Abungu, Cecil; I adukunda, Ma ie Vic oi e; Beggs, Duncan Cass; Hassan, Aquila;
Sayidali, Raqda
Wo king Pape
Why Global Sou h coun ies need o ca e abou highly
capable AI
CIGI Pape s, No. 311
P o ided in Coope a ion wi h:
Cen e o In e na ional Go e nance Inno a ion (CIGI), Wa e loo, On a io
Sugges ed Ci a ion: Abungu, Cecil; I adukunda, Ma ie Vic oi e; Beggs, Duncan Cass; Hassan, Aquila;
Sayidali, Raqda (2024) : Why Global Sou h coun ies need o ca e abou highly capable AI, CIGI
Pape s, No. 311, Cen e o In e na ional Go e nance Inno a ion (CIGI), Wa e loo (On a io)
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/311790
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h ps://c ea i ecommons.o g/licenses/by/4.0/
CIGI Pape s No. 311 — Decembe 2024
Why Global Sou h Coun ies
Need o Ca e Abou Highly
Capable AI
Cecil Abungu, Ma ie Vic oi e I adukunda,
Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali
CIGI Pape s No. 311 — Decembe 2024
Why Global Sou h Coun ies
Need o Ca e Abou Highly
Capable AI
Cecil Abungu, Ma ie Vic oi e I adukunda,
Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali
Abou CIGI
The Cen e o In e na ional Go e nance Inno a ion (CIGI) is an independen ,
non-pa isan hink ank whose pee - e iewed esea ch and us ed analysis
in luence policy make s o inno a e. Ou global ne wo k o mul idisciplina y
esea che s and s a egic pa ne ships p o ide policy solu ions o he digi al
e a wi h one goal: o imp o e people’s li es e e ywhe e. Headqua e ed
in Wa e loo, Canada, CIGI has ecei ed suppo om he Go e nmen o
Canada, he Go e nmen o On a io and ounde Jim Balsillie.
À p opos du CIGI
Le Cen e pou l’inno a ion dans la gou e nance in e na ionale (CIGI) es un
g oupe de é lexion indépendan e non pa isan don les eche ches é aluées
pa des pai s e les analyses iables inci en les décideu s à inno e . G âce
à son éseau mondial de che cheu s plu idisciplinai es e de pa ena ia s
s a égiques, le CIGI o e des solu ions poli iques adap ées à l’è e numé ique
dans le seul bu d’amélio e la ie des gens du monde en ie . Le CIGI, don le
siège se ou e à Wa e loo, au Canada, béné icie du sou ien du gou e nemen
du Canada, du gou e nemen de l’On a io e de son onda eu , Jim Balsillie.
C edi s
Execu i e Di ec o , Global AI Risks Ini ia i e Duncan Cass‑Beggs
Senio Resea ch Associa e and P og am Manage Ma hew da Mo a
Publica ions Edi o Lynn Schellenbe g
Publica ions Edi o Susan Bubak
G aphic Designe Sami Chouhda y
Copy igh © 2024 by he Cen e o In e na ional Go e nance Inno a ion
The opinions exp essed in his publica ion a e hose o he au ho s and do no
necessa ily e lec he iews o he Cen e o In e na ional Go e nance Inno a ion
o i s Boa d o Di ec o s.
Fo publica ions enqui ies, please con ac [email p o ec ed].
The ex o his wo k is licensed unde CC BY 4.0. To iew a copy o his licence,
isi h p://c ea i ecommons.o g/licenses/by/4.0/.
Fo euse o dis ibu ion, please include his copy igh no ice. This wo k may
con ain con en (including bu no limi ed o g aphics, cha s and pho og aphs)
used o ep oduced unde licence o wi h pe mission om hi d pa ies.
Pe mission o ep oduce his con en mus be ob ained om hi d pa ies di ec ly.
Cen e o In e na ional Go e nance Inno a ion and CIGI a e egis e ed
adema ks.
67 E b S ee Wes
Wa e loo, ON, Canada N2L 6C2
www.cigionline.o g
Table o Con en s
i Abou he Au ho s
i Ac onyms and Abb e ia ions
1 Execu i e Summa y
1 In oduc ion
2 The Con ou s o Highly Capable AI
3 Why Global Sou h Coun ies Should Ca e Abou Highly Capable AI
11 Concluding Recommenda ions
13 Wo ks Ci ed
i CIGI Pape s No. 311 — Decembe 2024 • Cecil Abungu, Ma ie Vic oi e I adukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali
Abou he Au ho s
Cecil Abungu (lead au ho ) is a Ph.D. s uden
a he Uni e si y o Camb idge and a esea ch
a ilia e wi h he Cen e o he S udy o Exis en ial
Risk (Uni e si y o Camb idge) and he Ins i u e
o Law and AI. He is also he coo dina o o
he ILINA P og am. His esea ch in e es s lie
in a i icial in elligence (AI) sa e y go e nance,
algo i hmic disc imina ion, in ellec ual
his o y, legal heo y and cons i u ional law.
Ma ie Vic oi e I adukunda (lead au ho ) is a
mas e o laws s uden a Ha a d Law School
and a policy ellow wi h he Ha a d AI Sa e y
S uden Team. He esea ch in e es s lie in AI
go e nance, pa icula ly he legal du ies o AI
de elope s and deploye s, and he egula ion o AI
use by go e nmen s in A ica. She has p e iously
se ed as a esea ch ellow a he ILINA P og am.
Duncan Cass-Beggs is execu i e di ec o o he
Global AI Risks Ini ia i e a CIGI, ocusing on
de eloping inno a i e go e nance solu ions
o add ess cu en and u u e global issues
ela ing o AI. Duncan has mo e han 25 yea s o
expe ience wo king on domes ic and in e na ional
public policy issues, mos ecen ly as head
o s a egic o esigh a he O ganisa ion o
Economic Co-ope a ion and De elopmen .
Aquila Hassan is a p ojec managemen
specialis a he Cen e o he Go e nance o
AI in Ox o d, England. She has a backg ound
in enginee ing and an in e es in esea ch
a ound go e ning AI o he global majo i y.
Raqda Sayidali is an unde g adua e s uden a
S a hmo e Uni e si y and a esea che a he
ILINA P og am. He esea ch in e es s include
AI sa e y go e nance and he legal challenges
su ounding eme ging echnologies.
Ac onyms and
Abb e ia ions
4G ou h-gene a ion
5G i h-gene a ion
AGI a i icial gene al in elligence
AI a i icial in elligence
compu e compu a ional powe
ICRC In e na ional Commi ee
o he Red C oss
IDF Is ael De ense Fo ces
LAWS le hal au onomous weapons sys ems
LLMs la ge language models
RLAIF ein o cemen lea ning
om AI eedback
RLHF ein o cemen lea ning
om human eedback
UAVs unmanned ae ial ehicles
1Why Global Sou h Coun ies Need o Ca e Abou Highly Capable AI
Execu i e Summa y
By ma ching and su passing human cogni i e
abili ies, highly capable a i icial in elligence
(AI) — ad anced AI sys ems o he o eseeable
u u e, which leading AI companies a e wo king
owa d as pa o hei b oade goal o c ea e
a i icial gene al in elligence (AGI) — could be
one o he mos ans o ma i e echnologies
he wo ld has e e seen. While his adical
echnology is being buil p ima ily in Global
No h coun ies, i s impac s a e likely o be el
wo ldwide, and disp opo iona ely so in hose
Global Sou h coun ies wi h long-s anding
ulne abili ies — weak s a e ins i u ions;
dependence on labou -in ensi e, manu ac u ing-
based and expo -led economic models; egula ly
ecu ing a med con lic ; high us in echnology;
and mo e globally subo dina ed cul u es.
The au ho s o his pape conside six ways in
which highly capable AI could in e ac wi h hese
ulne abili ies. Fi s , highly capable AI could
lea e Global Sou h peoples acing a much ha she
economic eali y. Second, highly capable AI could
lead o a mo e damaging a med con lic in Global
Sou h coun ies. Nex , highly capable AI could
enable ep essi e and endu ing au ho i a ianism
in Global Sou h coun ies and could expose Global
Sou h peoples o unpa alleled manipula ion. In
addi ion o ha , highly capable AI could also
deepen he cul u al subo dina ion o Global Sou h
peoples. Finally, AI de elope s and esea che s
ha e ye o de ise a oolp oo way o ensu ing ha
he mos ad anced models always ake ac ions
in line wi h posi i e human alues. Unless his
p oblem is emedied be o e he eme gence o
highly capable AI, he e is a chance such AI could
lead o ca as ophic ou comes including signi ican
loss o li e, possibly up o human ex inc ion.
Because o he signi ican socie al impac s ha
highly capable AI could ha e, being conce ned abou
AI in a gene al way will no su ice. The au ho s
a gue ha all s akeholde s who ca e abou hose
who li e in Global Sou h coun ies mus pull on he
le e s a ailable o hem wi h he goal o in luencing
he ongoing de elopmen o highly capable AI.
In oduc ion
We li e in an e a o apid echnological inno a ion
and each in which he ools and so wa e being
buil can ha e a wo ldwide impac . His o ically,
miles ones such as he eleg aph (1837), he
elephone (1876) and adio (la e 1890s) eme ged
a in e als o se e al decades. Howe e , he pace
o echnological ad ancemen has accele a ed
signi ican ly. The pas 25 yea s ha e wi nessed a
succession o echnological leaps (Rose 2023),
including ou h- and i h-gene a ion (4G and
5G) mobile ne wo ks, inc easingly powe ul
sma phones and he In e ne o Things. Wi h
ew excep ions, he so wa e and de ices buil as
a esul o such b eak h oughs ha e ound hei
way o people in e e y co ne o he wo ld. As
an example, when Myanma e-en e ed global
li e in 2011 ollowing decades o isola ion unde
mili a y ule, less han one pe cen o i s popula ion
had access o he in e ne . By 2020, his s a is ic
had su ged o oughly 44 pe cen .1 Myanma
essen ially expe ienced a “leap og e ec ,” mo ing
om minimal in e ne access o widesp ead
sma phone and social media use wi hin a decade.
These as -b eaking and a - eaching de elopmen s
in echnology ha e he capaci y o ans o m
socie ies in unexpec ed ways, pa icula ly
when deployed in di e se cul u al con ex s.
In Myanma , he apid sp ead o he in e ne
had a ange o complex consequences, bo h
bene icial and de imen al. On he upside, i
p o ided access o in o ma ion (Phyo 2023) and
new economic oppo uni ies (Jø gensen 2019,
48–50). On he downside, i led o he expansion
o pla o ms ha we e hen used o sp ead ha e
speech and disin o ma ion, exace ba ing e hnic
ensions and iolence (Kla k and Sagun 2023).
AI sys ems a e expec ed o ha e an e en mo e
p o ound global impac . Expe s an icipa e ha
AI will cause d ama ic changes ac oss almos
e e y domain o li e, including human igh s,
knowledge, well-being and go e nance (Ande son
and Rainie 2023). As wi h p e ious echnological
ad ancemen s, AI is bound o a ec he Global
Sou h signi ican ly, especially because he mos
capable AI sys ems a e being de eloped in Wes e n
coun ies, whe e go e nance amewo ks a e
1 See Jø gensen (2019) and h ps://da a.wo ldbank.o g/indica o /IT.NET.
USER.ZS?loca ions=MM.
2CIGI Pape s No. 311 — Decembe 2024 • Cecil Abungu, Ma ie Vic oi e I adukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali
shaped and designed wi h Wes e n con ex s
and eali ies in mind (Ayana e al. 2024, 3).
The e a e mo e han 130 coun ies in he egion
known as he “Global Sou h,” each wi h a di e se
socie y. Howe e , Global Sou h coun ies2 also
sha e some unique eali ies, ulne abili ies and
challenges. Mos o hem a e aced wi h simila
economic eali ies, such as se e ely limi ed access
o capi al and in es men , high le els o po e y
and subs an ial income inequali y (Ayub 2013).
Addi ionally, hey sha e socio-poli ical challenges,
including ecu ing iolen con lic (Palik, Obe meie
and Rus ad 2022), weak s a e ins i u ions (Choi 2023,
5–7) and lowe le els o educa ed popula ions.3 Due
o he complexi y o hese issues and he eno mous
demands o po en ial solu ions, Global Sou h
coun ies ha e g appled wi h hese p oblems o
decades, and i is easonable o assume will do so o
decades o come. These challenges consume much
o he ocus and esou ces o Global Sou h coun ies,
making i di icul o hem o add ess po en ial
isks posed by eme ging echnologies such as AI.
None heless, in his pape , he au ho s a gue
ha Global Sou h coun ies mus gi e se ious
a en ion o he de elopmen o highly capable AI,
gi en he signi ican con ex -speci ic isks i will
pose in hei socie ies. This a gumen is ounded
on expe p edic ions ega ding he powe ul
capabili ies o highly capable AI sys ems, and he
au ho s’ analysis o how highly capable AI sys ems
a e likely o engage wi h he ongoing eali ies
and speci ic ulne abili ies o hese coun ies.
In he sec ion ollowing his in oduc ion, “highly
capable AI” is de ined by can assing echnical
p edic ions abou he capabili ies ha he mos
powe ul AI sys ems will ha e in he o eseeable
u u e. The hi d sec ion hen ou lines he impac s
such AI sys ems could ha e in Global Sou h
coun ies. In pa icula , he pape a gues ha
he de elopmen and use o highly capable AI
could mean ha Global Sou h peoples will ace
a much ha she economic eali y; expe ience
ex emely damaging a med con lic ; expe ience
ep essi e and endu ing au ho i a ianism; li e
h ough unpa alleled manipula ion; and ind
hemsel es li ing in a wo ld whe e hei cul u es
2 Th oughou his pape , he e m “Global Sou h coun ies” will no e e o
China, Hong Kong, Macau and Singapo e.
3 See Ri chie e al. (2023) and www.wo ldeconomics.com/Indica o ‑Da a/
ESG/Social/Mean‑Yea s‑o ‑Schooling/.
a e e y deeply subo dina ed. The ou h and inal
sec ion o his pape con ains a sho discussion
on he way o wa d and o ms he conclusion.
The e is a ai chance ha cu en ad anced AI
sys ems could, ealis ically, gi e ise o some
a ia ion o he isks discussed in his pape .
Howe e , his s udy will show ha hese isks
will be g ea e and a mo e conce ning wi h
u u e highly capable AI sys ems. The au ho s
hope ha he esea ch p esen ed he e will p omp
all s akeholde s conce ned abou Global Sou h
peoples o cla i y and sha pen hei posi ions
on he de elopmen o highly capable AI.
The Con ou s o Highly
Capable AI
In his pape , “highly capable AI” e e s o AI
sys ems ha demons a e cogni i e capabili ies,
enabling hem o pe o m economically aluable
asks a o abo e he le el o human beings.
Leading AI companies such as Google DeepMind,
Me a and OpenAI a e wo king owa d building
hese ad anced AI sys ems as pa o hei
b oade goal o c ea e AGI.4 To achie e his goal,
hey a e in es ing hea ily in acqui ing da a se s,
compu e chips and da a cen es (Mu gia 2023;
Bengio e al. 2024, 843; Ga dizy and E a i 2024).
Cu en ad anced AI sys ems al eady exhibi
imp essi e capabili ies, including wha appea s
o be common-sense easoning, cause and e ec
easoning, s ep-by-s ep easoning and in-con ex
lea ning (P i i e a e al. 2024, 21). They ha e also
demons a ed imp o ed pe o mance on asks
such as image gene a ion and ecogni ion, ideo
gene a ion, and language-based asks such as
ex gene a ion, and in ields such as coding,
ma hema ics and biology (Ngo 2023; Bubeck e al.
2023, chap e 2). These imp essi e capabili ies a e
gene ally a ibu ed o deep lea ning (Ka no sky
2016; Pipe 2020; Ngo, Chan and Minde mann
2024, 1), and he scaling o model sizes, da a se s
and compu a ion powe (o , compu e) used in
aining (Kaplan e al. 2020, 3; Ha is, Ha is and
4 See Mu gia (2023); Hea h (2024); Pa el (2023); h ps://openai.com/
cha e .
9Why Global Sou h Coun ies Need o Ca e Abou Highly Capable AI
o his pape o esee highly capable AI as likely
o be e en mo e able o manipula e use s.
One key enable will be highly capable AI’s
inc eased access o con ex and mo e pe sonal
in o ma ion o use s (Gab iel e al. 2024, 27).
Resea ch in pe sona-based app oaches o
con e sa ional AI shows ha pe sonal in o ma ion
helps he sys em unde s and he con ex o
con e sa ion be e , making hese p og ams
in e ac mo e seamlessly and engage in mo e
human dialogue (Liu, Symons and Va sa ai 2022).
These capabili ies will be scaled up conside ably
in highly capable AI, making i exceedingly ha d
o o dina y human use s o know whe he he
AI sys em hey a e using is manipula ing hem.
The o he enable will be he abili ies ha highly
capable AI will ha e o eason, plan and espond
o mul imodal commands a an in e ence speed
ha makes human-machine in e ac ions seem
mo e na u al. This would inc ease he likelihood
o an h opomo phism, some hing ha spu s us
and allows manipula ion (Gab iel e al. 2024, 96,
102). Taken oge he , hese ea u es will c ea e AI
sys ems ha o e an imme si e expe ience and
c ea e he illusion o us by making use s eel like
hey a e in e ac ing wi h a iend o a con idan
(ibid., 112). Use s in such si ua ions will he e o e be
mo e likely o ollow any sugges ions o di ec i es
ha he AI sys em in ques ion p oposes. The e
a e al eady signs ha such a wo ld is no a o .
Conside , o example, Google’s Gemini,6 which
Google says seeks o c ea e a mo e imme si e
use expe ience h ough agen s ha can “see and
hea wha we do, be e unde s and he con ex
we’ e in and espond quickly in con e sa ion,
making he pace and quali y o in e ac ions
eel mo e na u al” (ci ed in Heikkilä 2024).
S udies ha e ound ha ci izens o de eloping
coun ies a e signi ican ly mo e op imis ic and
us ing abou he impac ha AI will ha e
on hei li es (Ipsos 2022). In a supplemen al
epo o i s 2024 us ba ome e , he Edelman
T us Ins i u e (2024, 3–7) epo ed ha us in
echnology is highe in de eloping coun ies,
while i has de e io a ed in he Uni ed S a es and
he Uni ed Kingdom. Ano he global s udy ound
ha people in B azil, China, India and Sou h A ica
showed highe le els o us in AI compa ed
o people in Global No h coun ies (Gillespie
6 Gemini is a “ amily” o highly capable mul imodal models whose quali y
inc eases wi h model size (Gemini Team 2023).
e al. 2023, 5). The s udy pa icula ly no ed ha
people in hese ou coun ies us in he abili y,
humani y and in eg i y o AI sys ems (ibid., 18).
The us AI enjoys in Global Sou h coun ies
seems o be a esul o i s pe cei ed bene i s.
E en in cases whe e pa icipan s a e ea ul abou
some isks, hei exci emen , op imism and us
appea o be undiminished (ibid., 24). In India,
o ins ance, echnology in gene al is seen as a
solu ion o many de elopmen al issues and is
o en us ed as an “au ho i y” in many si ua ions
(Kapania e al. 2022). Simila ly, in B azil he le el
o us in AI is signi ican ly highe han he le el
o unde s anding he e is abou i (Gillespie e al.
2023, 56). This si ua ion is likely o be eplica ed
in many o he Global Sou h coun ies.
Apa om le els o us in echnology, Global
Sou h peoples also ha e lowe le els o educa ion
in compa ison o he es o he wo ld. Fo ins ance,
in 2020 he adul li e acy a e in Sub-Saha an A ica
was only a 66 pe cen while he global a e age was
89 pe cen .7 S udies on misin o ma ion suscep ibili y
ha e shown ha educa ion is a key ac o in
de e mining how much someone is suscep ible o
belie ing alse in o ma ion. Fo example, one s udy
iden i ied an in e se s a is ical co ela ion be ween
highe le els o educa ion and belie in scien i ic
misin o ma ion (Siani and G een 2023, 8). The same
could la gely be ue o manipula ion as well, as
educa ion is impo an in de eloping he abili y
o de ec and e lec on nuances ac oss judgmen
domains (Knuu ila, Neude and Howa d 2022).
Highly Capable AI Could
Deepen he Subo dina ion o
Global Sou h Peoples’ Cul u es
Th ough di e en p ojec s and app oaches, ields
such as decolonial s udies ha e shown us ha
cul u al hegemonies can be mapped. In o he
wo ds, he e a e iden i iable pa e ns ha show
he wo ld iew dominance o mo e his o ically
powe ul and p i ileged g oups o people o e
less his o ically powe ul and p i ileged g oups o
people (Co es-Rami ez 2015, 117). This eali y has
knowingly and unknowingly been b ough o bea
h ough coe cion o pe suasion, o en acili a ed
by echnological de ices o so wa e. Fo example,
in he book Ame ica, as Seen on TV: How Tele ision
7 See h ps:// oyala icansa a is.com/ ounda ion/adul ‑educa ion‑in‑
olpalagilagi/.
10 CIGI Pape s No. 311 — Decembe 2024 • Cecil Abungu, Ma ie Vic oi e I adukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali
Shapes Immig an Expec a ions A ound he Globe,
sociologis Cla a Rod íguez shows how Ame ican
ele ision has been able o p ojec US-cen ic iews
on social ela ionships on o he es o he wo ld
(S oelke 2018). As his pape ’s au ho s a gue in he
sec ions ha ollow, he special capabili ies ha
highly capable AI will possess could lead o a much
mo e deeply en enched hegemony o Wes e n
cul u es o e Global Sou h peoples’ cul u es.
Se e al esea che s ha e al eady exp essed
conce ns abou how cu en s a e-o - he-a
gene a i e AI models can exace ba e he exis ing
dominance o Wes e n cul u es. Acco ding o
some o his esea ch, he models’ explici and
implici s ee ing in a ou o Wes e n cul u es is
mos ly a esul o how hey we e ained. Conside ,
o ins ance, some examples om LLMs, he
mos cu ing-edge AI sys ems a ailable oday.
LLMs a e ained on da a sc aped om he
in e ne , which o e - ep esen s some pa s o
he wo ld. Acco dingly, he language ha is
ep esen ed enhances cul u al alignmen whe e
he language in ques ion is p e alen . I is no
su p ise ha some s udies ha e also shown
ha LLMs’ esponses o cogni i e psychological
asks mos closely esemble hose o people
om Wes e n, educa ed, indus ialized, ich and
democ a ic socie ies. Indeed, Rohin Man i e al.
(2024, 1) ha e ecen ly un eiled esea ch ha
shows ha , due o hei aining co po a, LLMs
a e “clea ly biased agains loca ions wi h lowe
socio-economic condi ions (e.g., mos o A ica).”
Fu he mo e, he mos ad anced AI alignmen
me hods — ein o cemen lea ning om human
eedback (RLHF) and ein o cemen lea ning om
AI eedback (RLAIF) — p oduce AI sys ems ha
a e wedded o Wes e n cul u es. RLHF in ol es
human judgmen in c ea ing p e e ences in model
beha iou , while RLAIF, also called “cons i u ional
AI,” depends on a human-p oduced se o p inciples
o “cons i u ion” (Coni ze e al. 2024, 1). This
means ha bo h me hods ca y he biases inhe en
in humans as pe hei cul u al backg ounds
(Va shney 2024, 10). Fo ins ance, a s udy examining
he cul u al biases o LLMs ound ha he s a k
di e ences — a mo e owa d a mo e secula
esponses, o example — be ween he cul u al
alues obse able in Open AI’s GPT-3 model and
hose obse able in i s immedia e successo
(GPT-3.5 Tu bo) can be a ibu ed o he use o
RLHF in aining GPT-3.5 Tu bo (Tao e al. 2024, 3).
Unless he e is signi ican change, he a i al o
highly capable AI could wo sen he si ua ion o
Global Sou h peoples’ cul u es. I is likely ha
jus as wi h exis ing cu ing-edge AI models,
highly capable AI will mos ly be ained on da a
se s ha ha e Wes e n cul u al biases and no ms
embedded in o hem. I RLHF and RLAIF emain
he bes -pe o ming “alignmen ” me hods a ailable
o leading AI de elope s, we can imagine ha
bo h will con inue o p op up Wes e n cul u es
(as explained in he p eceding sec ion). And, as
many esea che s ha e sugges ed, highly capable
AI could be used e e ywhe e — and o do all so s
o hings — in he u u e. Because o he accu acy
and e iciency bene i s hese AI sys ems could
p esen , businesses and go e nmen s will sc amble
o in eg a e hem in o hei wo k in a way ha
hey ha e no ye done wi h exis ing AI sys ems.
In his e en , Global Sou h peoples migh be
o ced o li e wi h AI sys ems ha undamen ally
ca y Wes e n cul u al no ms and biases.
The cul u al biases and no ms ha will be
embedded in o highly capable AI sys ems could
be a odds wi h he cul u es o Global Sou h use s,
leading o misin e p e a ion and mis ep esen a ion
o ce ain cul u es, de elopmen s ha could in
u n c ea e cul u al ba ie s, impose a cul u al
hegemony and possibly e en esul in cul u al
e asu e (P abhaka an, Qad i and Hu chinson
2022, 2). On op o ha , by ep esen ing ce ain
cul u al alues as no ms h ough s e eo yped
esponses, AI sys ems can u he en ench
hese alues in use s who will hen hold hem
as hei own (Anwa e al. 2024, 80–81).
Al hough some may sugges ha ine- uning
hese AI sys ems on mo e localized da a could
c ea e mo e “local” models, ha would s ill do
li le o al e he undamen al philosophical
pilla s upon which he sys ems ha e been buil .
In o he wo ds, i a model has been ained o
iden i y a ce ain esponse as w ong o bad, i
is exceedingly ha d o change ha . O he s may
sugges ha he ac ha AI could be used o
p o ec languages ha a e a isk o disappea ance
shows i migh no lead o he subo dina ion o
Global Sou h peoples’ cul u es (Onome 2024).
Howe e , his iew con la es he exis ence o
aspec s o a cul u e wi h i s lack o subo dina ion.
The wo a e no necessa ily in e changeable.
11Why Global Sou h Coun ies Need o Ca e Abou Highly Capable AI
Highly Capable AI
Could Inad e en ly
Resul in Mass Dea h
Highly capable AI could also esul in ca as ophic
consequences, such as he dea h o e y
many humans (Bengio e al. 2024, 843), and
AI misalignmen is likely o be he cause. As
discussed in he sec ion “The Con ou s o Highly
Capable AI,” highly capable AI sys ems will be
goal-d i en. Tha means ha such sys ems could
lea n o pu sue bo h desi able and undesi able
goals, wi h he la e being goals no aligned wi h
he widely sha ed human alues (Ngo, Chan and
Minde mann 2024, 3). In he main, he pu sui o
hese undesi able goals will occu whe e highly
capable AI sys ems ha e ewa d unc ions ha a e
no pe ec ly aligned wi h human p e e ences (ibid.).
Tha possibili y is eal because he mos ad anced
AI alignmen me hods s ill all sho in wo ways.
Fi s , speci ying human in en ions using ha d-coded
ewa d unc ions emains e y ha d and he models
may s ill end up pu suing undesi able goals by
exploi ing some ewa d misspeci ica ion. Second,
al hough using me hods like RLHF migh esol e
some o hese mis akes, eedback om human
e alua o s is some imes un eliable, as human
e alua o s migh unin en ionally gi e app o al
o high ewa d o undesi able beha iou (ibid.),
which could ein o ce undesi ed beha iou s in he
AI. Simila ly, o he AI alignmen me hods being
de eloped by echnical esea che s s ill seem a
s ep behind he challenge (Aschenb enne 2023).
Some signs o he isk we ace a e cap u ed in he
small-scale misbeha iou ha can be obse ed
in, o example, an AI sys em ha wins a game by
exploi ing some gli ches a he han by playing he
game well (Pipe 2020; K ako na e al. 2020). I such
a possibili y pe sis s, he e is a ai chance ha he
misalignmen o highly capable AI sys ems will
esul in mo e la ge-scale consequences (Ngo, Chan
and Minde mann 2024, 8; O d 2020, 144). These AI
sys ems may pu sue ins umen al goals including
a oiding shu down, going a ound human a emp s
o al e hei ewa d unc ions, esou ce acqui ing
and, e en ually, powe seeking (Vold and Ha is
2021, 735; Ngo, Chan and Minde mann 2024,8; O d
2020, 145). Ul ima ely, such AI sys ems may acqui e
bo h he mo i a ion and abili y o pu sue goals
ha a e incompa ible wi h human well-being such
as al e ing he Ea h’s en i onmen o acili a e
compu ing speed o ac i ely disempowe ing
humani y o p e en in e e ence wi h AI goals.
This could e en ually lead o ca as ophic
consequences o humani y, including mass dea h.
Mass casual y e en s anywhe e a e inhe en ly bad,
bu a e pa icula ly unjus when caused in Global
Sou h coun ies by echnologies de eloped and
p ima ily bene i ing Global No h coun ies. In he
wo s -case scena io, misaligned highly capable AI
sys ems could cause human ex inc ion, leading
no only o he loss o cu en li es bu also o all
po en ial u u e gene a ions. Such an ou come
would be s ongly a odds wi h he alue o ca e o
u u e gene a ions ound in he adi ions o many
e hnic communi ies in Global Sou h coun ies.8
Concluding
Recommenda ions
The p eceding sec ion o his pape laid ou he
majo con ex -speci ic isks ha Global Sou h
peoples could ace when highly capable AI is
e en ually de eloped and used. Un o una ely,
any domes ically ocused e o s hese coun ies
ake o coun e he isks ou lined he ein may
ha e limi ed use ulness. While egula ing he use
o s andalone physical ools and so wa e migh
be possible, i highly capable AI is in he o m o
so wa e accessible o e he in e ne (as cu en
LLMs a e), any a emp s o block access could
be ci cum en ed ia i ual p i a e ne wo ks.
As e idence om China shows, e en he mos
sophis ica ed go e nmen -imposed in e ne i ewall
can s ill be ine ec i e a comple ely elimina ing
access o he ele an in e ne si es (Williams 2023).
Global Sou h go e nmen s could also a emp
o h o le in e ence bu i would be ha d o
hem o know which da a cen es o a ge , and
he ele an da a cen es migh be ou side hei
ju isdic ions (Lehdon i a, Wú and Hawkins 2024).
Global Sou h peoples should be u he conce ned
because hei go e nmen s a e cu en ly ill-
equipped o police he use o AI wi hin hei bo de s.
To he au ho s’ knowledge, a ime o w i ing, no
Global Sou h egula o y body has ye conduc ed
an in es iga ion in o he use o AI in hei socie y;
8 Fo one example o communi ies in sou he n A ica, see Isaac Schape a
(1955, 195–97).
12 CIGI Pape s No. 311 — Decembe 2024 • Cecil Abungu, Ma ie Vic oi e I adukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali
his si ua ion is unlikely o change any ime soon
because in hese coun ies AI expe ise and inancial
esou ces a e in sho supply, while o he p io i ies
a e endless. This lack o o e sigh could mean ha
any in e en ions migh a i e oo la e in he day.
Fo hese easons, i is c i ical ha Global Sou h
coun ies engage now, while highly capable AI is
s ill being de eloped. I would make a signi ican
di e ence i hese coun ies iden i ied and
pulled on he le e s a ailable o hem in o de
o in luence he u u e de elopmen o highly
capable AI sys ems and hei po en ial implica ions.
This could be achie ed by, o example:
→ Appoin ing expe s udy g oups whose ocus
is no jus AI in gene al, bu highly capable AI
speci ically. These s udy g oups could gi e Global
Sou h coun ies ad ice on wha speci ic s eps o
ake wi h ega d o highly capable AI.
→ Ensu ing ha hei na ional and egional policies
(and e en ually laws) pay special a en ion o
p e en ing and mi iga ing he isks ha highly
capable AI could pose o hei socie ies.
→ Fo ming coali ions o like-minded na ions
and pushing he leading AI companies and
Wes e n coun ies o pay special a en ion o
he global scale o impac s o highly capable
AI as well as he con ex -speci ic implica ions
o Global Sou h coun ies. This wo k could be
done h ough in e na ional o a o s a e- o-s a e
diplomacy.
Ou side o go e nmen , esea che s and ci il
socie y in Global Sou h coun ies can make a
signi ican di e ence i hey begin engaging as well.
Thei oices and in luence could p essu e hei
go e nmen s, p essu e Wes e n go e nmen s, o
pe suade esea che s and ci il socie y o ganiza ions
in Wes e n coun ies o p ess hei go e nmen s o
ask he leading AI companies mo e ques ions. Those
who ca e abou he a e o Global Sou h peoples
mus ealize ha ge ing se ious abou AI gene ally
is no enough. I is c ucial ha we speci ically
ge se ious abou he isks o highly capable AI.
Acknowledgemen s
The au ho s would like o specially hank Cullen
O’Kee e, Fai h Gakii, Lynn Schellenbe g, Ma hew
da Mo a, Seán Ó hÉigea aigh, Sienka Dounia,
Sumaya Nu and Vicky Mi i i o hei eedback and
assis ance in he p ocess o w i ing his pape .
13Why Global Sou h Coun ies Need o Ca e Abou Highly Capable AI
Wo ks Ci ed
Allen, G eg and Taniel Chan. 2017. “A i icial In elligence and
Na ional Secu i y.” July. Bel e Cen e S udy. Camb idge,
MA: Bel e Cen e o Science and In e na ional A ai s,
Ha a d Kennedy School. www.bel e cen e .o g/
publica ion/a i icial-in elligence-and-na ional-secu i y.
Allen, Ka en. 2021. “D ones and Violen Nons a e Ac o s
in A ica.” A ica Cen e o S a egic S udies,
Augus 6. h ps://a icacen e .o g/spo ligh /
d ones-and- iolen -nons a e-ac o s-in-a ica/.
Allen, Na haniel and Ma ian “I y” Okpali. 2022. “A i icial
in elligence c eeps on o he A ican ba le ield.” B ookings
Ins i u ion, Feb ua y 2. www.b ookings.edu/a icles/
a i icial-in elligence-c eeps-on- o- he-a ican-ba le ield/.
Amnes y In e na ional. 2024. The S a e o he Wo ld’s Human Righ s:
Ap il 2024. London, UK: Amnes y In e na ional.
www.amnes y.o g/en/documen s/pol10/7200/2024/en.
Ande son, Janna and Lee Rainie. 2023. As AI Sp eads, Expe s
P edic he Bes and Wo s Changes in Digi al Li e by 2035.
Pew Resea ch Cen e , June 21. www.pew esea ch.o g/
in e ne /2023/06/21/as-ai-sp eads-expe s-p edic -
he-bes -and-wo s -changes-in-digi al-li e-by-2035/.
Anwa , Usman, Abulhai Sapa o , Ja ie Rando, Daniel Paleka,
Miles Tu pin, Pe e Hase, Ekdeep Singh Lubana e al. 2024.
“Founda ional Challenges in Assu ing Alignmen and Sa e y
o La ge Language Models.” P ep in , a Xi , Sep embe 6.
h ps://doi.o g/10.48550/a Xi .2404.09932.
A ms ong, Ad ienne. 1981. “The Poli ical Consequences o
Economic Dependence.” Jou nal o Con lic Resolu ion
25 (3): 401–28. www.js o .o g/s able/173661.
A iola, Leona do R., Lise Rakne and Nicolas an de Walle. 2023.
Democ a ic Backsliding in A ica? Au oc a iza ion, Resilience,
and Con en ion. Ox o d, UK: Ox o d Uni e si y P ess.
Aschenb enne , Leopold. 2023. “Nobody’s on he ball on AGI
alignmen .” Fo Ou P ospe i y (blog), Ma ch 29.
www. o ou pos e i y.com/nobodys-on-
he-ball-on-agi-alignmen /.
Ayana, Gelan, Kokeb Dese, Hundessa Daba Nemomssa,
Bon u Hab amu, B uce Mellado, Kingsley Badu, Edmund
Yamba e al. 2024. “Decolonizing global AI go e nance:
assessmen o he s a e o decolonized AI go e nance in
Sub-Saha an A ica.” Royal Socie y Open Science 11(8):
231994. h ps://doi.o g/10.1098/ sos.231994.
Ay es, Ian and Jack M. Balkin. 2024. “The Law o AI is he Law
o Risky Agen s wi hou In en ions.” P ep in , Social Science
Resea ch Ne wo k, June 12. h ps://pape s.ss n.com/
sol3/pape s.c m?abs ac _id=4862025.
Ayub, Mahmood. 2013. “Po e y and Inequali y.” Global
Jou nal o Eme ging Ma ke Economies 5 (3): 329–46.
h ps://doi.o g/10.1177/0974910113505796.
Beaumon , Hila y. 2022. “‘Ne e sleeps, ne e e en blinks’: he
hi- ech Andu il owe s sp eading along he US bo de s.”
The Gua dian, Sep embe 16. www. hegua dian.com/
us-news/2022/sep/16/andu il- owe s-
su eillance-us-mexico-bo de -mig an s.
Bell, S ephanie A. and An on Ko inek. 2023. “AI’s Economic Pe il.”
Jou nal o Democ acy 34 (4): 151–61.
www.jou nalo democ acy.o g/a icles/ais-economic-pe il/.
Bengio, Yoshua. 2023. “AI and Ca as ophic Risk.” Jou nal o
Democ acy 34 (4):111–21. www.jou nalo democ acy.o g/
a icles/ai-and-ca as ophic- isk/.
Bengio, Yoshua, Geo ey E. Hin on, And ew Yao, Dawn Song, Pie e
Abbeel, T e o Da ell, Yu al N. Ha a i e al. 2024. “Managing
ex eme AI isks amid apid p og ess.” Science 384 (6698):
842–45. h ps://doi.o g/10.1126/science.adn0117.
B anwen, Gwe n. 2022. “The Scaling Hypo hesis.” Gwe n.ne ,
Janua y 2. h ps://gwe n.ne /scaling-hypo hesis.
B undage, Miles, Shaha A in, Jack Cla k, Helen Tone , Pe e
Ecke sley, Ben Ga inkel, Allan Da oe e al. 2018. The
Malicious Use o A i icial In elligence: Fo ecas ing,
P e en ion, and Mi iga ion. P ep in , a Xi , Feb ua y 20.
h ps://doi.o g/10.48550/a Xi .1802.07228.
Bubeck, Sébas ien, Va un Chand aseka an, Ronen Eldan, Johannes
Geh ke, E ic Ho i z, Ece Kama , Pe e Lee e al. 2023. “Spa ks
o A i icial Gene al In elligence: Ea ly expe imen s wi h GPT-4.”
P ep in , a Xi , Ap il 13. h ps://a xi .o g/abs/2303.12712.
Campbell-James, Jona han. 2024. “Cu bing Ou side In e en ion
in he Sudan Wa .” Policy analysis, Ap il 4. PolicyWa ch 3855.
Washing on, DC: The Washing on Ins i u e o Nea
Eas Policy. www.washing onins i u e.o g/policy-
analysis/cu bing-ou side-in e en ion-sudan-wa .
Capa ini, Ma ina and Ian Da is. 2022. “A med con lic and
peace p ocesses in he Ame icas.” In SIPRI Yea book
2022: A mamen s, Disa mamen and In e na ional
Secu i y, edi ed by S ockholm In e na ional Peace
Resea ch Ins i u e, 75–102. Ox o d, UK: Ox o d
Uni e si y P ess. www.sip i.o g/yea book/2022/03.
14 CIGI Pape s No. 311 — Decembe 2024 • Cecil Abungu, Ma ie Vic oi e I adukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali
Ca lsmi h, Joseph. 2024. “Is Powe -Seeking AI an Exis en ial Risk?”
P ep in , a Xi , Augus 13. h ps://doi.o g/10.48550/
a Xi .2206.13353.
Ca oll, Micah, Alan Chan, Hen y Ash on and Da id K uege . 2023.
“Cha ac e izing Manipula ion om AI Sys ems.” P ep in , a Xi ,
Oc obe 30. h ps://doi.o g/10.48550/a Xi .2303.09387.
Casals, Ma celo. 2024. “Cold Wa Dic a o ships in he Sou he n
Cone (B azil, A gen ina, U uguay, Chile).” Ox o d
Bibliog aphies, June 21. www.ox o dbibliog aphies.com/
display/documen /obo-9780199791279/
obo-9780199791279-0254.xml.
Cazzaniga, Mau a, Flo ence Jaumo e, Longji Li, Gio anni
Melina, Augus us J. Pan on, Ca lo Pizzinelli, Emma
Rockall and Ma ina M. Ta a es. 2024. “Gen-AI: A i icial
In elligence and he Fu u e o Wo k.” IMF S a Discussion
No e SDN/2024/001. Washing on, DC: In e na ional
Mone a y Fund. www.im .o g/en/Publica ions/
S a -Discussion-No es/Issues/2024/01/14/Gen-AI-
A i icial-In elligence-and- he-Fu u e-o -Wo k-542379.
Cen e o P e en i e Ac ion. 2024. “Con lic in he Democ a ic
Republic o Congo.” Council on Fo eign Rela ions,
June 20. www.c .o g/global-con lic - acke /
con lic / iolence-democ a ic- epublic-congo.
Chan, Alan, Rebecca Salganik, Al a Ma kelius, Ch is Pang,
Ni a shan Rajkuma , Dmi ii K asheninniko , Lau o
Langosco e al. 2023. “Ha ms om Inc easingly
Agen ic Algo i hmic Sys ems.” P ep in , a Xi , May 12.
h ps://doi.o g/10.48550/a Xi .2302.10329.
Choi, An hony. 2023. “The Global No h and Global Sou h in Poli ical
Dilemma: Pos ma e ialis E ec s o Age on Go e nmen T us .”
Ge ysbu g, PA: The Cupola: Schola ship a Ge ysbu g College.
h ps://cupola.ge ysbu g.edu/s uden _schola ship/1113.
Collie , Ben. 2024. “Ta ge ed social media ads a e in luencing
ou beha iou — and he go e nmen uses hem oo.” The
Con e sa ion, Feb ua y 27. h ps:// hecon e sa ion.com/
a ge ed-social-media-ads-a e-in luencing-ou -beha iou -
and- he-go e nmen -uses- hem- oo-223576.
Collie , Ru h B. 1982. Regimes in T opical A ica: Changing
Fo ms o Sup emacy, 1945–1975. Be kley,
CA: Uni e si y o Cali o nia P ess, 1982.
Coni ze , Vincen , Rachel F eedman, Jobs Hei zig, Wesley H.
Holliday, Bob M. Jacobs, Na han Lambe , Milan Mossé
e al. 2024. “Social Choice Should Guide AI Alignmen in
Dealing wi h Di e se Human Feedback.” P ep in , a Xi ,
June 4. h ps://doi.o g/10.48550/a Xi .2404.10271.
Co es-Rami ez, Eugenio-En ique. 2015. “Cul u al Hegemony
Today. F om Cul u al S udies o C i ical Pedagogy.”
Pos colonial Di ec ions in Educa ion 4 (2): 116–39.
h ps://co e.ac.uk/download/pd /84894568.pd .
C eme , Ca la Zoe and Jess Whi les one. 2021. “A i icial Cana ies:
Ea ly Wa ning Signs o An icipa o y and Democ a ic
Go e nance o AI.” In e na ional Jou nal o In e ac i e
Mul imedia and A i icial In elligence 6 (5): 100–109.
h ps://dx.doi.o g/10.9781/ijimai.2021.02.011.
C iddle, C is ina, Madhumi a Mu gia and Geo ge Hammond. 2024.
“A i icial in elligence b eak h oughs c ea e new ‘b ain’ o
ad anced obo s.” Financial Times, July 26. www. .com/
con en /bea9d 71-371c-4045-9cb4-64c22789b 7b.
Cummings, M. L., Hea he M. Ro , Kenne h Cukie , Jacob Pa akilas
and Hannah B yce. 2018. A i icial In elligence and
In e na ional A ai s: Dis up ion An icipa ed. Cha ham House
Repo , June. www.cha hamhouse.o g/si es/de aul / iles/
publica ions/ esea ch/2018-06-14-a i icial-in elligence-
in e na ional-a ai s-cummings- o -cukie -pa akilas-b yce.pd .
Da oe, Allan. 2018. “AI Go e nance: A Resea ch Agenda.” Ox o d,
UK: Fu u e o Humani y Ins i u e, Ox o d Uni e si y.
www. hi.ox.ac.uk/wp-con en /uploads/Go AI-Agenda.pd .
Depa men o Science, Inno a ion & Technology. 2023.
“Capabili ies and isks om on ie AI: A discussion
pape on he need o u he esea ch in o AI isk.”
Oc obe 25. London, UK: Go e nmen o he Uni ed
Kingdom. www.go .uk/go e nmen /publica ions/
on ie -ai-capabili ies-and- isks-discussion-pape .
De echos Digi ales. n.d. “A La in Ame ican pe spec i e on he use o
AI sys ems by he S a e.” San iago, Chile: De echos Digi ales.
www.ohch .o g/si es/de aul / iles/Documen s/Issues/
Digi alAge/Submissions/CSOs/De echos-Digi ales.pd .
Edelman T us Ins i u e. 2024. 2024 Edelman T us Ba ome e ,
Supplemen al Repo : Insigh s o he Tech Sec o . NewYo k,
NY: Edelman T us Ins i u e. www.edelman.com/
si es/g/ iles/aa uss191/ iles/2024-03/2024%20
Edelman%20T us %20Ba ome e %20Supplemen al%20
Repo %20Insigh s%20 o %20Tech.pd .
Egan, Jane and E ic Rosenbach. 2023. “Biosecu i y in he Age o AI:
Wha ’s he Risk?” Policy b ie , No embe 6. Camb idge, MA:
Bel e Cen e o Science and In e na ional A ai s.
www.bel e cen e .o g/publica ion/
biosecu i y-age-ai-wha s- isk.
15Why Global Sou h Coun ies Need o Ca e Abou Highly Capable AI
El-Sayed, Seliem, Can e Akbulu , Amanda McC oske y, Geo
Keeling, Zacha y Ken on, Za ia Jalan, Nahema Ma chal e al.
2024. “A Mechanism-Based App oach o Mi iga ing Ha ms
om Pe suasi e Gene a i e AI.” P ep in , a Xi , Ap il 23.
h ps://doi.o g/10.48550/a Xi .2404.15058.
Felds ein, S e en. 2019. “The Global Expansion o AI Su eillance.”
Wo king Pape , Sep embe . Washing on, DC: Ca negie
Endowmen o In e na ional Peace.
h ps://ca negie-p oduc ion-asse s.s3.amazonaws.com/
s a ic/ iles/ iles__WP-Felds ein-AISu eillance_ inal1.pd .
———. 2021. The Rise o Digi al Rep ession: How Technology
is Reshaping Powe , Poli ics, and Resis ance.
New Yo k, NY: Ox o d Uni e si y P ess.
———. 2024. “AI in wa : Can ad anced mili a y echnologies be
amed be o e i ’s oo la e?” Bulle in o he A omic Scien is s,
Janua y 11. h ps:// hebulle in.o g/2024/01/
ai-in-wa -can-ad anced-mili a y- echnologies-be-
amed-be o e-i s- oo-la e/#pos -heading.
F an z, E ica. 2012. “How and Why do Dic a o ships Su i e?
Lessons o he Middle Eas .” B idgewa e Re iew 31 (2):
16–18. h ps:// c.b idgew.edu/b _ e / ol31/iss2/7/.
F an z, E ica, And ea Kendall-Taylo and Joseph W igh . 2020.
“Digi al Rep ession in Au oc acies.” Use s Wo king Pape Se ies
2020:27. Go henbu g, Sweden: V-Dem Ins i u e.
h ps:// -dem.ne /media/publica ions/
digi al- ep ession17ma .pd .
F eedom House. 2024. F eedom in he Wo ld 2024: The Moun ing
Damage o Flawed Elec ions and A med Con lic . Feb ua y.
Washing on, DC: F eedom House. h ps:// eedomhouse.o g/
si es/de aul / iles/2024-02/FIW_2024_Digi alBookle .pd .
Funk, Allie, Ad ian Shahbaz and Kian Ves einsson. 2023. “The
Rep essi e Powe o A i icial In elligence.” In F eedom
on he Ne 2023, edi ed by Ad ian Shahbaz, Allie
Funk, Jenni e B ody, Kian Ves einsson, G an Bake ,
Ca h yn G o he, Ma hew Ba ak, e al. Washing on,
DC: F eedom House. h ps:// eedomhouse.o g/si es/
de aul / iles/2023-11/FOTN2023Final.pd .
Gab iel, Iason, A ianna Manzini, Geo Keeling, Lisa Anne Hend icks,
Ve ena Riese , Hasan Iqbal, Nenad Tomaše e al. 2024. “The
E hics o Ad anced AI Assis an s.” P ep in , a Xi , Ap il 28.
h ps://doi.o g/10.48550/a Xi .2404.16244.
Ga dizy, Anissa and Ami E a i. 2024. “Mic oso and OpenAI Plo
$100 Billion S a ga e AI Supe compu e .” The In o ma ion,
Ma ch 29. www. hein o ma ion.com/a icles/mic oso -
and-openai-plo -100-billion-s a ga e-ai-supe compu e .
Gemini Team. 2023. “Gemini: A Family o Highly
Capable Mul imodal Models.” London, UK: Google
DeepMind. h ps://asse s.bwbx.io/documen s/
use s/iqjWHBFd xIU/ 7G7R T6 nM/ 0.
Geo gie a, K is alina. 2024. “AI Will T ans o m he Global Economy.
Le ’s Make Su e I Bene i s Humani y.” Janua y 14.
www.im .o g/en/Blogs/A icles/2024/01/14/
ai-will- ans o m- he-global-economy-le s-
make-su e-i -bene i s-humani y.
Gillani, Noo . 2023. “Will AI e e each human-le el in elligence?
We asked i e expe s.” The Con e sa ion, Ap il 18.
h ps:// hecon e sa ion.com/will-ai-e e - each-human-
le el-in elligence-we-asked- i e-expe s-202515.
Gillespie, Nicole, S e e Lockey, Cai lin Cu is, Ja ad Pool and
Ali Akba i. 2023. T us in A i icial In elligence: A Global
S udy. Queensland, AU: Uni e si y o Queensland.
h ps://asse s.kpmg.com/con en /dam/kpmg/au/
pd /2023/ us -in-ai-global-insigh s-2023.pd .
Gi i, Felipe Suá ez. 2023. “La in Ame ican Ci ies in he Age o AI:
Na iga ing he Technological Re olu ion.” Ha a d Kennedy
School S uden Policy Re iew, June 21.
h ps://s uden e iew.hks.ha a d.edu/la in-ame ican-ci ies-
in- he-age-o -ai-na iga ing- he- echnological- e olu ion/.
Ha is, Je emie, Edoua d Ha is and Ma k Beall. 2023. “Su ey o
AI Technologies and AI R&D T ajec o ies.” Alexand ia, VA:
Glads one AI. h ps://cdn.p od.websi e- iles.com/
62c4c 7322be8ea59c904399/
65e83959 d414a488a4 a9a5_Glads one%20
Su ey%20o %20AI.pd .
Hea h, Alex. 2024. “Ma k Zucke be g’s new goal is
c ea ing a i icial gene al in elligence: And he doesn’
wan o con ol i . Maybe.” The Ve ge, Janua y 18.
www. he e ge.com/2024/1/18/24042354/
ma k-zucke be g-me a-agi- eo g-in e iew.
Heikkilä, Melissa. 2024. “Google’s As a is i s i s AI- o -
e e y hing agen .” MIT Technology Re iew, May 14.
www. echnology e iew.com/2024/05/14/1092407/
googles-as a-is-i s- i s -ai- o -e e y hing-agen /.
Helle , Pa ick. 2022. “Democ acy in he Global Sou h.” Annual
Re iew o Sociology 48: 463–84. h ps://doi.o g/
10.1146/annu e -soc-030320-123449.
Hend ycks, Dan, Man as Mazeika and Thomas Woodside. 2023.
“An O e iew o Ca as ophic AI Risks.” P ep in , a Xi ,
Oc obe 9. h ps://doi.o g/10.48550/a Xi .2306.12001.
16 CIGI Pape s No. 311 — Decembe 2024 • Cecil Abungu, Ma ie Vic oi e I adukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali
Hiebe , Kyle. 2023. “Rising digi al su eillance h ea ens A ica’s
democ a ic p og ess.” ISS A ican Fu u es (blog), Oc obe 4.
h ps:// u u es.issa ica.o g/blog/2023/Rising-digi al-
su eillance- h ea ens-A icas-democ a ic-p og ess.
Hö s e , Tim. 2021. “U ili y o A i icial In elligence o Au ho i a ian
Go e nance.” P ep in , Social Science Resea ch Ne wo k,
Sep embe 15. h ps://pape s.ss n.com/sol3/
pape s.c m?abs ac _id=4074964.
Human Righ s Wa ch. 2023. “DR Congo: A oci ies by Rwanda-
Backed M23 Rebels.” News, Feb ua y 6. New Yo k, NY:
Human Righ s Wa ch. www.h w.o g/news/2023/02/06/
d -congo-a oci ies- wanda-backed-m23- ebels.
ICRC. 2019. A i icial in elligence and machine lea ning in
a med con lic : A human-cen ed app oach. Repo ,
June 6. Gene a, Swi ze land: ICRC. www.ic c.o g/
en/documen /a i icial-in elligence-and-machine-
lea ning-a med-con lic -human-cen ed-app oach.
Ipsos. 2022. “Global Opinions and Expec a ions abou
A i icial In elligence: A Global Ad iso su ey.”
Janua y. Pa is, F ance: Ipsos. www.ipsos.com/si es/
de aul / iles/c /news/documen s/2022-01/Global-
opinions-and-expec a ions-abou -AI-2022.pd .
Jenkins, J. C aig and Augus ine J. Kposowa. 1990.
“Explaining Mili a y Coups D’É a : Black A ica,
1957–1984.” Ame ican Sociological Re iew 55 (6):
861–75. h ps://doi.o g/10.2307/2095751.
Jø gensen, Ka en Helene. 2019. “The Meaning o In e ne Access
in Myanma .” Mas e ’s hesis, No wegian Uni e si y o Li e
Sciences. h ps://nmbu.b age.uni .no/nmbu-xmlui/
bi s eam/handle/11250/2612476/
J%C3%B8 gensen2019.pd ?sequence=1.
Juling, Dominik. 2023. “Fu u e Bio e o and Biowa a e Th ea s
o NATO’s A med Fo ces un il 2030.” Jou nal o Ad anced
Mili a y S udies 14 (1): 118–43. h ps://doi.o g/
10.21140/mcuj.20231401005.
Kapania, Shi ani, Oli e Siy, Gabe Clappe , Azhagu Meena
and Ni hya Sambasi an. 2022. “‘Because AI is 100% igh
and sa e’: Use A i udes and Sou ces o AI Au ho i y in
India.” Pape p esen ed a he CHI Con e ence on Human
Fac o s in Compu ing Sys ems. New O leans, LA, Ap il 29–
May 5. h ps://doi.o g/10.1145/3491102.3517533.
Kaplan, Ja ed, Sam McCandlish, Tom Henighan, Tom B.
B own, Benjamin Chess, Rewon Child, Sco G ay, Alec
Rad o d, Je ey Wu and Da io Amodei. 2020. “Scaling
Laws o Neu al Language Models.” P ep in , a Xi ,
Janua y 23. h ps://a xi .o g/pd /2001.08361.
Ka no sky, Holden. 2016. “Some Backg ound on Ou Views
Rega ding Ad anced A i icial In elligence.” Open
Philan h opy (blog), May 6. www.openphilan h opy.o g/
esea ch/some-backg ound-on-ou - iews- ega ding-
ad anced-a i icial-in elligence/#2-when-should-
we-expec - ans o ma i e-a i icial-in elligence.
Kausch, K is ina, Ai o Bonsoms, Can Kasapoglu, Lewin Schmi
and Žil inas Š edkauskas. 2022. Libe y’s Doom?
A i icial In elligence in Middle Eas e n Secu i y. Eu opean
Ins i u e o he Medi e anean. www.eu omesco.ne /
wp-con en /uploads/2022/07/Policy-S udy27.pd .
Kendall-Taylo , And ea, E ica F an z and Joseph W igh . 2020. “The
Digi al Dic a o s: How Technology S eng hens Au oc acy.”
Fo eign A ai s, Feb ua y 6. www. o eigna ai s.com/
a icles/china/2020-02-06/digi al-dic a o s.
Ken on, Zacha y, Ramana Kuma , Sebas ian Fa quha ,
Jona han Richens, Ma MacDe mo and Tom E e i .
2022. “Disco e ing Agen s.” P ep in , a Xi , Augus 24.
h ps://doi.o g/10.48550/a Xi .2208.08345.
Kla k, A hu and Gab iela Sagun. 2023. “Misin o ma ion, Ha e
Speech and E hno-Religious Tensions in Myanma .” Analysis,
Ap il 27. Washing on, DC: Uni ed S a es Ins i u e o Peace.
www.usip.o g/publica ions/2023/04/misin o ma ion-
ha e-speech-and-e hno- eligious- ensions-myanma .
Knuu ila, Aleksi, Lisa-Ma ia Neude and Philip N. Howa d.
2022. “Who is a aid o ake news? Modeling isk
pe cep ions o misin o ma ion in 142 coun ies.”
Ha a d Kennedy School Misin o ma ion Re iew 3 (3):
1–13. h ps://doi.o g/10.37016/m -2020-97.
Ko inek, An on and Megan Juel s. 2022. “P epa ing o he
(non-exis en ?) u u e o wo k.” Cen e on Regula ion
and Ma ke s Wo king Pape 3. Washing on, DC:
B ookings Ins i u ion. www.b ookings.edu/a icles/
p epa ing- o - he-non-exis en - u u e-o -wo k/.
Ko inek, An on and Joseph E. S igli z. 2021. “A i icial
In elligence, Globaliza ion, and S a egies o Economic
De elopmen .” Na ional Bu eau o Economic Resea ch
Wo king Pape 28453. Augus . www.nbe .o g/sys em/
iles/wo king_pape s/w28453/w28453.pd .
K ako na, Vic o ia, Jona han Uesa o, Vladimi Mikulik,
Ma hew Rah z, Tom E e i , Ramana Kuma , Zac Ken on,
Jan Leike and Shane Legg. 2020. “Speci ica ion gaming:
he lip side o AI ingenui y.” Medium, Ap il 21.
h ps://deepmindsa e y esea ch.medium.com/speci ica ion-
gaming- he- lip-side-o -ai-ingenui y-c85bdb0deeb4.
17Why Global Sou h Coun ies Need o Ca e Abou Highly Capable AI
Lehdon i a, Vili, Bóxī Wú and Zoe Hawkins. 2024. “Compu e No h
s. Compu e Sou h: The Une en Possibili ies o Compu e-based
AI Go e nance A ound he Globe.” P ep in , SocA Xi Pape s,
Augus 25. h ps://doi.o g/10.31235/os .io/8yp7z.
Lehmann, Jean-Pie e. 1985. “Dic a o ship and de elopmen
in Paci ic Asia: wide implica ions.” In e na ional A ai s
61(4): 591–606. h ps://doi.o g/10.2307/2617705.
Liu, Jun eng, Ch is ophe Symons and Ranga Raju Va sa ai.
2022. “Pe sona-Based Con e sa ional AI: S a e o
he A and Challenges.” P ep in , a Xi , Decembe 4.
h ps://doi.o g/10.48550/a Xi .2212.03699.
Maas, Ma hijs M., Kayla Luce o-Ma eucci and Di Cooke. 2023.
“Mili a y A i icial In elligence as a Con ibu o o Global
Ca as ophic Risk.” In The E a o Global Risk: An In oduc ion
o Exis en ial Risk S udies, edi ed by S. J. Bea d, Ma in
Rees, Ca he ine Richa ds and Cla issa Rios Rojas, 237–84.
Camb idge, UK: Open Book Publishe s.
h ps://books.openbookpublishe s.com/
10.11647/obp.0336.10.pd .
Manda ia, Megha. 2024. “AI Is Coming o India’s Famous Tech
Hub.” The Wall S ee Jou nal, Augus 6. www.wsj.com/ ech/
ai/ai-is-coming- o -indias- amous- ech-hub-97261085.
Man i, Rohin, Sama Khanna, Ma shall Bu ke, Da id Lobell
and S e ano E mon. 2024. “La ge Language Models a e
Geog aphically Biased.” P ep in , a Xi , Oc obe 5.
h ps://doi.o g/10.48550/a Xi .2402.02680.
Ma losa, Khabele. 2023. “Global ends and impac o democ a ic
ecession: Ha d choices o he Global Sou h.” Sou h
A ican Jou nal o In e na ional A ai s 30 (3): 337–55.
h ps://doi.o g/10.1080/10220461.2023.2269149.
Mo a, F ank and B ian Fonseca. 2019. “I ’s No he 1970s Again o
La in Ame ica’s Mili a ies. He e’s Why.” Ame icas Qua e ly,
Decembe 11. www.ame icasqua e ly.o g/a icle/i s-no -
he-1970s-again- o -la in-ame icas-mili a ies-he es-why/.
Mu gia, Madhumi a. 2023. “OpenAI chie seeks new
Mic oso unds o build ‘supe in elligence.’” Financial
Times, No embe 13. www. .com/con en /
dd9ba2 6- 509-42 0-8e97-4271c7b84ded.
Ngo, Richa d. 2020. “AGI Sa e y F om Fi s P inciples.” AI Alignmen
Fo um, Sep embe 28. www.alignmen o um.o g/s/
mzg mmTKKn5MuCzFJ.
———. 2023. “Visualizing he deep lea ning e olu ion.”
Medium, Janua y 5. h ps://medium.com/@ icha dcngo/
isualizing- he-deep-lea ning- e olu ion-722098eb9c5.
Ngo, Richa d, Law ence Chan and Sö en Minde mann. 2024. “The
Alignmen P oblem om a Deep Lea ning Pe spec i e.” P ep in ,
a Xi , Ma ch 19. h ps://a xi .o g/pd /2209.00626.
O’B ien, John T. and Cassidy Nelson. 2020. “Assessing he Risks
Posed by he Con e gence o A i icial In elligence and
Bio echnology.” Heal h Secu i y 18 (3): 219–27.
h ps://liebe pub.com/doi/pd /10.1089/hs.2019.0122.
Omeje, Kenne h. 2018. “Ex ac i e Economies and Con lic s in he
Global Sou h: Re-Engaging Ren ie Theo y and Poli ics.”
In Ex ac i e Economies and Con lic s in he Global Sou h:
Mul i-Regional Pe spec i es on Ren ie Poli ics, edi ed by
Kenne h Omeje, 1–25. Abingdon, UK: Rou ledge.
Onome. 2024. “How AI is Keeping Old Languages Ali e.”
Au oGPT by Minds eam, June 26. h ps://au ogp .ne /
he-impac -o -ai-in-languages-p ese a ion/#:~: ex =AI%20
aids%20language%20documen a ion%20and,and%20
suppo ing%20 eal%2D ime%20 ansla ion.
Onyekwena, Chukwuka and Mma Ama a Eke uche. 2019.
“The Global Sou h and de elopmen assis ance.”
Commen a y, Sep embe 16. Washing on, DC:
B ookings Ins i u ion. www.b ookings.edu/a icles/
he-global-sou h-and-de elopmen -assis ance/.
O d, Toby. 2020. The P ecipice: Exis en ial Risk and he Fu u e
o Humani y. New Yo k, NY: Hache e Books.
Palik, Júlia, Anna M. Obe meie and Si i A. Rus ad. 2022.
“Con lic T ends in A ica, 1989–2021.” PRIO
Pape . Oslo, No way: Peace Resea ch Ins i u e
Oslo. www.p io.o g/publica ions/13215.
Pa el, Nilay. 2023. “Inside Google’s big AI shu le — and how i
plans o s ay compe i i e, wi h Google DeepMind CEO Demis
Hassabis.” The Ve ge, July 10. www. he e ge.com/23778745/
demis-hassabis-google-deepmind-ai-alpha old- isks.
Phyo. 2023. “The Digi al Landscape o Ac i ism in Myanma .”
EngageMedia, No embe 29. h ps://engagemedia.o g/
2023/you h-myanma /.
Pipe , Kelsey. 2020. “The case o aking AI se iously as a h ea o
humani y: Why some people ea AI, explained.” Vox,
Oc obe 15. www. ox.com/ u u e-pe ec /2018/
12/21/18126576/ai-a i icial-in elligence-
machine-lea ning-sa e y-alignmen .
P abhaka an, Vinodkuma , Rida Qad i and Ben Hu chinson. 2022.
“Cul u al Incong uencies in A i icial In elligence.” P ep in ,
a Xi , No embe 19. h ps://doi.o g/10.48550/
a Xi .2211.13069.
18 CIGI Pape s No. 311 — Decembe 2024 • Cecil Abungu, Ma ie Vic oi e I adukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali
P i i e a, Daniel, Tamay Besi oglu, Rishi Bommasani, S ephen
Caspe , Yejin Choi, Danielle Gold a b, Hoda Heida i and
Leila Khala ba i. 2024. In e na ional Scien i ic Repo on
he Sa e y o Ad anced AI: In e im Repo . May 17. London,
UK: Depa men o Science, Inno a ion and Technology.
www.go .uk/go e nmen /publica ions/in e na ional-
scien i ic- epo -on- he-sa e y-o -ad anced-ai.
Pu zie , Kon ad. 2023. “Nex Wa e o Remo e Wo k Is Abou
Ou sou cing Jobs O e seas.” The Wall S ee Jou nal,
Ap il 11. www.wsj.com/a icles/nex -wa e-o - emo e-
wo k-is-abou -ou sou cing-jobs-o e seas-54a 39ba.
Ri chie, Hannah, Ve onika Sambo ska, Na asha Ahuja, Es eban
O iz-Ospina and Max Rose . 2023. “Global Educa ion.”
Ou Wo ldinDa a.o g. h ps://ou wo ldinda a.o g/global-
educa ion?insigh =child en-ac oss- he-wo ld- ecei e- e y-
di e en -amoun s-o -quali y-lea ning#key-insigh s.
Ro ey, Roge . 2004. “Biological weapons and po en ial
indica o s o o ensi e biological weapon ac i i ies.” In
SIPRI Yea book 2004: A mamen s, Disa mamen and
In e na ional Secu i y, edi ed by S ockholm In e na ional
Peace Resea ch Ins i u e, 557–71. Ox o d, UK: Ox o d
Uni e si y P ess. www.sip i.o g/yea book/2004/13.
Rose , Max. 2023. “Technology o e he long un: zoom ou o see
how d ama ically he wo ld can change wi hin a li e ime.”
Ou Wo ldinDa a.o g, Feb ua y 22.
h ps://ou wo ldinda a.o g/ echnology-long- un#.
Rus ad, Si i A. 2024. “Con lic T ends: A Global O e iew,
1946–2023.” PRIO Pape . Oslo, No way: Peace Resea ch
Ins i u e Oslo. www.p io.o g/publica ions/14006.
Schape a, Isaac. 1955. A Handbook o Tswana Law and
Cus om. 2nd ed. London, UK: F ank Cass & Co.
Schmid , E ic. 2023. “Inno a ion Powe : Why Technology
Will De ine he Fu u e o Geopoli ics.” Fo eign A ai s,
Feb ua y 28. www. o eigna ai s.com/uni ed-s a es/
e ic-schmid -inno a ion-powe - echnology-geopoli ics.
Sha i , Yonada , Sandhini Aga wal, Miles B undage, S e en
Adle , Cullen O’Kee e, Rosie Campbell, Teddy Lee e
al. 2023. “P ac ices o Go e ning Agen ic AI Sys ems.”
OpenAI pape , Decembe 14. h ps://cdn.openai.com/
pape s/p ac ices- o -go e ning-agen ic-ai-sys ems.pd .
Siani, Alessand o and Imogen G een. 2023. “Scien i ic
Misin o ma ion and Mis us o COVID-19 P e en i e Measu es
among he UK Popula ion: A Pilo S udy.” Vaccines1(2):301.
h ps://doi.o g/10.3390/ accines11020301.
So ala, Kaj and Roman V. Yampolskiy. 2015. “Responses
o ca as ophic AGI isk: a su ey.” Physica
Sc ip a 90: 018001. h ps://iopscience.iop.o g/
a icle/10.1088/0031-8949/90/1/018001/pd .
S oelke , Tom. 2018. “Ame ican TV, In e na ional Pe cep ions, and
Reali y.” Fo dham Now, Ma ch 28. h ps://now. o dham.edu/
inside- o dham/ acul y- eads/ame ican- -in e na ional-
pe cep ions-and- eali y/.
Szczepański, Ma cin. 2019. “Economic impac s o a i icial
in elligence (AI).” B ie ing, July. B ussels, Belgium: Eu opean
Pa liamen a y Resea ch Se ice. www.eu opa l.eu opa.eu/
RegDa a/e udes/BRIE/2019/637967/
EPRS_BRI(2019)637967_EN.pd .
Taleb, Abdel ahman Abu. 2024. “E idence o I an and UAE d ones
used in Sudan wa .” BBC, June 12. www.bbc.com/
news/a icles/c2 jz652j1o.
Tan, Jun-E, Ka hleen Azali and Ka e ina F ancisco. 2021. Go e nance
o A i icial In elligence (AI) in Sou heas Asia. Melbou ne,
AU: EngageMedia. Decembe . h ps://engagemedia.o g/
wp-con en /uploads/2021/12/Engage_Repo -Go e nance-
o -A i icial-In elligence-AI-in-Sou heas -Asia_12202021.pd .
Tao, Yan, Olga Vibe g, Ryan S. Bake and René F. Kizilcec. 2024.
“Cul u al Bias and Cul u al Alignmen o La ge Language
Models.” P ep in , a Xi , June 26.
h ps://a xi .o g/abs/2311.14096 2.
Tech Wi e Asia. 2018. “The impac o AI is as g ea as he
s eam engine: McKinsey.” Tech Wi e Asia, Augus 31.
h ps:// echwi easia.com/09/2018/ai-will-b ing-
global-g ow h-on-pa -wi h-s eam-engine/.
The Economis . 2023. “Cha GPT could eplace elema ke e s,
eache s and ade s.” The Economis , Ap il 14.
www.economis .com/g aphic-de ail/2023/04/14/
cha gp -could- eplace- elema ke e s- eache s-and- ade s.
Timcke, Sco and Hanani Hlomani. 2024. Decoding he
Ballo : How Migh AI Reshape Democ acy on he A ican
Con inen ? Cape Town, Sou h A ica: Resea ch ICT A ica.
h ps:// esea chic a ica.ne / esea ch/how-migh -ai-
eshape-democ acy-on- he-a ican-con inen /.
Toda o, Michael P. and S ephen C. Smi h. 2020. Economic
De elopmen . 13 h ed. Ha low, UK: Pea son.
T anspa ency In e na ional. 2024. Co up ion Pe cep ions Index
2023. Be lin, Ge many: T anspa ency In e na ional.
h ps:// anspa encia.o g.es/wp-con en /uploads/
Repo _CPI2023_English.pd .
