Shiwangi Singh; Su abhi Singh; K aus, Sascha; Sha ma, Anuj; Dhi , Sanjay
A icle
Cha ac e izing gene a i e a i icial in elligence
applica ions: Tex -mining-enabled echnology
oadmapping
Jou nal o Inno a ion & Knowledge (JIK)
P o ided in Coope a ion wi h:
Else ie
Sugges ed Ci a ion: Shiwangi Singh; Su abhi Singh; K aus, Sascha; Sha ma, Anuj; Dhi , Sanjay (2024) :
Cha ac e izing gene a i e a i icial in elligence applica ions: Tex -mining-enabled echnology
oadmapping, Jou nal o Inno a ion & Knowledge (JIK), ISSN 2444-569X, Else ie , Ams e dam, Vol.
9, Iss. 3, pp. 1-12,
h ps://doi.o g/10.1016/j.jik.2024.100531
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/327434
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-nc-nd/4.0/
Cha ac e izing gene a i e a ificial in elligence applica ions: Tex -
mining-enabled echnology oadmapping
Shiwangi Singh
a
, Su abhi Singh
b
, Sascha K aus
c,d,
*, Anuj Sha ma
b
, Sanjay Dhi
e
a
Indian Ins i u e o Managemen Ranchi, Jha khand, India
b
Jindal Global Business School, O. P. Jindal Global Uni e si y, Sonipa , Ha yana, India
c
F ee Uni e si y o Bozen-Bolzano, Facul y o Economics & Managemen , Piazza Uni e si
a 1, 39100 Bolzano, I aly
d
Uni e si y o Johannesbu g, Depa men o Business Managemen , Johannesbu g, Sou h A ica
e
Depa men o Managemen S udies, Indian Ins i u e o Technology Delhi, New Delhi, India
ARTICLE INFO
A icle His o y:
Recei ed 19 Ap il 2024
Accep ed 28 July 2024
A ailable online 8 Augus 2024
ABSTRACT
This s udy aims o iden i y gene a i e AI (GenAI) applica ions and de elop a oadmap o he nea , mid, and
a u u e. S uc u al opic modeling (STM) is used o disco e la en seman ic pa e ns and iden i y he key
applica ion a eas om a ex co pus comp ising 2,398 pa en s published be ween 2017 and 2023. The s udy
iden ifies six la en opics o GenAI applica ion, including objec de ec ion and iden ifica ion; medical appli-
ca ions; in elligen con e sa ional agen s; image gene a ion and p ocessing; financial and in o ma ion secu-
i y applica ions; and cybe -physical sys ems. Eme gen opic e ms a e lis ed o each opic, and in e - opic
co ela ions a e explo ed o unde s and he hema ic s uc u es and summa ize he seman ic ela ionships
among GenAI applica ion a eas. Finally, a echnology oadmap is de eloped o each iden ified applica ion
a ea o he nea , mid, and a u u e. This s udy p o ides aluable insigh s in o he e ol ing GenAI landscape
and helps p ac i ione s make s a egic business decisions based on he GenAI oadmap.
© 2024 The Au ho s. Published by Else ie España, S.L.U. on behal o Jou nal o Inno a ion & Knowledge. This
is an open access a icle unde he CC BY-NC-ND license
(h p://c ea i ecommons.o g/licenses/by-nc-nd/4.0/)
Keywo ds:
Gene a i e AI
Technology oadmapping
Pa en s
Tex -mining
S uc u al opic modeling
Pa en da a mining
JEL classifica ions:
O30
O32
O33
In oduc ion
A ificial in elligence (AI) ad ancemen s, including gene a i e AI
(GenAI), ha e in oduced my iad oppo uni ies o bo h indi iduals
and business o ganiza ions (San ana & Díaz-Fe n
andez, 2023;Eapen
e al., 2023;Spanjol & Noble, 2023). Wi h he abili y o gene a e ex s
ha a e simila o hose w i en by humans (Pa lik, 2023), GenAI
algo i hms ha e a b oad ange o applica ions ac oss di e en indus-
ies (Ameen e al., 2023;Hend iksen, 2023). GenAI allows cha bo s
and i ual assis an s o ha e con ex ually ele an and human-like
con e sa ions, adding pe sonalized de ails o con e sa ions and
enhancing cus ome se ice e ficiency (Sila d e al., 2023). Addi ion-
ally, GenAI can enhance c ea i e exp essions and acili a e he c ea-
ion o a , music, and li e a u e con en . Fo ins ance, Cha GPT can
gene a e a wide a ie y o con en based on he use command,
including essays, poe y, concise summa ies, and answe s o use
ques ions (Ray, 2023).
As he echnology con inues o apidly e ol e, mapping he land-
scape o GenAI is c ucial (Ma iani & Dwi edi, 2024). Pa k e al. (2020)
defined oadmapping as “a p ocess ha mobilizes s uc u ed sys ems
hinking, isual me hods (e.g., oadmap ‘can as’) and pa icipa i e
app oaches o add ess o ganiza ional challenges and oppo uni ies,
suppo ing communica ion and alignmen o s a egic planning and
inno a ion managemen wi hin and be ween o ganiza ions a fi m
and sec o le els”(p. 2). Mo e specifically, echnology oadmapping
(TRM) is “ he p ocess o c ea ing isualiza ions o elemen s ela ed o
echnologies”(Naza ko e al., 2022). TRM can help iden i y ends
and u u e de elopmen oppo uni ies ac oss mul iple sec o s,
enabling fi ms o explo e new p oduc lines and ma ke oppo uni-
ies. The insigh s gene a ed can be use ul o guiding s a egic deci-
sions and ensu ing he u u e de elopmen o echnologies (Ca alho
e al., 2013). In addi ion, TRM can p o ide s a egic di ec ion and
esou ce op imiza ion by ou lining he sequence o echnology de el-
opmen (i.e., nea -, mid-, and a - u u e), acili a e s akeholde com-
munica ion by isualizing he dimensions o echnology
* Co esponding au ho .
E-mail add esses: [email p o ec ed] (S. Singh), su abhi.ii d1@gmail.
com (S. Singh), [email p o ec ed] (S. K aus), [email p o ec ed] (A. Sha ma),
[email p o ec ed] (S. Dhi ).
h ps://doi.o g/10.1016/j.jik.2024.100531
2444-569X/© 2024 The Au ho s. Published by Else ie España, S.L.U. on behal o Jou nal o Inno a ion & Knowledge. This is an open access a icle unde he CC BY-NC-ND license
(h p://c ea i ecommons.o g/licenses/by-nc-nd/4.0/)
Jou nal o Inno a ion & Knowledge 9 (2024) 100531
Jou nal o Inno a ion
&Knowledge
h ps://www.jou nals.else ie .com/jou nal-o -inno a ion-and-knowledge
de elopmen (Lee e al., 2013;Ramos e al., 2022), and help fi ms
iden i y in en o s o pa en applican s wi h whom o o m s a egic
pa ne ships.
Al hough p e ious s udies on GenAI ha e ocused on a a ie y o
ace s, including implica ions o inno a ion managemen (Ma iani &
Dwi edi, 2024;Ob eja e al., 2024), managemen educa o s (Ra en &
Jones, 2023), a el decision-making (Wong e al., 2023), human
esou ce managemen (Budhwa e al., 2023), inno a ion manage-
men (Id ees e al., 2023;Spanjol & Nobel, 2023), and in o ma ion
sys ems (Susa ala e al., 2023), he e has been a limi ed ocus on
oadmapping GenAI, explo ing how i eme ged, and iden i ying
u u e p ospec s. This s udy aims o map he echnological landscape
o GenAI using a ex -mining app oach (i.e., s uc u al opic model-
ing), ex ac ing GenAI- ela ed pa en s om pa en da ase s. Pa en s
ha e been shown o be a alid p oxy measu e o inno a ion and
echnological de elopmen s (Noh e al., 2021;Su e al., 2023) and
p e ious s udies on he blockchain (Zhang e al., 2021), e-comme ce
(Singh & Vijay, 2024), and au onomous d i ing (Su e al., 2023) ha e
used pa en da a o cons uc a echnology oadmap. To be e unde -
s and he GenAI landscape, he ollowing esea ch objec i es a e
p oposed:
To conduc a comp ehensi e mapping o he echnological clus-
e s and applica ions
To de elop a echnological oadmap o GenAI and o esigh o
he u u e
The findings o his s udy p o ide h ee con ibu ions. Fi s , he
esul s con ibu e o s udies on echnological o ecas ing and oad-
mapping, in eg a ing he cu en li e a u e on AI wi h oadmapping
p ac ices. By analyzing pa en da a ela ed o GenAI, his s udy
p esen s a oadmap o GenAI ad ancemen s ac oss a ious indus ies.
Second, while adi ional oadmapping me hodologies ha e elied
upon expe opinions, pa en -da a-d i en TRM o e s comp ehensi e
applica ion a eas o he GenAI. Thi d, he findings o his s udy can
assis decision-make s in AI esea ch and de elopmen .
Li e a u e e iew
Gene a i e AI
GenAI is a apidly e ol ing ca ego y o AI sys ems ha de elops
c ea i e con en based on p e- ained da a (Nguyen-Duc e al., 2023;
Ma iani & Dwi edi, 2024). Acco ding o Kalo a (2024), GenAI is “algo-
i hms (such as Cha GPT) ha can be used o c ea e new con en ,
including audio, code, images, ex , simula ions, and ideos”(p. 7).
Ad ancemen s in deep lea ning, la ge language models (LLMs), gen-
e a i e p e- ained ans o me s, na u al language p ocessing (NLP),
and di usion models ha e accele a ed he echnological capabili ies
o GenAI (Dwi edi e al., 2023a). By au oma ically gene a ing in o -
ma i e con en based on use inpu , GenAI can educe human e o s.
I can be used, o example, o es and debug codes, gene a e new
NFTs, and p o ide in o ma i e da a o aid in decision-making. GenAI
can speed up he de elopmen o c ea i e p ocesses, simpli y busi-
ness ope a ions, os e inc emen al o adical inno a ion, and gene -
a e in o ma i e con en ha can enhance business pe o mance
(Amankwah-Amoah e al., 2024).
Fi ms can use GenAI o au oma e ou ine ope a ions, pe sonalize
in o ma ion o specific p e e ences, imp o e ope a ional e ficiency,
and quickly adap o dynamic ma ke s. P edic i e models gene a ed
h ough GenAI can help fi ms iden i y new ma ke oppo uni ies and
cus omize p oduc s and se ices o mee indi idual cus ome s’
equi emen s (Fosso Wamba e al., 2023). In o he wo ds, GenAI can
help simpli y complex p ocedu es, enhance c ea i e alen s, and
enable cos educ ions.
GenAI can enhance human c ea i i y by boos ing di e gen hink-
ing, imp o ing unde s anding, and add essing knowledge bias. I can
gene a e ideas mo e quickly han humans, enabling indi iduals o
assess he iabili y o he ideas. GenAI can be applied o a wide ange
o p ac ical business scena ios, such as enhancing cus ome sa is ac-
ion, imp o ing ma ke ing s a egies, and ad ancing heal hca e se -
ices (Wamba e al., 2024). Fi ms can u ilize GenAI p edic i e
modeling o p edic cus ome p e e ence, ma ke ends, and com-
pe i i e ad an age. Medical esea che s a e explo ing he use o
GenAI echniques, such as a ificial neu al ne wo ks (ANN), o c ea e
an ibodies. E ec i e, sus ainable in eg a ion o GenAI echnologies
in o exis ing echnological sys ems mus add ess e hical issues, iden-
i y cons ain s, and encou age human−AI associa ion. Ma iani and
Dwi edi (2024) explained ha “GenAI can enable he usion and
hyb idiza ion o di e en ypes o inno a ion −such as p oduc , p o-
cess and ma ke ing inno a ion − hus pa ing he way o he eme -
gence o en i ely new business models”(p. 5). GenAI p o ides
dis inc i e ad an ages ac oss mul iple indus ies and can easily be
in eg a ed in o fi ms’exis ing echnological capabili ies o enhance
compe i i e ad an age.
Al hough GenAI can imp o e cus ome expe iences by p o iding
esponses, he e is he isk o dissa is ac ion i GenAI does no mee
use expec a ions in domains such as accu acy o esponsi eness.
Aydın and Ka aa slan (2022) highligh ed he possibili y o e o o
inadequa e con en c ea ion when employing GenAI, explaining ha
i can educe consume us . Addi ionally, he e is a secu i y isk
ha GenAI-gene a ed con en will e eal confiden ial in o ma ion
abou cus ome s o fi ms. Despi e hese challenges, GenAI p o ides
businesses wi h he oppo uni y o s eamline p ocesses, imp o e
consume engagemen , and os e inno a ion (Rubel e al., 2022).
Roadmapping
Co po a e o esigh is he “applica ion o u u es and o esigh
p ac ices by an o ganiza ion o ad ance i sel ”(Go don e al., 2020). I
in ol es analyzing ends, iden i ying signals, and o mula ing co po-
a e s a egies o plan o an unce ain u u e (Ge shman e al., 2016).
Co po a e o esigh con ibu es o inno a ion by p o iding s a egic
guidance, assis ing inno a ion ini ia i es, and challenging assump-
ions (Go don e al., 2020). Mul iple app oaches can be applied o co -
po a e o esigh , including compe i i e in elligence (Hakmaoui,
Oub ich, Calo , & Ghazi, 2022), benchma king (Calo e al., 2020), sce-
na io analysis (Fink & Schlake, 2000), and oadmapping analysis, also
e e ed o as TRM (Go don e al., 2020). Roadmapping analysis, in
pa icula , is an in eg al me hod o co po a e o esigh (Ozcan,
Homayoun a d, Simms, & Wasim, 2021).
TRM in eg a es echnology and ma ke -o ien ed elemen s in o a
cohesi e mul i- ie oadmap o p o ide a sys ema ic iew o
ad ancemen wi hin a echnological domain (Naza enko e al., 2022).
I can be used o map a b oad ange o echnologies o achie e di e se
pu poses, including he acquisi ion o compe i i e in elligence, o e-
cas ing, po olio managemen , s a egic planning, echnology man-
agemen , and echnology planning (Ding & He n
andez, 2023;Lee &
Pa k, 2005;Lee e al., 2007;Vasconcellos e al., 2014). P e ious s ud-
ies ha e documen ed he success ul applica ion o TRM (Chak abo y
e al., 2022;Le aba & P e o ius, 2022;Naza enko e al., 2022;Ozcan
e al., 2021;Wa anabe e al., 2020). Wi hin a echnology oadmap,
essen ial componen s include ime ame, know-why (i.e., ac o s
con ibu ing o alue c ea ion), know-how (i.e., encompassing ech-
nological de elopmen s), and linkages (A
s €
om e al., 2022;Ding &
He n
andez, 2023). Phaal e al. (2004) highligh ed eigh p ima y pu -
poses o TRM: s a egic planning, p oduc planning, knowledge asse
planning, capabili y planning, in eg a ion planning, long- ange plan-
ning, p og am planning, and p ocess.
De Alcan a a and Ma ens (2019) explained ha TRM “has he
abili y o show he in e ela ionship be ween ma ke , p oduc , and
S. Singh, S. Singh, S. K aus e al. Jou nal o Inno a ion & Knowledge 9 (2024) 100531
2
echnology and has been applied in a la ge numbe o indus ies”(p.
128). TRM is an impo an s ep in he s a egic planning p ocess ha
should be ini ia ed be o e a icula ing he p ojec po olio and de el-
ops o e mul iple i e a ions and efinemen s, in alignmen wi h he
o ganiza ion’s echnology s a egy. TRM u ilizes a ime-based o ches-
a ed amewo k o c ea e, demons a e, and dissemina e s a egic
plans o de eloping echnology, p oduc s, se ices, o ma ke s. The
TRM echnique is highly adap able, so i may be used o add ess a i-
ous o ganiza ional objec i es. TRM helps fi ms achie e a compe i i e
ad an age by de eloping and u ilizing inpu , ans o ma ion, and
ou pu -based capabili ies.
In TRM, he “ ocus should be on s a egic planning, wi h oadmap-
ping p o iding a mechanism, ca alys , and common language o ca y
he s a egic planning p ocess o wa d”(Phaal e al., 2005). TRM ena-
bles he fi m o achie e s a egic ans o ma ion by iden i ying s a-
egic oppo uni ies in he eme ging indus y, de eloping a echnical
oadmap, and p o iding ecommenda ions on whe e he fi m can
enhance i s echnological compe encies.
Me hodology
P io s udies on TRM ha e u ilized a a ie y o me hods, such as
ex mining (Liu e al., 2023;Ozcan e al., 2021); ex clus e ing
(Zhang e al., 2016); seman ic analysis (Miao, Wang, Li, & Wu, 2020);
keywo d ne wo k analysis and link p edic ion (Kim & Geum, 2021);
Bayesian ne wo ks (Jeong, Jang, & Yoon, 2021); he Delphi echnique
(Pa k e al., 2020); mo phological analysis (Bloem da Sil ei a e al.,
2018); and opic modeling based on la en Di ichle alloca ion
(Zhang, Daim, & Zhang, 2021).
This s udy adap s he s uc u al opic model (STM) by Robe s e
al. (2016) o ex ac la en opics om an ex ensi e collec ion o pa -
en ex documen s and unde s and ela ed ends. STM is an unsu-
pe ised s a is ical machine-lea ning me hod ha iden ifies a opic
as a p obabilis ic dis ibu ion o seman ically associa ed e ms (Rob-
e s, S ewa , & Ai oldi, 2016;S
anchez-F anco & A amendia-Mune a,
2023). In o he wo ds, STM clus e s equen ly co-occu ing and
seman ically ela ed e ms in a ex co pus, and hese clus e s a e
defined as la en opics (K aus e al., 2023). STM is p e e able o adi-
ional opic modeling app oaches because he opic modeling p ocess
inco po a es documen -le el co a ia es ha imp o e he causal
in e ence and quali a i e in e p e abili y o he la en hema ic s uc-
u es (Sha ma, Rana, & Nunkoo, 2021). STM enables esea che s o
app oxima e he ela ionship be ween me ada a co a ia es and opi-
cal p e alence, acili a ing he analysis o how opic con en and
p e alence a y as pe documen -le el co a ia es (Dwi edi e al.,
2023b).
The da a-gene a ing p ocess unde STM is depic ed in Fig. 1. Each
node has a sepa a e ole; he obse ed a iables a e shown in shaded
nodes and la en a iables a e ep esen ed as unshaded nodes. The
ec angles cha ac e ize eplica ion as he ex co pus has D-indexed
documen s, and each documen , d, has e ms indexed by N
d
. The
numbe o opics (K) is selec ed empi ically by he esea che s. The
opic- e m dis ibu ion and pe -documen opic p opo ions a e wo
key la en a iables ha cap u e he mix u e o opics wi hin he
documen s and he p obabili y dis ibu ion o each opic o e e ms.
The co e language model gene a es opic p opo ions o each docu-
men and hen es ima es pe - e m opic assignmen and opic-wo d
dis ibu ion o each wo d in he documen .
Da a and da a p e-p ocessing
This s udy began wi h he iden ifica ion o sea ch que y key-
wo ds. Based on keywo ds in he p e ious li e a u e (e.g., K aus e al.,
2022;Saue and Seu ing, 2023), he ollowing sea ch s ing was used
o sea ch he pa en s: “la ge language models”OR “LLMs”OR “Con-
e sa ional Agen s”OR GPT* OR *GPT OR “Dall-E”OR BARD OR
LaMDA OR “Gene a i e P e- ained T ans o me ”OR “Gene a i e
Models”OR “P e- ained Gene a i e Models”OR “Gene a i e Ad e -
sa ial Ne wo k.”Using hese keywo ds, 2,985 pa en s we e iden ified
be ween 2017 and 2023. The yea 2017 was chosen as he s a ing
poin o his oadmapping s udy as i ma ks he beginning o a sig-
nifican pe iod o ad ancemen s in he field o GenAI, including P o-
g essi e GAN (2017), GPT-2 and GPT-3 (2019, 2020), DALL-E 2
(2023), Cha GPT (2022), and GPT-4 (2023) (Bengesi e al., 2024). A e
fil a ion o ele ancy, 2,398 pa en s we e e ained o final analysis.
The pa en s we e lis ed in a ious pa en o fices, including he U.S.
Pa en and T adema k O fice, he Eu opean Pa en O fice, and he
Japan Pa en O fice. The op coun ies ha ha e filed pa en s include
in he Uni ed S a es, Sou h Ko ea, China, Japan, India, G ea B i ain,
Taiwan, Ge many, Canada, and Aus alia.
The ex co pus o his s udy was p epa ed by conca ena ing he
i le and abs ac o pa en documen s, which is a s anda d p ocedu e
in opic modeling (Madzík, Fal
a , Yada , Liza elli, &
Ca nogu sk
y,
2024). Tex p e-p ocessing in ol ed he emo al o non-English cha -
ac e s, basic English s op wo ds, punc ua ion ma ks, and coun y
names (Gao, Wang, & Wu, 2023;Singh e al., 2020;Singh e al.,
2023). An n-g am okenize was implemen ed in he R language o
iden i y he mos equen big ams and ig ams, which we e hen
con e ed in o unig ams o p ese e he seman ics o hese wo ds
(Goodell, Kuma , Li, Pa naik, & Sha ma, 2022). Topic models wi h
a ying numbe s o opics we e es ed o empi ically selec he op i-
mal numbe o opics. Pas s udies ha e confi med ha he op imal
numbe o opics can be chosen based on exclusi i y sco es, seman ic
Fig. 1. Pla e No a ion o STM (adap ed om Robe s e al. (2016)).
S. Singh, S. Singh, S. K aus e al. Jou nal o Inno a ion & Knowledge 9 (2024) 100531
3
cohe ence, and held-ou likelihood (Sha ma, Koohang, Rana, Abed, &
Dwi edi, 2023;Sha ma e al., 2021;Singh, Singh, Koohang, Sha ma,
& Dhi , 2023). Fig. 2 illus a es ha seman ic cohe ence d ops sha ply
when he numbe o opics is g ea e han six, hus, a model wi h six
opics was used.
The s udy adap ed i s app oach o de eloping a echnology oad-
map om p e ious s udies, including Ozcan e al. (2021) and Lee e
al. (2008). The laye s o he echnology oadmap, i.e., ma ke d i e s
and p ocesses, we e iden ified and he ime lag was adjus ed based
on he pa en applica ion da e o classi y i as in he nea -, mid-, o
a - u u e. The p ocesses we e iden ified om he pa en da ase and
linked o he ma ke d i e s. Based on he iden ified laye s and ime
lag o pa en s, he GenAI oadmap was p epa ed.
Resul s
A keywo d analysis was pe o med on he i le and de ails o each
pa en o unde s and he ends and concep s. A o al o 5,102 key-
wo ds we e iden ified. The keywo ds wi h he highes equency
included “ne wo k”(9,135 ins ances), “da a”(6,995), “ad e sa ial”
(5,565), “lea ning”(2,577), “inpu ”(2,454), “neu al”(2,283), “disc im-
ina o ”(2,006), “plu ali y”(1,936), “gene a ion”(1,402), and “appa a-
us”(1,114). Va ious neu al ne wo k a chi ec u es a e used in GenAI
models, including con olu ional neu al ne wo ks (CNNs), ecu en
neu al ne wo ks (RNNs), and gene a i e ad e sa ial ne wo ks
(GANs). Fu he , da ase s a e equi ed o ain hese GenAI models.
“Ad e sa ial”deno es an ad e sa ial aining echnique used in
GANs. “Lea ning”is ela ed o supe ised and unsupe ised lea ning
echniques. The inpu da a can exis in a a ie y o o ma s, including
ex , audio, and image. Fu he , he “disc imina o ”is a componen in
GANs ha dis inguishes be ween eal and gene a ed da a. “Plu ali y”
deno es he mul iplici y o gene a ed ou pu s. “Appa a us” e e s o
he amewo ks, ools, o pla o ms used o de elop and deploy
GenAI models.
STM associa es documen s wi h key opics ha can be exp essed
using he op eme gen e ms. Table 1 summa izes he key opics and
he associa ed op e ms based on he p obabili y o occu ence. A
ew exempla y pa en documen s a e also p o ided o help u he
explo a ion and analysis. The p oposed opic labels and defini ions
a e based on he op e ms and associa ed pa en documen s o each
opic (Ammi a o, Felice i, Linzalone, Co ello, & Kuma , 2023). Mos
pa en s wi hin he co pus ha e been egis e ed unde financial and
in o ma ion secu i y applica ions (22.8%), al hough image gene a ion
and p ocessing (21.7%) has also a ac ed significan a en ion om
GenAI in en o s and esea che s.
Topic 1: objec de ec ion and iden ifica ion
Topic 1, Objec De ec ion and Iden ifica ion, ep esen s he domi-
nan esea ch ela ed o he de ec ion o objec s, loca ions, pa e ns,
and ou lie s om digi al images and ideos. The wide- anging appli-
ca ions o GAN models in de ec ing s ee objec s, finding anomalies
in medical images, and ges u e con ol in human− obo in e ac ion
(B ophy e al., 2023;Ho man e al., 2023;Wu e al., 2022) a e well-
add essed in pa en documen s. Con empo a y compu e ision ech-
niques hea ily exploi deep gene a i e models o a wide ange o
objec de ec ion and iden ifica ion applica ions.
Topic 2: medical applica ions
Topic 2, Medical Applica ions, mainly ocuses on applica ions o
GenAI in he heal hca e indus y, including medical imaging echnol-
ogies, d ug disco e y and de elopmen , and medical esea ch and
da a analysis (Chen & Esmaeilzadeh, 2024;Haz a & Byun, 2020;Yi e
al., 2019). Se e al pa en s ha e been egis e ed ha use ensemble
GANs o simula e biomedical signals ela ed o ca dio ascula dis-
ease. GANs a e also used o gene a e p o ein sequences and de ec
eal o coun e ei chemicals in medical d ugs.
Topic 3: in elligen con e sa ional agen s
Topic 3, In elligen Con e sa ional Agen s, encompasses pa en s
ela ed o he applica ion o AI-based con e sa ional agen s like cha -
bo s o i ual agen s ha can in e ac wi h humans o e ex o
oice in e aces (Mekni, 2021). Mul ipu pose con e sa ional agen s
based on deep lea ning echniques o p ocessing na u al language
que ies ha e been p oposed and deployed o a ious business appli-
ca ions (Allouch e al., 2021). Filed pa en s o in elligen con e sa-
ional agen s p opose sys ems and me hods o in eg a ing hese
Fig. 2. Es ima ing he op imal numbe o opics wi hin he model.
S. Singh, S. Singh, S. K aus e al. Jou nal o Inno a ion & Knowledge 9 (2024) 100531
4
Table 1
Topics p opo ion and eme gen opic e ms.
Topic and Topic P opo ion Defini ion Eme gen Topic Te ms Sample Pa en s
Objec De ec ion and Iden ifica ion
(12.7%)
Sys ems and me hods o de ec ing
objec s, loca ions, and Abno mali ies
Image, Resolu ion, Vehicle, Embodimen , Gene a i e Ad e sa ial Ne -
wo k, P oduc , Componen , Damage, Roadway, De ec o
1. Gene a i e ad e sa ial ne wo k models o de ec ing small s ee
objec s
2. Sys em and me hod o u ilizing weak supe ision and a gene a-
i e ad e sa ial ne wo k o iden i y a loca ion
Medical Applica ions (14.6%) Compu e -aided diagnosis, disease p e-
dic ion, and physiological
in e p e a ion
Elec oca diog am, An ibody, Cellula Image, Amino Acid Sequence,
P edic ion, Abno mal Flow De ec ion, F ame, Sequence, Reflec ion,
Composi ion
1. Ensemble gene a i e ad e sa ial ne wo k-based simula ion o ca -
dio ascula disease-specific biomedical signals
2. Elec oca diog am gene a ion de ice based on gene a i e ad e -
sa ial ne wo k algo i hm and me hod he eo
In elligen Con e sa ional Agen s (13.9%) Sys ems and me hods o de elopmen ,
deploymen , in eg a ion, and moni o -
ing o con e sa ional agen s such as
cha bo s
Con e sa ional Agen , Response, Na u al Language Que y, De ec ion,
Ve ifica ion, Deep Lea ning Technique, Cha bo , Me chan , Sel -Dis-
closu e, Suppo
1. Me hod and sys em o swi ching and hando e be ween one o
mo e in elligen con e sa ional agen s
2. Sys em o moni o ing and in eg a ion o one o mo e in elligen
con e sa ional agen s
Image Gene a ion and P ocessing (21.7%) P ocessing, gene a ing, edi ing, and
es o ing digi al images
Image, Syn he ic, Image P ocessing, Ou pu , In o ma ion, Image Gen-
e a ion, Lea ning, Fea u e, Disc imina o , Segmen a ion
1. Gene a i e ad e sa ial ne wo k o p ocessing and gene a ing
images and label maps
2. Me hod o gene a ing image gene a ion model based on gene a-
i e ad e sa ial ne wo k
Financial and In o ma ion Secu i y Appli-
ca ions (22.8%)
Classifica ion and p edic ion sys ems o
finance and da a secu i y- ela ed asks
Da a, T aining, Model, Co a iance, Gene a i e Ad e sa ial Ne wo k,
Loan, Bo owe , Time-Se ies, Beha io In e ence Model, Semi-
Supe ised
1. Me hod and appa a us o examina ion o financial c edi using
a ificial neu al ne wo k, gene a i e ad e sa ial ne wo k, and
ein o cemen lea ning
2. Me hod o pe o ming con inual lea ning on c edi sco ing wi h-
ou ejec in e ence and eco ding medium eco ding compu e
eadable p og am o execu ing he me hod
Cybe -Physical Sys ems (14.3%) In elligen sys ems o sensing, compu a-
ion, con olling, moni o ing, op imiz-
ing, and communica ing wi h o he
sys ems
Model, De ice, Con olle , Gene a i e Ad e sa ial Ne wo k, In o ma-
ion, Ta ge , Signal, P ocess, Gene a i e-AI, Cybe -Physical Sys em
1. Sys em and Me hod o Abs ac ing Cha ac e is ics o Cybe -Physi-
cal Sys ems
2. Me hod o Con olling a Home Appliance
S. Singh, S. Singh, S. K aus e al. Jou nal o Inno a ion & Knowledge 9 (2024) 100531
5
echniques in di e en sec o s. Al hough he esea ch in his a ea is
s ill eme ging, i has significan po en ial o e ol e in he u u e.
Topic 4: image gene a ion and p ocessing
Topic 4, Image Gene a ion and P ocessing, ep esen s he use o
GenAI and deep lea ning echniques o gene a e digi al images and
apply di e en e ec s o images. Tex - o-image syn hesis, he manip-
ula ion o image e ec s, and he es o a ion o images a e he mos
common use cases o GenAI in ela ion o digi al images (Gu e al.,
2022;Liu e al., 2021). The ecen pa en s filed and g an ed wi hin
his opic p opose he applica ion o ans o me s and GANs in image
p ocessing and digi al image gene a ion.
Topic 5: financial and in o ma ion secu i y applica ions
Topic 5, Financial and In o ma ion Secu i y Applica ions, includes
pa en s ela ed o financial c edi analysis, anomaly de ec ion in
financial da a, financial o ecas ing, isk assessmen , and c edi sco -
ing. GenAI’s subs an ial p oduc i i y and ope a ional e ficiency has
he po en ial o e olu ionize he financial indus y and ela ed sec-
o s (Kanbach, Heiduk, Bluehe , Sch ei e , & Lahmann, 2024;Zheng
e al., 2024). GenAI is e olu ionizing he finance sec o by analyzing
da a a iances o suppo aud de ec ion (Rane, 2023). Simila ly,
applica ions o GenAI in in o ma ion secu i y and da a p i acy ocus
on iden i ying po en ial b eaches and ulne abili ies, gene a ing
cybe a ack simula ions, p io i izing isk modeling, and au oma ing
secu i y asks.
Topic 6: cybe -physical sys ems
Topic 6, Cybe -Physical Sys ems, ocuses on in elligen , compu e -
based sys ems (P o en e al., 2021) ha can p ocess subs an ial
amoun s o da a and in eg a e sensing, moni o ing, con ol, and ne -
wo king in o physical p ocesses in a digi al en i onmen (Nayak,
Naik, Vimal, & Fa o skaya, 2024). Mos pa en s wi hin his opic
apply o cybe -physical sys ems such as ehicle con olle s, cabin
moni o ing sys ems, sma home de ices, secu e p i a e ne wo ks,
and indus ial au oma ion. Cybe -physical sys ems p ima ily empha-
size he need o de elop in e aces and p ocesses o acili a ing
de ice-le el in e ac ion o moni o and con ol in e ne -o - hings-
based sys ems wi hin cybe space.
Discussion
A co ela ion analysis u ilizing an es ima ed ma ginal opic p o-
po ion co ela ion ma ix was pe o med o u he in es iga e and
quan i y he associa ions be ween he iden ified opics. Fig. 3
p esen s he co ela ions among he six opics, he alues o which
a e all less han 0.3. The nega i e alues confi m no documen s
wi hin he co pus con ain equal e e ences o any wo opics.
One o he main ad an ages o STM is ha i can be used o in es-
iga e he in e ac ions be ween co a ia es and opics. The opic p o-
po ion is es ima ed as a unc ion o he publica ion yea . The
empo al dynamics in opic p opo ions indica e changes in he pop-
ula i y o each opic o e ime and can be used o iden i y eme gen
hemes based on inc easing ends in opic p opo ion. Fig. 4 depic s
he ends o each o he six opics o e he s udy pe iod. Topic 2
(Medical Applica ions), Topic 3 (In elligen Con e sa ional Agen s),
and Topic 6 (Cybe -Physical Sys ems) all show a ising end. Topic 4
(Image Gene a ion and P ocessing) shows a sligh decline a e 2022,
hough i con inues o a ac a significan amoun o schola ly ocus.
Finally, Topic 1 (Objec De ec ion and Iden ifica ion) and Topic 5
(Financial and In o ma ion Secu i y Applica ions) show a g adually
declining end.
Roadmapping (nea - u u e, mid- u u e, and a - u u e)
Based on he iden ified opics, a echnology oadmap was gene -
a ed o GenAI o he nea , mid, and a u u e, mapping ma ke d i -
e s and p ocesses o each clus e . The nea - u u e ad ancemen s a e
hose ha a e expec ed in he nex 0−2 yea s; he mid- u u e
ad ancemen s a e an icipa ed in 2−5 yea s; and a - u u e de elop-
men s a e an icipa ed in 5−10 yea s (Table 2).
In he nea u u e, objec de ec ion and iden ifica ion ad ance-
men s can ocus on le e aging deep lea ning models o gene a e
no el chemical compounds o d ug design, enhancing oad sa e y by
in eg a ing ad anced d i ing assis ance sys ems, and implemen ing
p edic i e main enance solu ions ha o ecas equipmen damage
and op imize main enance schedules o minimize down ime in
Fig. 3. Co ela ion among opics.
S. Singh, S. Singh, S. K aus e al. Jou nal o Inno a ion & Knowledge 9 (2024) 100531
6
indus ial ope a ions. Medical de elopmen s om GenAI e ol e
a ound enhancing diagnos ic capabili ies and ea men planning
h ough medical image syn hesis, s eamlining documen a ion asks
o gene a e accu a e and ele an medical epo s, and acili a ing
pe sonalized in e ac ion expe iences. By imp o ing he accu acy and
e ficiency o medical p ocesses and clinical wo kflows, hese de elop-
men s a e poised o significan ly enhance pa ien ca e.
In he sho e m, he an h opomo phic na u e o in elligen
con e sa ional agen s and hei con ex ual esponses a e
expec ed o imp o e, enhancing use engagemen . These de elop-
men s include he gene a ion o dialogue esponses, que y−key-
wo d ma ching, and he gene a ion o cus omized con en . Using
GenAI echniques, con e sa ional agen s will be able o ca e o
indi idual use needs, enhancing use sa is ac ion, os e ing que y
unde s anding, and c ea ing mo e ele an and pe sonalized con-
en . A he same ime, image gene a ion and p ocessing
capabili ies will ocus on enhancing p e en i e main enance p ac-
ices h ough a isual anomaly de ec ion sys em, imp o ing he
accu acy o heal hca e diagnos ics h ough medical image noise
educ ion, and ad ancing imme si e expe iences h ough i ual
ea u e maps. Th ough he use o GenAI echniques, fi ms can
make significan imp o emen s in ope a ional e ficiency, diagnos-
ics capabili ies, and use sa is ac ion.
Nea - e m ad ancemen s in financial and in o ma ion secu i y
applica ions can ocus on enhancing and op imizing pe o mance
h ough p e- aining sys ems o sel -lea ning agen s, p edic i e
main enance and aul p edic ion o a oid de ice ailu es and mi i-
ga e isks, and ne wo k op imiza ion o enhance applica ion pe o -
mance, eliabili y, and secu i y. GenAI de elopmen s in cybe -
physical sys ems include cybe -secu i y measu es o ackle e ol ing
isks. Fu he , GenAI can also be le e aged o imp o e he cla i y and
esolu ion o moni o ing sys ems.
Fig. 4. E olu ion o eme gen opics.
S. Singh, S. Singh, S. K aus e al. Jou nal o Inno a ion & Knowledge 9 (2024) 100531
7
Table 2
Technology oadmap.
Objec De ec ion
and Iden ifica ion
P ocesses D ug design D i ing assis ance
sys em
Equipmen damage
p edic ion sys em
Machine
ansla ions using
LLMs
Collision a oidance
and mic o
na iga ion
Resol ing ime delays Au oma ed isual
inspec ion
Syn he ic human
finge p in s
Se ice- obo
Ma ke -
d i e
Gene a e chemical
compounds
wi h desi ed
cha ac e is ics
Sa e y enhancemen
Op imiza ion
Risk-mi iga ion
Da a-d i en decision
making
In eg a ion wi h
business
wo kflows
Au onomous
na iga ion
Cus ome expe ience
enhancemen
Quali y assu ance Biome ic secu i y
solu ions
Aging popula ion
Human- obo
in e ac ion
Medical
Applica ions
P ocesses Medical image syn hesis Medical ex gene a ion Opinion exp ession
based on consis en
s yle o pe sonali y
Elec onic medical
eco d en i y
ecogni ion
Dep ession diagno-
sis in e iew
Speech ecogni ion
(unspoken ex and
speech syn hesis)
Gene a ion o p o-
ein sequence
P ognosis e alua-
ion o b eas
cance
Disease ea ly
wa ning
p edic ion
Ma ke -
d i e
Diagnosis and ea men
planning
Gene a ing accu a e and
con ex ually
ele an medical
epo s
Pe sonalized in e ac ion
s yles
Au oma ion o da a
en y
Scalabili y o men-
al heal h
se ices
Accessibili y imp o e-
men s (speech
impai men s)
Applica ion in
pha maceu ical
esea ch and
bio echnology
T ea men
planning
Disease p e en ion
In elligen Con e -
sa ional Agen s
P ocesses Gene a ion o dialogue
esponses
Que y-keywo d
ma ching
Gene a ion o cus om-
ized con en
E ficien da a co-
clus e ing
Pe sona-based dia-
logue modeling
Gene a ing acial
exp essions in a use
in e ace
Con e sa ional
in e ace o
APIs
Swi ching and hando e be ween one o
mo e in elligen con e sa ional agen s
Ma ke -
d i e
Na u al and con ex u-
ally ele an
esponses
Accu a e que y
unde s anding
Pe sonalized con en
c ea ion
Collabo a i e
fil e ing
Enhance use
engagemen
Vi ual eali y
Augmen ed eali y
Simpli y API
in e ac ion
Agen coo dina ion
Mul i-agen sys ems
Image Gene a ion
and P ocessing
P ocesses Visual anomaly de ec-
ion sys em
Medical image noise
educ ion me hod
Vi ual ea u e maps Pose-in a ian ace
ecogni ion
Medical image
segmen a ion
Ta ge iden ifica ion De-iden ifica ion o
pe sonal
in o ma ion
Tu ning a 2-D
image in o a
Skybox
Sys em o o ming
e al a ec ion
Ma ke -
d i e
P e en i e main enance
Risk mi iga ion
Diagnosis and ea men
planning
VR, AR
Imme si e
expe iences
In a ian ea u e
ex ac ion
Disease diagnosis
T ea men
planning
Au oma ed a ge
iden ifica ion
Anonymize pe -
sonal
in o ma ion
Imme si e media
expe iences
Enhance pa en al
bonding
Financial and In o -
ma ion Secu i y
Applica ions
P ocesses P e- aining sys em o
sel -lea ning agen
P edic ing ailu e in
de ices
Ne wo k op imiza ion Examina ion o
financial c edi
using a ificial
neu al ne wo k
De eloping and
deploying
anomaly de ec-
ion sys ems
Communica ion e fi-
cien machine lea n-
ing o da a
Robus deep gene -
a i e models
Gene a ing syn-
he ic poin
cloud da a
De ec ing unde-
ec ed ne wo k
in usion ypes
Ma ke -
d i e
Adap and op imize
pe o mance
P edic i e main enance
Faul p edic ion
Imp o e pe o mance,
eliabili y, and
secu i y
E alua e c edi -
wo hiness
Mi iga e finan-
cial isks
Cybe secu i y
h ea s
E ficien communica ion
p o ocols
In eg a ion o
obus ness
echniques
Risk assessmen
Asse alua ion
Cybe h ea s
Malwa e a acks
Cybe -Physical
Sys ems
P ocesses Fine- uning AI models Enhancing images Modi ying esponses o
imp o e accu acy
Abs ac ing cha ac-
e is ics o
cybe -physical
sys ems
Reading compu e
files
Di ec ional ecommen-
da ions (ac i i y
acking)
P edic ing de ice
main enance
Gene a ing syn-
he ic da a
Con e sa ion cu a-
o sys em o
s uc u ed
in e ac ions
Ma ke -
d i e
Cybe secu i y h ea s Imp o e cla i y and
esolu ion
Enhancing sys em pe -
o mance
Enhancing use
sa is ac ion
Op imiza ion
Faul de ec ion
Au oma ed da a
analysis
Pe sonalized ecom-
menda ions
Na iga ion assis ance
Fo ecas de ice
ailu es
Schedule
main enance
Robus aining
Valida ion o AI
models
S uc u ed in e ac-
ions
Dialogue
managemen
Nea Fu u e (0-2 yea s) Mid-Fu u e (2-5 Yea s) Fa -Fu u e (5-10 yea s)
S. Singh, S. Singh, S. K aus e al. Jou nal o Inno a ion & Knowledge 9 (2024) 100531
8
