de­grees of em­pa­thy.” How­ever, some psy­chopaths can read other peo­ple ex­tremely well and would thus score highly on cer­tain items of the em­pa­thy ques­tion­naires Baron-Co­hen de­scribes in his book. Fur­ther­more, as Baron-Co­hen ac­knowl­edges, peo­ple on the autism spec­trum tend to have less em­pa­thy—at least cer­tain forms of it—but they are not more malev­olent than the pop­u­la­tion av­er­age. There­fore, re­duc­ing malev­olence to “zero de­grees of em­pa­thy” could be prob­le­matic or at least cru­cially de­pends on how we define and op­er­a­tional­ize em­pa­thy. ↩︎

  4. As of now, there is no es­tab­lished treat­ment of malev­olence. Har­ris and Rice (2006) re­view the em­piri­cal find­ings on the treat­ment of psy­chopa­thy but are quite pes­simistic about their effec­tive­ness. ↩︎

  5. A more rigor­ous anal­y­sis would be valuable, though it would also be method­olog­i­cally challeng­ing—as­sess­ing the per­son­al­ity traits of his­tor­i­cal figures, for ex­am­ple, is rather difficult. ↩︎

  6. One rea­son to be hes­i­tant here is that it seems plau­si­ble, for in­stance, that we, as well as jour­nal­ists and his­to­ri­ans, see more signs of malev­olent per­son­al­ity traits in lead­ers who have caused great harm, and will tend to over­look malev­olent per­son­al­ity traits in lead­ers who have done more good. ↩︎

  7. Again, we re­fer to Ap­pendix A for more de­tails. ↩︎

  8. For ex­am­ple, with­out Mao and Stalin the prob­a­bil­ity of a com­mu­nist China is smaller. A non-com­mu­nist China may have bet­ter re­la­tions with the U.S., and the prob­a­bil­ity of great power wars and (AI) arms races may be re­duced. How­ever, such claims are nec­es­sar­ily very spec­u­la­tive. For in­stance, one could also ar­gue that World War II may have in­creased longer-term sta­bil­ity by lead­ing to the for­ma­tion of the UN. ↩︎

  9. Or, as Robert Hare, one of the most well-known re­searchers of psy­chopa­thy, puts it: “Se­rial kil­ler psy­chopaths ruin fam­i­lies. Cor­po­rate, poli­ti­cal and re­li­gious psy­chopaths ruin economies [and] so­cieties.” (Ron­son, 2012, p. 117). ↩︎

  10. Gal­lagher & Allen (2014) found that U.S. pres­i­dents scor­ing higher on the Big Five facet “al­tru­ism” were less likely to em­ploy mil­i­tary force. ↩︎

  11. Also com­pare MacAskill: “I still en­dorse the view of push­ing re­sources into the fu­ture. The biggest caveat ac­tu­ally I’d have is about the rise of fas­cism and Stal­inism as the thing to push on [...] even though you might not think that a par­tic­u­lar ide­ol­ogy will last for­ever, well, if it lasts as long un­til you get like some eter­nal lock-in event, then it lasts for­ever. [...] I kind of think the rise of fas­cism and Stal­inism was a big­ger deal in the 20th cen­tury than the in­ven­tion of nu­clear weapons.” (MacAskill, 2020). ↩︎

  12. Dark Triad traits in poli­ti­cal can­di­dates cor­re­late with more nega­tive cam­paigns and fear ap­peals (Nai, 2019). ↩︎

  13. Since Dark Triad traits cor­re­late with so­cial dom­i­nance ori­en­ta­tion (SDO, Jones & Figueredo, 2013), malev­olent lead­ers will, on av­er­age, ex­hibit higher SDO and pre­fer poli­cies re­sult­ing in higher so­cial and eco­nomic in­equal­ity. ↩︎

  14. Az­i­zli et al. (2016) find that psy­chopa­thy and Machi­avel­li­anism are as­so­ci­ated with a greater propen­sity to lie and en­gage in high-stakes de­cep­tion. ↩︎

  15. Bowler & Karp (2004) find that scan­dals in­volv­ing poli­ti­ci­ans tend to lower poli­ti­cal trust. It seems plau­si­ble that malev­olent poli­ti­cal lead­ers are more likely to be in­volved in scan­dals. ↩︎

  16. Ac­cord­ing to Ulrich et al. (2001), the rate of an­ti­so­cial per­son­al­ity di­s­or­der among crim­i­nal offen­ders, 45% of whom were con­victed of “rob­bery or ex­tor­tion,” is more than 10 times higher than that of the con­trol sam­ple. Jona­son et al. (2012) also find that Machi­avel­li­anism, nar­cis­sism, and psy­chopa­thy all cor­re­late with the use of “hard tac­tics” in the work­place, in­clud­ing “threats of pun­ish­ment” (see Table 1, p. 451). ↩︎

  17. MacAskill (2020, em­pha­sis added): “[...] when you look at his­tory of well what are the worst catas­tro­phes ever? They fall into three main camps: pan­demics, war and to­tal­i­tar­i­anism. Also, to­tal­i­tar­i­anism or, well, au­toc­racy has been the de­fault mode for al­most ev­ery­one in his­tory. And I get quite wor­ried about that. So even if you don’t think that AI is go­ing to take over the world, well it still could be some in­di­vi­d­ual. And if it is a new growth mode, I do think that very sig­nifi­cantly in­creases the chance of lock-in tech­nol­ogy.” ↩︎

  18. Kaja Pe­rina on the Man­i­fold pod­cast (Pe­rina, 2020): “[M]ost of the stud­ies on psy­chopaths [...] are done on in­mates. For that rea­son, we’re forced to con­jec­ture about the re­ally suc­cess­ful ones be­cause I think the more suc­cess­ful, the more they evade de­tec­tion, per­haps, lifelong. So there is this dis­con­nect wherein a lot of them, the vi­o­lent ones, the less in­tel­li­gent ones, re­ally end up in jail, and these are the ones who are stud­ied, but these are not the ones who are highly Machi­avel­lian, nec­es­sar­ily, these are not the ones who are brilli­antly ma­nipu­la­tive. Th­ese are the ones who are com­mit­ting vi­o­lent crimes and get caught.” ↩︎

  19. Ex­ten­sive back­ground checks, for ex­am­ple with the help of pri­vate in­ves­ti­ga­tors, would be an­other promis­ing pos­si­bil­ity. In­tel­li­gence agen­cies do this already for some­what re­lated pur­poses. Gen­er­ally, the com­pet­i­tive na­ture of the poli­ti­cal pro­cess can of­ten un­cover past im­moral be­hav­ior—though sway­ing par­ti­san views seems to re­quire ev­i­den­tial strength that is difficult to achieve. Thanks to Mo­jmír Stehlík for rais­ing these points. ↩︎

  20. Yet an­other pos­si­bil­ity would be to use “ob­jec­tive” per­son­al­ity tests that don’t rely on self- or other-re­port but use ac­tual perfor­mance tests to eval­u­ate per­son­al­ity traits (with­out the test-taker know­ing which trait is sup­posed to be mea­sured). How­ever, ac­cord­ing to our cur­sory read­ing of the liter­a­ture, the few “ob­jec­tive” per­son­al­ity tests that ex­ist seem to have low val­idity (e.g., Kline & Cooper, 1984). ↩︎

  21. How­ever, one needs ex­am­ples to train such pre­dic­tors in the first place. One could start by look­ing for differ­ences in the brains of nor­mal peo­ple and, say, di­ag­nosed psy­chopaths, but this met­ric will be bi­ased to­wards di­ag­nosed psy­chopaths who are at least some­what un­rep­re­sen­ta­tive of non-di­ag­nosed malev­olent hu­mans (as ex­plained above). One needs to cor­rect for this as­cer­tain­ment bias. ↩︎

  22. Re­lat­edly, neu­ro­science re­search is of­ten un­der­pow­ered, re­sult­ing in low re­pro­ducibil­ity of the ac­cu­mu­lated find­ings (But­ton et al., 2013). ↩︎

  23. How­ever, the ed­u­ca­tion sys­tem also in­volves a lot of tests and grad­ing; and is at least some­what re­lated to ca­reer ad­vance­ment. Such tests are also com­mon for mil­i­tary en­try and some­times civil ser­vice. In the con­text of elec­tions, the key ques­tion is why vot­ers do not gen­er­ally de­mand such tests (in­clud­ing re­lated ob­jec­tive mea­sures, such as tax re­turns). ↩︎

  24. How­ever, there ex­ists the fright­en­ing pos­si­bil­ity that some vot­ers want their poli­ti­cal lead­ers to be at least mod­er­ately malev­olent. Most Rus­si­ans and Chi­nese, for ex­am­ple, seem to think highly of Stalin and Mao, re­spec­tively—though this is likely at least partly due to pro­pa­ganda. Gen­er­ally, many vot­ers seem to like “strong men” like Putin and over­look or even ap­pre­ci­ate ele­vated Dark Te­trad traits in their poli­ti­cal lead­ers. Also, ac­cord­ing to the (po­ten­tially bi­ased) as­sess­ment of “ex­perts”, poli­ti­ci­ans with au­to­cratic ten­den­cies—many of whom nonethe­less re­ceived the ma­jor­ity of votes—score sig­nifi­cantly higher on Dark Triad traits than the av­er­age poli­ti­cian (Nai & Toros, 2020). ↩︎

  25. It is also worth not­ing that in some forms of gov­ern­ment, such as al­lo­cat­ing poli­ti­cal po­si­tions to ran­domly se­lected in­di­vi­d­u­als or hered­i­tary monar­chy, those in po­si­tions in power are ex­actly as likely to be malev­olent as the pop­u­la­tion at large. This may be bet­ter than fierce com­pe­ti­tion for po­si­tions of power if the lat­ter ad­van­tages the most ruth­less and malev­olent in­di­vi­d­u­als. On the other hand, good se­lec­tion pro­ce­dures could also re­duce malev­olence in po­si­tions of power be­low the baseline; and of course this is only one con­sid­er­a­tion among many when eval­u­at­ing differ­ent forms of gov­ern­ment. ↩︎

  26. Thanks to Richard Ngo for mak­ing this point. ↩︎

  27. How­ever, ini­tial dis­tri­bu­tions may change due to com­pet­i­tive pres­sures or other fac­tors. Even if none of the first ems are malev­olent, there is no guaran­tee that malev­olence will re­main ab­sent in the long run. ↩︎

  28. Ex­treme events like se­vere abuse or vi­o­lence can make a huge differ­ence for the vic­tims, but such events are rel­a­tively rare and there­fore do not ex­plain much var­i­ance in the gen­eral pop­u­la­tion (Plomin, 2019). ↩︎

  29. Also note that shift­ing per­son­al­ity traits by more than this would likely be very difficult even if one wanted to do this. ↩︎

  30. Some dark traits, such as Machi­avel­li­anism, can be benefi­cial un­der cer­tain cir­cum­stances. It might be bet­ter if one could sin­gle out a dark trait, such as sadism, and only se­lect against it while leav­ing other dark traits un­changed. ↩︎

  31. How­ever, GWA stud­ies of per­son­al­ity are still fairly weak even at such scales. Even higher sam­ple sizes might be achieved by iden­ti­fy­ing proxy vari­ables of malev­olence, such as pub­lic records on crime. How­ever, this could eas­ily back­fire and cause great harm in nu­mer­ous ways, so one would have to be very care­ful. ↩︎

  32. Cur­rently, Dark Te­trad traits seem to be ne­glected in many rele­vant ar­eas. Most sperm banks mea­sure traits such as height, at­trac­tive­ness, phys­i­cal and men­tal health but not Dark Te­trad traits. Ser­vices that offer pre-im­ple­men­ta­tion di­ag­nos­tics screen for all sorts of ge­netic dis­eases and some even for IQ but not for Dark Te­trad traits. Test bat­ter­ies of enor­mous gov­ern­ment pro­jects like the UK Biobank mea­sure thou­sands of vari­ables—in­clud­ing phys­i­cal health, height, and preferred coffee and ce­real type—but they don’t mea­sure Dark Te­trad traits or even most per­son­al­ity traits in any sort of rigor­ous man­ner. ↩︎

  33. Gen­er­ally, re­duc­ing the in­fluence of malev­olent ac­tors can be done in myr­iad (of­ten mu­tu­ally re­in­forc­ing) ways. For in­stance, the more wide­spread be­lief sys­tems are that put great value on non-malev­olent traits such as com­pas­sion and al­tru­ism, the more par­ents might de­mand screen­ing for malev­olent traits, and the more fu­ture (gov­ern­ment) pro­jects will in­clude mea­sures of malev­olence in their test bat­ter­ies. ↩︎

  34. Adopt­ing cer­tain ide­olo­gies could also make one more malev­olent. How­ever, we think it’s plau­si­ble that most of the cor­re­la­tion is ex­plained by cau­sa­tion from malev­olent traits to dan­ger­ous ide­olo­gies, partly be­cause per­son­al­ity traits seem less amenable to change than be­liefs. ↩︎

  35. The Dark Triad also pre­dicts sex­ism (Gluck et al., 2020; O’Con­nell & Mar­cus, 2016), na­tion­al­ism (Matthews et al., 2018) as well as cog­ni­tive and af­fec­tive prej­u­dice (Koehn et al., 2019). Psy­cho­pathic traits pre­dict op­po­si­tion to­wards free speech and an­i­mal rights as well as sup­port for us­ing war as a tool for diplo­macy (Pre­ston & Anestis, 2018, Table 3). Machi­avel­li­anism and nar­cis­sism also seem to cor­re­late with over­con­fi­dence (Camp­bell et al., 2002; 2004; Ma­cenczak et al., 2016; Jain & Bear­den, 2011), and thus plau­si­bly nega­tively cor­re­late with epistemic hu­mil­ity which should serve a pro­tec­tive func­tion against all kinds of ex­trem­ism and fa­nat­i­cism. (Note that we don’t in­tend to con­vey that all these as­so­ci­a­tions are equally dan­ger­ous.) ↩︎

  36. Many of these and similar quotes could have been made solely for poli­ti­cal rea­sons, e.g., to strengthen al­li­ances. ↩︎