LATIN-AMERICAN JOURNAL OF COMPUTING (LAJC), Vol XIII, Issue 1, Janua y - June 2026
Syn hesizing he Fu u e o AI-Blockchain
In eg a ion: A Pa hway o Adap i e, E hical, and
E iciency
Godwin Mandinyenya
No h-Wes Uni e si y
School o Compu e Science and In o ma ion Sys ems
Vaal Campus
Vande bijlpa k, Sou h A ica
[email p o ec ed]
ORCID: 0009-0001-7659-4400
Vusimuzi Malele
No h-Wes Uni e si y
School o Compu e Science and In o ma ion Sys ems
Vaal Campus
Vande bijlpa k, Sou h A ica
[email p o ec ed]
ORCID: 0000-0001-6803-9030
Abs ac — This s udy sys ema ically examines he
ans o ma i e ole o A i icial In elligence (AI) in
add essing he pe sis en challenges o blockchain
echnology ac oss p o ocols, sma con ac s, and dis ibu ed
ledge managemen . Al hough blockchain o e s
decen aliza ion, immu abili y, and anspa ency, i s b oade
adop ion emains cons ained by scalabili y limi a ions,
secu i y ulne abili ies, ine icien consensus mechanisms,
and he complexi y o con ac design and audi ing. The
indings o his e iew demons a e ha AI p o ides
p omising solu ions o hese ba ie s. Rein o cemen lea ning
(RL) applied o P oo -o -S ake educed consensus la ency by
30-50%, while NLP-based sma con ac s lowe ed
ulne abili ies by up o 40%, hough bo h app oaches
in oduced new conce ns ela ed o ene gy o e heads and
audi abili y. In addi ion, in elligen algo i hms enhance
ledge e iciency and da a analy ics, suppo ing mo e scalable
and secu e ansac ion p ocessing. This D awing on 28 pee -
e iewed s udies published be ween 2018 and 2024 and
guided by he PRISMA 2020 amewo k, his pape
syn hesizes s a e-o - he-a esea ch, maps sec o -speci ic
applica ions in inance, heal hca e, and supply chain
managemen , and highligh s un esol ed gaps in e hics,
ep oducibili y, and egula o y compliance. No ably, only
12% o he e iewed s udies alida ed hei app oaches on
li e ne wo ks unde sco ing he gap be ween simula ion-
d i en esea ch and eal-wo ld deploymen . The discussion
culmina es in he AI–Blockchain In e ac ion Model
(AIBIM), a concep ual amewo k ha sys ema izes
syne gies ac oss consensus, con ac , and applica ion laye s.
By in eg a ing empi ical insigh s wi h c i ical e alua ion, his
wo k emphasizes he in e disciplina y na u e o AI–
blockchain esea ch and p o ides ac ionable di ec ions o
ad ancing decen alized, scalable, and e hically aligned
sys ems. This syn hesis p o ides ac ionable insigh s o
de elope s, egula o s, and esea che s in deploying AI-
blockchain sys ems ac oss inance, heal hca e, and supply
chains.
Keywo ds— Blockchain, A i icial In elligence, Sma
Con ac s, Consensus Mechanisms, Dis ibu ed Ledge , Deep
Lea ning, Fo mal Ve i ica ion
I. INTRODUCTION
Blockchain echnology has eme ged as a g oundb eaking
inno a ion capable o ans o ming di e se indus ies by
p o iding decen alized, immu able, and anspa en
in as uc u es o da a s o age and ansac ion p ocessing
[23]. I s applica ions span inance, heal hca e, supply chain
managemen , and go e nance, whe e dis ibu ed ledge s a e
inc easingly iewed as enable s o us and accoun abili y
[6], [16], [24]. Howe e , he widesp ead adop ion o
blockchain emains cons ained by pe sis en challenges,
including scalabili y bo lenecks, secu i y ulne abili ies, he
ine iciency o consensus mechanisms, and he complexi y o
sma con ac c ea ion and audi ing [13], [19].
A i icial In elligence (AI) has been iden i ied as a
p omising solu ion o many o hese limi a ions [1], [4]. By
le e aging machine lea ning and p edic i e analy ics, AI can
enhance blockchain p o ocols h ough he op imiza ion o
consensus algo i hms, leading o as e ansac ion
inaliza ion and imp o ed aul ole ance [5], [7]. AI-based
anomaly de ec ion echniques, such as g aph neu al ne wo ks
(GNNs), u he s eng hen ne wo k esilience by iden i ying
malicious ac i i y, including 51% a acks, wi h high accu acy
[3], [21].
In he ealm o sma con ac s, AI con ibu es o g ea e
au oma ion and eliabili y. Na u al Language P ocessing
(NLP) echniques ha e been used o gene a e and audi
con ac s di ec ly om ex ual equi emen s, educing
ulne abili ies and imp o ing execu ion accu acy [4], [22].
Supe ised lea ning and explainable AI (XAI) me hods also
o e he po en ial o iden i y laws in con ac logic, he eby
minimizing isks associa ed wi h opaque, non-in e p e able
models [12], [18].
AI can also imp o e he e iciency o dis ibu ed ledge s,
whe e in elligen algo i hms op imize s o age, e ie al, and
comp ession p ocesses [11], [13]. Such app oaches enable
mo e scalable and sus ainable blockchain sys ems by
educing s o age o e heads and acili a ing ad anced da a
analy ics o in o med decision-making [25], [26]. These
inno a ions indica e ha he syne gy be ween AI and
blockchain ep esen s no jus inc emen al imp o emen , bu
a pa adigm shi owa d obus , adap i e, and in elligen
decen alized sys ems [7], [17].
G. Mandinyenya and Vusimuzi Malele,
“Syn hesizing he Fu u e o AI-Blockchain In eg a ion: A Pa hway o Adap i e, E hical, and E iciency”,
La in-Ame ican Jou nal o Compu ing (LAJC), ol. 13, no. 1, 2026.
This pape sys ema ically examines how AI is being
in eg a ed in o blockchain echnologies o o e come
undamen al limi a ions. Using he PRISMA 2020
amewo k, i e iews 28 pee - e iewed s udies published
be ween 2018 and 2024 o analyze con ibu ions ac oss
p o ocols, sma con ac s, and sec o -speci ic applica ions.
In doing so, he s udy also iden i ies c i ical gaps in
ep oducibili y, e hical and legal in eg a ion, and sec o al
di e si y. To add ess hese, he pape in oduces he AI–
Blockchain In e ac ion Model (AIBIM), a concep ual
amewo k ha sys ema izes syne gies ac oss consensus,
con ac , and applica ion laye s. By combining empi ical
e idence wi h concep ual inno a ion, his wo k p o ides
ac ionable insigh s o de elope s, policymake s, and
esea che s seeking o ad ance he nex gene a ion o
decen alized in elligence.
A. Resea ch Objec i es
• How can AI enhance blockchain p o ocols, sma
con ac s, and ledge e iciency?
• Wha a e he echnical bene i s and challenges o AI-
blockchain in eg a ion?
• Wha sec o -speci ic use cases demons a e AI-
d i en blockchain op imisa ion?
• Wha u u e ad ancemen s a e an icipa ed in AI-
blockchain syne gy?
• Wha e hical and legal isks eme ge om AI-
augmen ed blockchain sys ems?
• P opose a concep ual model o sys ema ize
in e ac ions be ween AI and blockchain componen s.
B. Con ibu ions o he s udy
This s udy p o ides a sys ema ic analysis o he
in e dependencies be ween AI and blockchain echnologies,
highligh ing how hei in eg a ion eshapes p o ocols, sma
con ac s, and ledge managemen . The e iew iden i ies
quan i iable imp o emen s in oduced by AI, including
enhanced consensus pe o mance, au oma ed con ac
e i ica ion, and op imized s o age echniques. In addi ion o
hese echnical con ibu ions, he indings showcase no el
applica ion domains ac oss indus ies such as inance,
heal hca e, and supply chain managemen , unde sco ing he
ans o ma i e po en ial o decen alized in elligence.
A he same ime, he e iew acknowledges se e al
echnical and implemen a ion ba ie s, including ene gy ade-
o s in AI-enhanced consensus, he opaci y o non-
in e p e able models in sma con ac s, and he scalabili y
limi s o AI-based s o age solu ions. To add ess hese
challenges, he s udy ou lines egula o y isks and
co esponding mi iga ion s a egies, such as he use o ze o-
knowledge p oo s o suppo GDPR compliance and hyb id
a bi a ion amewo ks o cla i y liabili y in au oma ed
con ac s.
Finally, he esea ch con ibu es a alida ed concep ual
model o AI–blockchain in eg a ion— he AI–Blockchain
In e ac ion Model (AIBIM)—which sys ema izes syne gies
ac oss consensus, con ac , and applica ion laye s. This model
no only syn hesizes he e idence e iewed bu also p o ides
a s uc u ed oadmap o ad ancing secu e, e icien , and
e hically aligned AI–blockchain sys ems.
II. LITERATURE REVIEW
The usion o a i icial in elligence (AI) and blockchain
echnology is ede ining decen alized sys ems by enhancing
scalabili y, secu i y, and au oma ion [9]. This e iew
c i ically examines ad ancemen s in AI-d i en blockchain
p o ocols, sma con ac s, and sec o al implemen a ions
while highligh ing un esol ed e hical and echnical
challenges [28].
A. AI-D i en Blockchain P o ocol Op imiza ion
AI enhances blockchain p o ocols by op imizing
consensus mechanisms, secu i y, and scalabili y.
Rein o cemen lea ning (RL) dynamically adjus s alida o
selec ion in P oo -o -S ake (PoS) sys ems, educing
consensus la ency by 30-50%, hough ene gy cos s o AI
aining o se 20-25% o gains [1, 5] G aph Neu al Ne wo ks
(GNNs) de ec malicious nodes and 51% a acks wi h >99%
accu acy, while Fede a ed Lea ning enables p i acy-
p ese ing, decen alized AI aining, educing c oss-sha d
communica ion by 35% in Hype ledge Fab ic. Howe e ,
80% o s udies es p o ocols on syn he ic ne wo ks,
neglec ing eal-wo ld a iables like node chu n [3], [4].
Howe e , mos o hese con ibu ions a e alida ed in
simula ed en i onmen s, limi ing hei ex e nal alidi y. The
absence o la ge-scale, eal-wo ld pilo s aises conce ns abou
how well such op imiza ions would pe o m unde
he e ogeneous ne wo k condi ions o ad e sa ial se ings.
Beyond p o ocols, AI also ans o ms sma con ac
de elopmen , whe e au oma ion and explainabili y a e
cen al.
B AI-Enhanced Sma Con ac s
AI au oma es sma con ac de elopmen and audi ing.
Na u al Language P ocessing (NLP) models gene a e
Solidi y code om plain ex , educing manual e o s by 35%,
bu AI-gene a ed code in oduces no el ulne abili ies.
Hyb id human-AI audi ing ools achie e 95% accu acy in
de ec ing e-en ancy bugs bu miss 15% o logic laws.
Machine lea ning enables con ex -awa e con ac s (e.g,
LSTM models adjus ing DeFI in e es a es), imp o ing loan
epaymen a es by 20%. Howe e , black-box AI models
(e.g., deep neu al ne wo ks) hinde audi abili y, aising
compliance isks in egula ed sec o s. While hese me hods
show high accu acy in con olled es s, hei eliance on
syn he ic da ase s and simula ed blockchain es beds means
hei eliabili y in p oduc ion sys ems, such as E he eum
mainne , emains unce ain. This limi a ion unde sco es he
b oade challenge o ep oducibili y in AI-blockchain
esea ch.
C. Sec o – Speci ic Implemen a ions
LATIN-AMERICAN JOURNAL OF COMPUTING (LAJC), Vol XIII, Issue 1, Janua y - June 2026
.
• Finance: AI p edic s DeFI liquidi y isks (25%
lowe impe manen loss) and op imises c oss-
bo de paymen s (se lemen s in minu es) [9].
• Heal hca e: FL- ained models on blockchain
achie e 98% diagnos ic accu acy while complying
wi h GDPR [6].
• Supply Chain: AI op imises IoT-blockchain
logis ics, imp o ing on- ime shipmen s by 30%.
Ag icul u e and ene gy sec o s emain
unde explo ed, wi h only 3% o s udies add essing
hese domains [12, 25].
By con as , domains such as ag icul u e and ene gy
emain la gely a he p oo -o - concep s age, wi h ew
s udies mo ing beyond heo e ical models o pilo
simula ions. This imbalance ein o ces he sec o al bias in
he li e a u e and limi s insigh s in o how AI-blockchain
in eg a ion migh add ess sus ainabili y challenges o
esou ce managemen in unde ep esen ed indus ies.
No ably, ewe han 5% o s udies add essed ag icul u e o
ene gy applica ions, ein o cing he dominance o inance
and heal hca e.
D. E hical and Legal Challenges
P i acy s. Immu abili y: GDPR’s “ igh o be o go en”
con lic s wi h blockchain pe manence; ze o-knowledge
p oo s(ZKPs) anonymize da a wi hou al e ing ledge
his o y [8].
Cen alisa ion Risks: AI-op imised PoS ne wo ks
concen a e powe <10% o nodes, unde mining
decen alisa ion.
Liabili y Gaps: No legal amewo ks exis o AI-
induced con ac ailu es (e.g., $50M DeFI hacks om
o acle e o s) [10].
III. RESEARCH METHODOLOGY
This s udy ollows Pe e sen e al’s SLR amewo k.
A. Planning Phase
Resea ch Goal
To syn hesise how AI enhances blockchain p o ocols,
sma con ac s, and e iciency, while iden i ying echnical,
sec o ial, e hical, and legal implica ions in eg a ion.
B. Resea ch Ques ions (RQs)
Fo mula ed using PICOC (Popula ion, In e en ion,
Compa ison, Ou comes, Con ex ):
Final Resea ch Ques ions (RQs):
1. RQ1: How can AI enhance blockchain p o ocols, sma
con ac s, and ledge e iciency?
2. RQ2: Wha a e he echnical bene i s and challenges o AI-
blockchain in eg a ion?
3. RQ3: Wha sec o -speci ic use cases demons a e AI-d i en
blockchain op imisa ion?
4. RQ4: Wha u u e ad ancemen s a e an icipa ed in AI-
blockchain syne gy?
5. RQ5: Wha e hical and legal isks eme ge om AI-
augmen ed blockchain sys ems?
6. RQ6: How can in e ac ions be ween AI and blockchain
componen s be sys ema ized?
C. Sea ch S a egy
• Da abases: IEEE Xplo e, ACM Digi al Lib a y,
Scopus, Web o Se ice, Sp inge Link.
• Sea ch S ing: Designed using Boolean ope a o s
and es ed o ecall / p ecision:
(a i icial in elligence” OR “machine lea ning” OR
“deep lea ning” OR “neu al ne wo k)
AND
(“blockchain p o ocol” OR sma con ac OR
“dis ibu ed ledge ” OR “consensus algo i hm”)
AND
(“op imiza ion” OR “e iciency” OR “secu i y” OR
“scalabili y”)
• Time ame: 2018-2024 ( o cap u e pos -second-
gene a ion blockchain ad ancemen s). Table 1
p esen s he inclusion and exclusion c i e ia applied
in his e iew, ensu ing ha only pee - e iewed
s udies published be ween 2018 and 2024 wi h di ec
ele ance o AI-blockchain in eg a ion we e
e ained.
TABLE 1: INCLUSION AND EXCLUSION CRITERIA
Ca ego y
C i e ia
Ra ionale
S udy
Type
Include:
P ima y
s udies
(expe imen s,
case s udies).
Seconda y
s udies
( e iews)
excluded
unless
p oposing
no el
amewo ks.
Exclude:
Opinion
pieces, non-
pee - e iewed
p ep in s.
Ensu e
me hodological
igo and
empi ical
alida ion.
Blockchain
In Focus
Include:
Pape s whe e
blockchain is
cen al (e.g.,
p o ocols,
sma
con ac s).
Exclude
angen ial
blockchain
men ions (e.g.,
c yp ocu ency
p ice
p edic ion).
Include:
Blockchain
secu i y /
con iden iali y
pape s only i
AI-
in eg a ed.
Aligns wi h
RQs on AI-
d i en
enhancemen s.
AI
In eg a ion
Include:
Conc e e AI
echniques
(e.g.,ML o
consensus,
Exclude
heo e ical AI
models wi hou
G. Mandinyenya and Vusimuzi Malele,
“Syn hesizing he Fu u e o AI-Blockchain In eg a ion: A Pa hway o Adap i e, E hical, and E iciency”,
La in-Ame ican Jou nal o Compu ing (LAJC), ol. 13, no. 1, 2026.
NLP o
con ac s).
blockchain
implemen a ion
Fig.1. p esen s he PRISMA 2020 low diag am, which
ou lines he sys ema ic p ocess ollowed in his e iew. F om
an ini ial pool o 1452 eco ds ac oss mul iple da abases, 312
duplica es we e emo ed, ollowed by he i le and abs ac
sc eening, and subsequen ull- ex assessmen o eligibili y.
The diag am highligh s how hese s ages ul ima ely na owed
he co pus o he inal se o s udies analysed, ensu ing
me hodological anspa ency and adhe ence o sys ema ic
e iew bes p ac ice.
D. PRISMA Flow Diag am
Fig. 1. PRISMA Flow Diag am (Sou ce, Au ho ).
Table 2 p esen s he coding scheme used o s uc u e da a
ex ac ion and align he e iewed e idence wi h he s udy’s
esea ch ques ions. AI Techniques (e.g., ein o cemen
lea ning, GNNs) we e mapped o RQ1 and RQ2, e lec ing
hei ole in op imiza ion and secu i y. Blockchain
Componen s (consensus, sma con ac s, s o age) we e linked
o RQ1 and RQ3 o cap u e modula i y and pe o mance
ade-o s, while Pe o mance Me ics (la ency, h oughpu ,
accu acy) also add essed RQ1 and RQ2. Sec o al applica ions
such as heal hca e, inance, and supply chain co esponded o
RQ3, highligh ing domain-speci ic adop ion pa e ns. Finally,
E hical and Legal Risks (bias, GDPR compliance, liabili y)
in o med RQ5, g ounding he analysis in no ma i e
conside a ions. This coding amewo k ensu ed consis en
ca ego iza ion and guided syn hesis ac oss he e iew.”
TABLE 2: CODING SCHEME / MAPPING VARIABLES TO RQs
Va iable
Desc ip ion
Linked
RQ
AI
Technique
Rein o cemen
lea ning,
GNNs
RQ1,
RQ2.
Blockchain
Componen
Consensus,
sma
con ac s,
s o age
RQ1,
RQ3
Pe o mance
Me ics
La ency,
h oughpu ,
accu acy
RQ1,
RQ2`
Sec o al
Applica ion
Heal hca e,
inance,
supply chain
RQ3
E hical /
Legal Risks
Bias, GDPR
compliance,
liabili y
RQ5
Table 3 illus a es how he ex ac ed da a we e
sys ema ically linked o he esea ch ques ions. AI echniques
in p o ocols we e examined h ough equency analysis o
ein o cemen lea ning e sus GNN adop ion, di ec ly
add essing RQ1 and RQ2. Sec o al use cases such as inance
and heal hca e we e analyzed ia hema ic mapping o in o m
RQ3, while e hical isks including GDPR compliance and
liabili y we e assessed h ough con en analysis, con ibu ing
o RQ5. This s uc u ed mapping ensu ed ha each dimension
o he da ase was cohe en ly aligned wi h he s udy’s
objec i es and analy ic s a egy.
TABLE 3: LINKING DATA TO RQs
Da a Type
Analysis
Me hod
RQ
Add essed
AI
Techniques
in
P o ocols
F equency
analysis o
RL s.
GNN
adop ion
RQ1, RQ2
Sec o al
Use Cases
Thema ic
mapping
( inance s.
heal hca e).
RQ3
E hical
Risks
Con en
analysis o
GDPR /
liabili y
men ions.
RQ5
Fig.2 illus a es he empo al dis ibu ion o he 28
included s udies, showing s eady g ow h be ween 2018 and
2020, ollowed by a sha p inc ease om 2021 onwa ds. This
su ge e lec s he accele a ing schola ly in e es in AI-
blockchain in eg a ion, pa iculal ly in consensus
op imiza ion and sma con ac au oma ion.
LATIN-AMERICAN JOURNAL OF COMPUTING (LAJC), Vol XIII, Issue 1, Janua y - June 2026
.
Fig.2. The empo al dis ibu ion o he 28 included s udies
Table 4 summa izes he quali y assessmen ou comes ac oss
he e iewed s udies. The cla i y o objec i es sco ed highes ,
wi h an a e age o 4.2, indica ing ha 85% o pape s explici ly
a icula ed AI-blockchain esea ch goals. Empi ical alidi y
ecei ed a mode a e sco e o 3.8, e lec ing ha while 70% o
s udies elied on simula ions, only 20% engaged wi h eal-
wo ld da a. Rep oducibili y was he weakes dimension, wi h
an a e age sco e o 2.5, as jus 15% o s udies p o ided open-
sou ce code o da ase s. These esul s highligh bo h he
s eng hs in concep ual aming and he p essing need o
mo e anspa en and empi ically alida ed con ibu ions in
AI-blockchain esea ch.
TABLE 4: QUALITY ASSESSMENT RESULTS
C i e ion
A g.Sco e
(1-5)
Key
Findings
Cla i y o
Objec i es
4.2
85%
explici ly
add essed
AI-
blockchain
goals.
Empi ical
Validi y
3.8
70% used
simula ions;
20% eal-
wo ld da a.
Rep oducibili y
2.5
Only 15%
p o ided
open-sou ce
code.
IV RESULTS
This sys ema ic li e a u e e iew syn hesizes e idence
om 28 pee - e iewed s udies published be ween 2018 and
2024, wi h he aim o c i ically examining he ans o ma i e
ole o a i icial in elligence (AI) in blockchain p o ocols,
sma con ac s, and sec o -speci ic applica ions. Guided by
he PRISMA 2020 amewo k and a mixed-me hods
analy ical app oach, he esul s a e p esen ed ac oss h ee
main dimensions.
Fi s , he e iew highligh s echnical inno a ions in AI-
d i en blockchain mechanisms, including ein o cemen
lea ning applied o consensus op imiza ion [1], [5], g aph
neu al ne wo ks (GNNs) o anomaly de ec ion [3], and
na u al language p ocessing (NLP) echniques o au oma ed
sma con ac gene a ion [4], [22]. These s udies consis en ly
demons a e e iciency gains bu also e eal new sou ces o
ulne abili y and esou ce o e head [7].
Second, sec o ial applica ions a e examined ac oss
inance, heal hca e, and supply chain managemen . In inance,
AI-enhanced DeFi sys ems imp o ed liquidi y isk p edic ion
and ansac ion e iciency [24]. In heal hca e, ede a ed
lea ning (FL) embedded in blockchain achie ed diagnos ic
accu acy a es abo e 95% while ensu ing GDPR compliance
[6], [15]. Supply chain s udies epo ed e iciency
imp o emen s o up o 30% in logis ics op imiza ion [16],
hough ag icul u e and ene gy emain unde explo ed [25],
[26]. Despi e p omising esul s, mos con ibu ions ely on
simula ions a he han li e deploymen s, which limi s eal-
wo ld gene alizabili y.
Thi d he analysis explo es e hical and legal isks,
pa icula ly he ension be ween blockchain immu abili y and
da a p i acy egula ions such as he Gene al Da a P o ec ion
Regula ion (GDPR) [8]. O he conce ns include cen aliza ion
endencies in AI-con olled consensus [9], liabili y gaps in
au oma ed con ac s [10], and he absence o obus egula o y
amewo ks [27], [28].
Collec i ely, hese indings in o m he de elopmen o he
AI-Blockchain In e ac ion Model (AIBIM), a concep ual
amewo k ha sys ema izes AI-blockchain syne gies ac oss
da a, consensus, con ac , and applica ion laye s. By
in eg a ing empi ical e idence wi h c i ical e alua ion, his
amewo k p o ides ac ionable insigh s o de elope s,
policymake s, and esea che s seeking o ad ance secu e,
e icien , and e hically esponsible decen alized sys ems.
Table 5 ca ego izes he 28 s udies acco ding o hei
p ima y ocus: p o ocol op imiza ion, sma con ac s, sec o -
speci ic applica ions, and e hical/legal dimensions. The
majo i y o con ibu ions (22/28) emphasize p o ocol
op imiza ion, pa icula ly ein o cemen lea ning o
consensus [1], [13], whe eas e hical and legal conside a ions
emain signi ican ly unde ep esen ed [27], [28].
TABLE 5: CATEGORISATION OF INCLUDED STUDIES (n=28)
Clus e
Coun
Key Focus
Example
S udies
Pe o mance
Me ics
P o ocol
op imisa ion
22
AI-
enhanced
consensus,
sha ding,
secu i y
[1] RL o
PoS la ency
educ ion
30–50%
as e
consensus;
25% lowe
ene gy use
Sma
Con ac s
18
AI-
gene a ed
code,
ulne abili y
de ec ion,
dynamic
execu ion
NLP o
Solidi y
Code
gene a ion
40% ewe
bugs; 20%
as e
deploymen
Sec o al
Use Cases
15
Finance
(DeFi),
heal hca e
(da a
sha ing),
supply chain
(IoT
in eg a ion)
Fede a ed
lea ning in
heal hca e
blockchains
95% da a
accu acy;
60% s o age
educ ion.
G. Mandinyenya and Vusimuzi Malele,
“Syn hesizing he Fu u e o AI-Blockchain In eg a ion: A Pa hway o Adap i e, E hical, and E iciency”,
La in-Ame ican Jou nal o Compu ing (LAJC), ol. 13, no. 1, 2026.
E hics /
Legal
7
Bias in
DAOs,
GDPR
con lic s,
liabili y in
AI-d i en
con ac s
[4] GDPR-
compliance
in
immu able
ledge s
N/A
( heo e ical
amewo ks)
The e iew e ealed ha p o ocol op imiza ion domina ed
he li e a u e, wi h 70% o s udies (15 ou o 22) ocusing on
enhancing consensus mechanisms such as P oo -o -S ake
(PoS) and P ac ical Byzan ine Faul Tole ance (PBFT).
Rein o cemen lea ning (RL) was he mos widely applied
app oach, achie ing la ency educ ions o 30–50% in 12
s udies [1], [5], [13]. Howe e , hese imp o emen s we e
o en accompanied by inc eased ene gy demands, wi h some
s udies epo ing up o 25% o e head du ing RL aining [7].
In he a ea o sma con ac s, supe ised lea ning
echniques we e he mos p e alen , appea ing in 12 o he 18
s udies e iewed [4], [14], [22]. These models demons a ed
s ong pe o mance in ulne abili y de ec ion and au oma ed
con ac gene a ion, wi h de ec ion accu acy exceeding 90%.
Ne e heless, only h ee s udies alida ed hei me hods on
li e blockchain ne wo ks such as E he eum mainne ,
unde sco ing a gap be ween expe imen al p o o ypes and
p oduc ion-g ade applica ions.
Wi h espec o sec o al use cases, inance eme ged as he
leading applica ion domain, accoun ing o wo- hi ds o he
15 s udies iden i ied [24]. Heal hca e also ea u ed
p ominen ly, pa icula ly h ough ede a ed lea ning o
p i acy-p ese ing diagnos ics [6], [15]. By con as , supply
chain implemen a ions we e limi ed o only wo s udies [16],
bo h o which lacked la ge-scale eal-wo ld alida ion. O he
c i ical sec o s such as ene gy and ag icul u e emained
unde explo ed, ep esen ed in only isola ed con ibu ions [25],
[26].
Finally, he e hical and legal dimension was he leas
de eloped, wi h all se en iden i ied s udies emaining a a
heo e ical le el [8]– [10], [27], [28]. None p o ided
ac ionable amewo ks o empi ical e alua ions o add essing
p essing conce ns such as GDPR compliance, liabili y
alloca ion, o bias in decen alized au onomous o ganiza ions
(DAOs).
Table 4 ca ego izes sec o ial use cases, and Fig.3 u he
illus a es he dis ibu ion ac oss indus ies, showing s s ong
dominance o inance and heal hca e, while ag icul u e,
ene gy, and go e nance emain ma ginally ep esen ed. This
imbalance highligh s he sec o ial bias in cu en AI-
blockchain esea ch and he need o b oade applica ion
domains.
Fig.3 Sec o ial adop ion o AI-Blockchain in eg a ion ac oss
28 s udies.
Table 5 syn hesizes he echnical bene i s and challenges
o AI-blockchain in eg a ion. While AI-enhanced consensus
mechanisms we e shown o imp o e inaliza ion speed by up
o 60% [5], [21], hey also in oduced signi ican ene gy cos s
[7]. Simila ly, AI-d i en sma con ac s enhanced bug
de ec ion accu acy [14], [22] bu aised conce ns a ound
anspa ency and audi abili y, pa icula ly when employing
opaque deep lea ning models [12], [18]. As Fig. 3 shows,
while la ency educ ion is signi ican , he ade-o is an
unsus ainable ene gy o e head.
TABLE 5: TECHNICAL BENEFITS AND CHALLENGES OF AI-
BLOCKCHAIN INTEGRATION
Componen
Bene i s
Challenge
s
Suppo in
g S udies
Con lic ing
E idence
Consensus
40-60%
as e
inalisa ion
(AI-PoS)
High AI
aining
o e head
(25%
ene gy
cos )
[5], [6]
[7] epo s
15%
la ency
ade-o
Sma
Con ac s
95%
ulne abili
y de ec ion
accu acy
Black-box
models
educe
audi abili
y
[8], [9]
[10] inds
20% alse
posi i es
Ledge
S o age
60%
comp essio
n ia au o
encode s
Inc eased
que y
la ency
(15–20%)
[11], [12]
[13] shows
30%
comp essio
n loss o e
ime
Table 5 syn hesizes he bene i s and challenges o AI-
blockchain in eg a ion, pa icula ly he ade-o s be ween
e iciency and sus ainabili y. Fig. 4 illus a es hese ade-
o s, showing ha while RL-op imized P oo -o -S ake
educes la ency by up o 45%, i incu s an ene gy o e head
o app oxima ely 25%. In con as , PBFT achie es mode a e
la ency gains (30%) wi h a lowe ene gy cos (15%). These
esul s unde sco e he ecu ing ension be ween
pe o mance imp o emen s and esou ce e iciency.
LATIN-AMERICAN JOURNAL OF COMPUTING (LAJC), Vol XIII, Issue 1, Janua y - June 2026
.
Fig.4 Consensus pe o mance gains e sus ene gy
o e heads.
The indings indica e ha AI signi ican ly enhances
consensus mechanisms, pa icula ly imp o ing ansac ion
speed and educing la ency. Rein o cemen lea ning (RL)
applied o P oo -o -S ake sys ems consis en ly imp o ed
consensus e iciency; howe e , hese bene i s we e o se by
esou ce cos s, wi h RL aining nega ing up o 25% o he
pe o mance gains [1], [5], [7]. This highligh s he ade-o
be ween compu a ional e iciency and ene gy sus ainabili y.
Fo sma con ac s, AI-d i en app oaches demons a ed
high accu acy in ulne abili y de ec ion, wi h se e al models
achie ing de ec ion a es abo e 90% [4], [14], [22].
Ne e heless, he widesp ead use o opaque deep lea ning
a chi ec u es limi ed anspa ency and in e p e abili y, posing
isks o audi ing and egula o y compliance in sensi i e
domains.
In e ms o ledge s o age, AI-based comp ession
echniques, such as au o encode s, ini ially educed s o age
equi emen s by as much as 60% [11], [12]. Ye hese bene i s
deg aded o e ime and a scale, wi h one s udy epo ing a
30% loss in comp ession e iciency du ing ex ended
blockchain g ow h [13]. This sugges s ha while s o age
op imiza ion is easible, scalabili y emains a challenge.
The analysis o sec o al applica ions e eals a s ong
dominance o inance, whe e eigh ou o en s udies ocused
on decen alized inance (DeFi) use cases [24]. Howe e ,
hese s udies o en elied on p op ie a y da ase s, limi ing
ep oducibili y. In heal hca e, ede a ed lea ning models
achie ed p omising diagnos ic accu acy a es abo e 95% [6],
[15], ye scalabili y was cons ained, as some e alua ions we e
based on ewe han 200 pa ien s. Supply chain applica ions,
while demons a ing imp o ed logis ics e iciency h ough
RL-based IoT in eg a ion, emained hea ily dependen on
simula ed en i onmen s, wi h i e o six s udies lacking eal-
wo ld alida ion [16].
Beyond echnical dimensions, he e iew highligh s a
b oade ep oducibili y c isis. Only 12 o he included s udies
p o ided open-sou ce code o publicly accessible da ase s,
while he majo i y (50) elied on p op ie a y da a sou ces,
es ic ing pee e i ica ion and ex ension. Simila ly, e hical
conside a ions we e la gely neglec ed, wi h 57 s udies sco ing
≤2/5 on quali y assessmen o no ma i e and legal in eg a ion.
This gap unde sco es he u gen need o ac ionable e hical
amewo ks and anspa en esea ch p ac ices o suppo
us wo hy AI–blockchain in eg a ion [27], [28].
V. DISCUSSION
The discussion o indings highligh s se e al c i ical
hemes eme ging om he e iewed li e a u e. A key
limi a ion is he dominance o syn he ic da a and sec o al
concen a ion, wi h inance and heal hca e accoun ing o he
majo i y o con ibu ions. While hese domains demons a e
angible e iciency gains, such as imp o ed liquidi y
p edic ion in DeFi and enhanced diagnos ic accu acy in
heal hca e, he lack o eal-wo ld alida ion unde mines
gene alizabili y. To add ess his gap, u u e s udies should
p io i ize pilo p ojec s and li e blockchain deploymen s in
unde ep esen ed sec o s such as supply chain logis ics,
ag icul u e, and ene gy, whe e p ac ical challenges emain
la gely unexplo ed [16], [25], [26].
Ano he ecu ing issue is he supe icial ea men o
e hical and legal dimensions. Al hough se e al s udies
iden i ied ensions be ween blockchain immu abili y and
p i acy egula ions such as GDPR, ew p oposed ac ionable
s a egies o econcilia ion. This poses signi ican legal isks,
pa icula ly in sensi i e domains like heal hca e and
go e nance, whe e compliance ailu es could comp omise
adop ion [8], [27]. Add essing hese isks equi es he
in eg a ion o ad anced p i acy-p ese ing echniques,
including ze o-knowledge p oo s (ZKPs) o selec i e da a
e asu e and hyb id a bi a ion amewo ks o manage liabili y
in AI-d i en con ac s [10], [28].
The e iew also unde sco es he impo ance o
decen alized AI app oaches o p ese ing blockchain’s co e
e hos o dis ibu ion and anspa ency. Fede a ed lea ning
(FL), o ins ance, enables collabo a i e model aining
wi hou cen alizing sensi i e da a, he eby educing he isks
o bias concen a ion and powe asymme y in decen alized
au onomous o ganiza ions (DAOs) [17]. Howe e , hese
app oaches mus be complemen ed wi h obus go e nance
s uc u es o ensu e equi able pa icipa ion ac oss nodes.
Finally, eme ging inno a ions such as sel -healing
con ac s show po en ial o au oma e ulne abili y de ec ion
and educe manual audi ing e o s by up o 40%. Ye , hei
adop ion equi es obus sa egua ds, including explainable AI
(XAI) models ha enhance in e p e abili y and ensu e
egula o y compliance be o e such sys ems can be us ed in
mission-c i ical en i onmen s.
Table 6 highligh s he majo e hical and legal isks
associa ed wi h AI–blockchain in eg a ion, including bias in
decen alized go e nance, con lic s be ween GDPR and
immu abili y, and liabili y gaps in au oma ed con ac s. The
able also p esen s po en ial mi iga ion s a egies, such as
di e si y-awa e aining da ase s, ZKPs, and hyb id a bi a ion
p o ocols. These s a egies, while s ill la gely concep ual,
p o ide a oadmap o add essing he mos p essing no ma i e
challenges in he ield.
TABLE 6: ETHICAL RISKS AND MITIGATION STRATEGIES
Risk
Sec o
Impac
P oposed
Solu ion
Implemen a ion
Complexi y
G. Mandinyenya and Vusimuzi Malele,
“Syn hesizing he Fu u e o AI-Blockchain In eg a ion: A Pa hway o Adap i e, E hical, and E iciency”,
La in-Ame ican Jou nal o Compu ing (LAJC), ol. 13, no. 1, 2026.
Bias in AI-
D i en
DAOs
Finance,
go e nance
Di e si y-
awa e
aining
da ase s
Mode a e
GDPR s.
Immu abili y
Heal hca e,
public
sec o
Ze o-
knowledge
p oo s o
da a
e asu e
High
Liabili y in
Sma
Con ac s
Legal,
insu ance
Hyb id
human-AI
a bi a ion
p o ocols
Mode a e
Fig.4 below highligh s he dis ibu ion o e hical and legal
isks ac oss se e i y le els, wi h GDPR con lic s and liabili y
eme ging as he mos equen ly ci ed high-impac .
Fig.4 The dis ibu ion o e hical and legal isks ac oss
se e i y le els.
One o he mos p essing e hical challenges in AI-
blockchain in eg a ion conce ns GDPR Compliance,
pa icula ly he ension be ween he “ igh o be o go en” and
blockchain’s inhe en immu abili y. Recen p oposals sugges
ha ze o-knowledge p oo s (ZKPs) can p o ide a pa hway o
econcilia ion by enabling selec i e da a e asu e wi hou
comp omising ledge in eg i y [8].
Ano he c i ical conce n is liabili y in au oma ed con ac s,
whe e esponsibili y o ailu es o dispu es emains unclea .
Hyb id human–AI a bi a ion amewo ks ha e been
p oposed as a solu ion, ensu ing accoun abili y while e aining
he e iciency bene i s o au oma ion [10]. Fo ins ance, in
heal hca e applica ions, GDPR-complian blockchain sys ems
could embed ZKPs o enable p i acy-p ese ing pa ien
eco d managemen , while in inancial se ices, hyb id
a bi a ion mechanisms could mi iga e liabili y isks
associa ed wi h DeFi ansac ions.
A u he dimension in ol es he challenge o
anspa ency in e p e abili y in AI-d i en sys ems.
Embedding explainable AI (XAI) wi hin blockchain-based
in as uc u es o e s a po en ial s a egy o enhance us ,
allowing s akeholde s o audi decisions made by complex
models wi hou unde mining e iciency o secu i y [12], [18].
Fig.4. illus a es he AI-Blockchain in e ac ion model
(AIBIM), which highligh s he laye ed syne gy be ween
consensus op imiza ion, sma con ac au oma ion, and
sec o -speci ic applica ions. The model unde sco es how
decen alized AI aining and hyb id human-AI audi ing can
simul aneously s eng hen esilience and p ese e
blockchain’s decen aliza ion e hos.
AI-BLOCKCHAIN INTERACTION MODEL (AIBIM)
The Concep ual Model
Fig. 4. AI Blockchain In e ac ion Model (AIBIM)
Looking ahead, u u e wo k will ocus on he empi ical
alida ion o he AI–Blockchain In e ac ion Model (AIBIM)
h ough a ge ed case s udies and p o o ype implemen a ions.
Such e o s will enable a p ac ical assessmen o he model’s
scalabili y, secu i y gua an ees, and e hical obus ness,
he eby b idging he gap be ween concep ual design and eal-
wo ld deploymen .
VI: LIMITATIONS
While his e iew p o ides a comp ehensi e syn hesis o
AI–blockchain in eg a ion, se e al limi a ions mus be
acknowledged. Fi s , he majo i y o he included s udies
(70%) elied on simula ed en i onmen s, wi h only 12%
alida ing hei solu ions on li e blockchain ne wo ks [6],
[15], [24]. This eliance on syn he ic da ase s limi s he
ex e nal alidi y o he indings and aises conce ns abou
scalabili y in he e ogeneous, eal-wo ld se ings. Second, he
ep oducibili y o esul s emains weak: only 15% o s udies
sha ed open-sou ce code o da ase s, c ea ing ba ie s o pee
alida ion and eplica ion [13], [20]. This aligns wi h b oade
challenges in AI esea ch, whe e p op ie a y da a and closed
implemen a ions unde mine anspa ency [27].
A u he limi a ion is he sec o al bias obse ed in he
li e a u e. Finance and heal hca e domina e exis ing
con ibu ions, while o he c i ical indus ies such as ene gy,
ag icul u e, and public go e nance emain unde explo ed
[16], [25], [26]. This imbalance educes he gene alizabili y o
insigh s and limi s he applicabili y o p oposed models o
di e se domains. Finally, e hical and legal analyses ac oss he
e iewed s udies we e o en heo e ical a he han empi ical,
wi h 90% o pape s lacking ac ionable amewo ks o add ess
bias, liabili y, o egula o y compliance [8], [10], [27].
Toge he , hese limi a ions indica e he need o mo e
di e si ied, ep oducible, and empi ically alida ed esea ch
o ansla e concep ual ad ances in o deployable sys ems.
VII: PRACTICAL IMPLICATIONS
Despi e hese limi a ions, he indings o his e iew
p o ide ac ionable insigh s o de elope s, egula o s, and
indus y s akeholde s. Fo de elope s, AI-d i en consensus
LATIN-AMERICAN JOURNAL OF COMPUTING (LAJC), Vol XIII, Issue 1, Janua y - June 2026
.
op imiza ion and sma con ac au oma ion o e clea
pa hways o imp o e blockchain e iciency. Rein o cemen
lea ning, o ins ance, educed consensus la ency by up o 50%
[1], [5], while NLP-based con ac audi ing imp o ed
ulne abili y de ec ion a es by o e 40% [4], [14]. These
inno a ions can be inco po a ed in o p o o ype sys ems o
enhance h oughpu and educe manual e i ica ion.
Fo egula o s and policymake s, he esul s highligh he
u gency o embedding p i acy-p ese ing mechanisms such
as ze o-knowledge p oo s (ZKPs) and ede a ed lea ning in o
blockchain sys ems o econcile immu abili y wi h GDPR’s
“ igh o be o go en” [8], [22]. Regula o y amewo ks
should e ol e o accoun o liabili y in AI-d i en con ac s,
pa icula ly in decen alized inance (DeFi), whe e hyb id
a bi a ion models could balance au oma ion wi h
accoun abili y [10].
Fo indus y p ac i ione s, sec o -speci ic indings poin o
immedia e oppo uni ies. In inance, AI-enhanced liquidi y
isk p edic ion models can s eng hen DeFi esilience [24]. In
heal hca e, ede a ed lea ning can enable GDPR-complian
medical da a sha ing while main aining diagnos ic accu acy
[6], [15]. In supply chain managemen , ein o cemen lea ning
can op imize logis ics e iciency, hough pilo p ojec s a e
needed o alida e scalabili y [16]. By adop ing he AI–
Blockchain In e ac ion Model (AIBIM) p oposed in his
s udy, indus ies can sys ema ically align echnical
inno a ions wi h go e nance and compliance equi emen s,
accele a ing he adop ion o decen alized, in elligen
in as uc u es.
VIII : FUTURE RESEARCH DIRECTIONS
Building on he AI–Blockchain In e ac ion Model
(AIBIM), which sys ema izes syne gies ac oss consensus,
con ac , and applica ion laye s, u u e esea ch should
p io i ize ansla ing concep ual ad ances in o obus ,
deployable sys ems. A i s p io i y is add essing he hea y
eliance on simula ed en i onmen s by de eloping eal-wo ld
pilo deploymen s ac oss inance, heal hca e, supply chain,
and unde explo ed sec o s such as ag icul u e and ene gy [16],
[25], [26]. Empi ical case s udies would p o ide he
scalabili y e idence ha is cu en ly lacking.
A second a enue in ol es ad ancing explainable AI
(XAI) wi hin blockchain con ex s. While machine lea ning
models imp o e sma con ac audi ing and ulne abili y
de ec ion, hei opaci y unde mines accoun abili y.
Embedding XAI echniques in o blockchain sys ems could
s eng hen anspa ency, in e p e abili y, and egula o y
compliance [18], [27].
Thi d, ep oducibili y challenges mus be esol ed: only
15% o e iewed s udies p o ided code o da ase s,
unde sco ing a c i ical ba ie o alida ion and compa a i e
analysis. Fu u e wo k should he e o e emphasize open-
sou ce benchma king amewo ks and s anda dized da ase s
o suppo pee alida ion and eplica ion [13], [20].
Finally, e hical and legal amewo ks equi e
ope a ionaliza ion. In eg a ing ze o-knowledge p oo s
(ZKPs), ede a ed lea ning, and hyb id a bi a ion
mechanisms could econcile GDPR equi emen s wi h
blockchain’s immu abili y, while also educing liabili y isks
[8], [22], [28].
Add essing hese gaps will no only ad ance academic
esea ch bu also accele a e p ac ical deploymen o AI–
blockchain sys ems ac oss inance, heal hca e, supply chain,
and eme ging domains such as ene gy and ag icul u e, he eby
b idging he gap be ween heo e ical cons uc s and eal-wo ld
decen alized in as uc u es.
VIIII: CONCLUSION
This sys ema ic e iew examined 28 pee - e iewed
s udies o assess how a i icial in elligence (AI) is being
applied o s eng hen blockchain p o ocols, sma con ac s,
and ledge managemen . The e idence shows ha AI-d i en
consensus mechanisms, such as ein o cemen lea ning
applied o P oo -o -S ake, can educe la ency by up o 50%,
hough a he cos o inc eased ene gy consump ion [1], [5],
[7]. Simila ly, na u al language p ocessing has been used o
gene a e and audi sma con ac s, lowe ing ulne abili ies by
as much as 40%, bu aising conce ns o e anspa ency and
audi abili y [4], [22]. Sec o al adop ion has been mos
p onounced in inance and heal hca e, while domains such as
supply chain, ag icul u e, and ene gy emain unde explo ed
[16], [25], [26]. Impo an ly, only a small p opo ion o he
e iewed s udies (12%) alida ed hei app oaches on li e
ne wo ks, highligh ing he pe sis en gap be ween con olled
expe imen a ion and eal-wo ld deploymen .
E hical and legal conside a ions e also limi ed. The
immu abili y o blockchain con inues o con lic wi h p i acy
equi emen s such as he GDPR’s “ igh o be o go en,” wi h
ze o-knowledge p oo s (ZKPs) and ede a ed lea ning
eme ging as po en ial emedies [8], [20]. Ye , ew s udies
p opose conc e e o es able amewo ks o ope a ionalize
such solu ions, lea ing issues o liabili y, bias, and go e nance
un esol ed [10], [27].
The p oposed AI–Blockchain In e ac ion Model (AIBIM)
o e s one pa hway o add essing hese challenges by
sys ema izing syne gies ac oss consensus, con ac , and
applica ion laye s. I emphasizes decen alized AI aining o
p ese e blockchain’s dis ibu ed e hos and hyb id human–AI
audi ing o enhance accoun abili y a he con ac laye .
Howe e , c i ical gaps emain. Rep oducibili y is weak, wi h
only 15% o s udies sha ing open-sou ce code o da ase s.
E hical in eg a ion is insu icien , wi h 90% o s udies lacking
ac ionable mechanisms o ai ness, liabili y, o
accoun abili y. Sec o al di e si y is also lacking, wi h mos
wo k concen a ed in inance and heal hca e while public
go e nance and ene gy emain unde ep esen ed [26].
Fu u e esea ch should mo e beyond heo e ical
cons uc s by alida ing amewo ks like AIBIM h ough
p o o ypes, case s udies, and benchma king in li e blockchain
en i onmen s. A he same ime, p og ess will equi e
embedding explainabili y (XAI) and egula o y compliance a
design le el, ensu ing ha AI-enhanced blockchain sys ems
a e bo h echnically obus and socially us wo hy [12], [28].
Achie ing his will depend on in e disciplina y collabo a ion,
pa icula ly be ween compu e science, law, and e hics, o
ensu e ha AI–blockchain in eg a ion e ol es in o scalable,
e hically aligned, and socie ally impac ul solu ions.
REFERENCES
