Resea ch Pape
Recommended ci a ion: Xu, D., & Ma in, D. (2025). Explo ing The Use o
Cha GPT by Compu e Science S uden s in So wa e De elopmen : Applica ions,
E hical Conside a ions, and Insigh s o Enginee ing Educa ion. In Kangaslampi,
R., Langie, G., Jä inen, H.-M., & Nagy, B. (Eds.), SEFI 53 d Annual Con e ence.
Eu opean Socie y o Enginee ing Educa ion (SEFI), Tampe e, Finland. DOI:
10.5281/zenodo.17631596.
This Con e ence Pape is b ough o you o open access by he 53 d Annual Con e ence
o he Eu opean Socie y o Enginee ing Educa ion (SEFI) a Tampe e Uni e si y in
Tampe e, Finland. This wo k is licensed unde a C ea i e Commons
A ibu ion-NonComme cial-Sha e Alike 4.0 In e na ional License.
EXPLORING THE USE OF CHATGPT BY COMPUTER SCIENCE
STUDENTS IN SOFTWARE DEVELOPMENT:
APPLICATIONS, ETHICAL CONSIDERATIONS, AND INSIGHTS FOR
ENGINEERING EDUCATION
D. Xu a, D. Ma in b,
1
a Uni e si y College London, London, Uni ed Kingdom, 0009-0001-9211-7876
b Uni e si y College London, London, Uni ed Kingdom, 0000-0002-9368-4100
Con e ence Key A eas: Digi al ools and AI in enginee ing educa ion
Keywo ds: compu e science educa ion, so wa e p ojec s, la ge language models,
Cha GPT
ABSTRACT
Cha GPT has been inc easingly used in enginee ing educa ion, pa icula ly in
compu e science, o e ing e icien suppo ac oss so wa e de elopmen asks. While
i helps s uden s na iga e p og amming challenges, i s use also aises conce ns abou
academic in eg i y and o e eliance. Despi e g owing in e es in his opic, p io
esea ch has la gely elied on su eys, emphasizing ends o e in-dep h analysis o
s uden s’ s a egies and e hical awa eness. This s udy complemen s exis ing wo k
h ough a quali a i e in es iga ion o how compu e science s uden s in one UK
ins i u ion s a egically and e hically engage wi h Cha GPT in so wa e de elopmen
p ojec s. D awing on semi-s uc u ed in e iews, i explo es wo key ques ions: (1)
How do compu e science s uden s e hically and s a egically epo using Cha GPT
in so wa e de elopmen p ojec s? (2) How do s uden s unde s and and pe cei e he
e hical issues associa ed wi h using Cha GPT in academic and p o essional con ex s?
Findings e eal a shi in s uden s’ lea ning models, mo ing om adi ional
“independen hinking–manual coding–i e a i e debugging” o “AI-assis ed idea ion–
in e ac i e p og amming–collabo a i e op imiza ion.” Impo an ly, many use Cha GPT
con e sa ionally o deepen unde s anding, while consciously ese ing c ea i e and
high-le el decision-making asks o hemsel es. S uden s end o cap Cha GPT’s
con ibu ion o oughly 30%, and e alua e i s ou pu o mi iga e o e eliance. Howe e ,
only a mino i y ho oughly analyze AI-gene a ed code, aising conce ns abou educed
c i ical engagemen . Meanwhile, s uden s ejec unc edi ed use, highligh isks such
as p i acy b eaches and skill deg ada ion, and call o clea usage guidelines se by
hei eache s. This esea ch o e s no el insigh s in o he e ol ing lea ne -AI dynamic
and highligh s he need o explici guidance o suppo esponsible and pedagogically
sound use o such ools.
1
Co esponding Au ho
1 INTRODUCTION
Gene a i e A i icial In elligence (GenAI) is apidly e ol ing and pe mea ing educa ion,
esea ch, and p ac ice, wi h he po en ial o undamen ally ans o m eaching and
lea ning (Adel e al., 2024). Cha GPT, known o i s widesp ead use and ad anced
echnical capabili ies, has become a key suppo ool in enginee ing educa ion (Salas-
Pilco & Yang, 2022). Howe e , while i o e s signi ican lea ning ad an ages, i aises
e hical conce ns (Fe gus e al., 2023). This dual na u e unde sco es he need o
deepe esea ch.
The s udy ocuses on compu e science (CS) s uden s o hei unique ole as use s
who a e expec ed o unde s and he p inciples behind GenAI echnologies. Thei
echnical aining allows hem o unde s and he capabili ies and limi a ions o la ge
language models, ecognizing hem as "s a is ical co ela ion machines" a he han
sys ems based on causal easoning (Zhou e al., 2024). Addi ionally, CS s uden s
ha e a heigh ened awa eness o e hical issues, a ising om hei aining and
engagemen wi h he socie al impac s o AI echnologies (Rahman & Wa anobe,
2023).
Mo eo e , a co e aspec o CS educa ion a e so wa e de elopmen p ojec s. They
in ol e mul iple dimensions including coding and epo w i ing, making hem well-
sui ed o GenAI applica ion esea ch (Akba e al., 2023). The e o e, his s udy aims
o explo e he s uden s’ usage pa e ns and e hical conside a ions o Cha GPT in
so wa e de elopmen , ocusing on add essing he ollowing wo esea ch ques ions:
How do compu e science s uden s e hically and s a egically epo using Cha GPT
in so wa e de elopmen p ojec s?
How do s uden s unde s and and pe cei e he e hical issues associa ed wi h using
Cha GPT in academic and p o essional con ex s?
2 BACKGROUND
2.1 Cu en Resea ch o GenAI in Enginee ing Educa ion
The de elopmen o GenAI is d i ing inno a ion in enginee ing educa ion, b inging
unp eceden ed possibili ies o eaching me hods and p ac ical app oaches (Nikolic e
al., 2023). As a powe ul ool, Cha GPT enhances lea ning e iciency h ough
pe sonalized lea ning suppo , adap i e lea ning asks, and ins an eedback (Mai y
and De oy, 2024). I demons a es unique ad an ages in simpli ying he so wa e
de elopmen p ocess (Fe gus e al., 2023). Cha GPT is widely used in a ious s ages
o so wa e de elopmen , including code gene a ion, debugging, epo w i ing, and
gene a ing es cases, o e ing e icien solu ions o s uden s’ p og amming challenges
(Adiguzel e al., 2023). By au oma ing epe i i e asks, hese ools allow s uden s o
ocus on highe -le el asks such as algo i hm op imiza ion and p oblem-sol ing
(Rahman & Wa anobe, 2023).
Howe e , he use o his ool also b ings po en ial nega i e e ec s and e hical
challenges. O e eliance on Cha GPT may unde mine s uden s’ independen hinking
abili ies and e en lead o academic misconduc , such as plagia ism and code
ab ica ion (Lieb enz e al., 2023). Meanwhile, da a p i acy, bias, inaccu acies, and
o e eliance on hese ools a e ecu ing hemes in cu en esea ch (Sallam, 2023).
These challenges no only a ec s uden s' lea ning p ocesses bu also ha e a -
eaching implica ions o educa ional equi y.
Cu en esea ch p ima ily elies on su eys, ocusing mo e on ends a he han
p o iding quali a i e analyses o s uden s' hough p ocesses (Alshwiah, 2024). I lacks
in-dep h explo a ion o s uden s' a ionale o using Cha GPT, which hinde s
educa o s’ unde s anding o s uden s' ac ual needs and beha io s. This s udy aims o
u he explo e his a ea by ocusing on he speci ic usage pa e ns and e hical issues
o Cha GPT in so wa e de elopmen educa ion, as epo ed by CS s uden s.
3 METHODOLOGY
To explo e he usage pa e ns o Cha GPT in so wa e de elopmen educa ion and he
e hical issues i aises, his s udy employed a quali a i e esea ch app oach, p ima ily
collec ing da a h ough semi-s uc u ed in e iews.
This da a collec ion me hod was chosen as i allows o an in-dep h explo a ion o
s uden s’ pe sonal expe iences and he beha io al logic unde lying hei use.
Compa ed o su eys, in e iews can cap u e mo e nuanced emo ions and a i udes,
pa icula ly when dealing wi h complex and mul i ace ed e hical issues (Denscombe,
2009). This me hod is be e sui ed o unco e ing s uden s’ au hen ic hough s and
conce ns (Bell & Wa e s, 2018).
In p e ious s udies, Dai (2024) used p ac ice obse a ion and design jou nals o eco d
s uden s’ beha io s wi h Cha GPT, while Guillén-Ypa ea and He nández-Rod íguez
(2024) u ilized Like -scale su eys o collec da a on s uden s’ a i udes owa d and
equency o GenAI use. Howe e , obse a ion me hods may ocus p ima ily on
ex e nal beha io s, while su eys may p o ide limi ed insigh in o he unde lying
mo i a ions and hough p ocesses o s uden s ega ding GenAI. By adop ing an
in e iew-based app oach, his s udy mi iga es hese limi a ions, p o iding iche and
mo e de ailed da a while allowing o lexible explo a ion o eme ging hemes.
The in e iews co e ed he ollowing aspec s:
• How s uden s epo using Cha GPT o comple e so wa e de elopmen asks.
• S uden s’ unde s anding and a i udes owa d he e hical issues po en ially
in ol ed in using Cha GPT.
The in e iews aimed o cap u e s uden s’ expe iences, s udy how hey epo
in e ac ing wi h gene a i e AI ools in hei lea ning p ac ices, and explo e he
unde lying logic and a i udes d i ing hese beha io s. The esea ch adhe ed s ic ly
o ins i u ional e hical guidelines o ensu e pa icipan p i acy, consen , and he op ion
o wi hd aw om he s udy [E hics e e ence numbe : 2025-0753-534].
The pa icipan s we e 6 s uden s majo ing in Compu e Science du ing hei bachelo ’s
s udies, and cu en ly en olled a a UK uni e si y. Thei academic backg ound
e lec ed uppe -le el unde g adua e aining in compu e science. They had comple ed
co e modules in da a science, a i icial in elligence, and so wa e enginee ing, wi h a
solid ounda ion in AI p inciples ele an o GenAI. Pa icipan s we e ec ui ed h ough
an open call and olun a ily consen ed a e ecei ing de ailed in o ma ion abou he
s udy. To minimise po en ial powe dynamics, nei he o he esea che s had any
eaching o supe iso y ela ionship wi h he s uden s. In e iewees we e selec ed o
hei amilia i y wi h Cha GPT in so wa e de elopmen and hei con ex ual
unde s anding o enginee ing educa ion.
The in e iews we e conduc ed ia online Zoom mee ings. Each in e iew las ed
a ound 30 minu es and co e ed s uden s’ expe iences o using Cha GPT, pe cei ed
bene i s and d awbacks, e hical conce ns, and sugges ions o u u e educa ional
applica ions. Be o e in e iews, pa icipan s signed consen o ms and we e b ie ed on
he s udy's pu pose, da a usage, and p i acy measu es. In e iews we e audio-
eco ded and ansc ibed.
The in e iew ansc ip s we e analysed using hema ic analysis o iden i y pa e ns
and e hical dilemmas associa ed wi h Cha GPT usage. Fi s , au ho 1 ansc ibed and
closely ead he in e iews o de elop amilia i y and ake ini ial no es. The nex s age
in ol ed coding and heme ex ac ion. Ini ial codes we e consolida ed, and key hemes
and sub hemes we e iden i ied by he i s au ho . Finally, he key hemes we e
in e p e ed and syn hesized in ela ion o he esea ch ques ions.
To ensu e he eliabili y o indings, he wo au ho s acked he coding and discussed
he assigned codes and he a ionale o coding speci ic exce p s o ensu e consensus
ac oss he code s’ in e p e a ions. A ocused IRR app oach was adop ed using one
code as e e ence, and calcula ed condi ional ag eemen . This app oach is
pa icula ly sui able when he objec i e is o ensu e eliabili y in iden i ying c i ical
codes (Campell, 2013). Among he 25 da a segmen s coded by he i s au ho , 23
we e also independen ly coded in o he same ca ego ies by he second au ho ,
yielding a pe cen ag eemen a e o 92%, which exceeds he commonly accep ed
h eshold o 80% (Bazen, 2021). Disc epancies be ween code s we e analyzed and
discussed, helping o e ine code de ini ions and enhance analy ical cla i y. In e code
eliabili y ensu es ha he coding p ocess is no idiosync a ic o a single esea che
and suppo s analy ical anspa ency and eplicabili y (Cole, 2024).
Conside ing he limi a ions o his s udy, he e is cu en ly no widely accep ed s anda d
o e alua ing AI con ibu ions, and he s udy elies on sel - epo ed da a, which may
in oduce subjec i e bias (Kasneci e al., 2023). Fo example, s uden s’ es ima es o
AI-gene a ed con en (e.g., 30%) a e based on pe sonal pe cep ion, a he han
objec i e measu es such as he ime spen using AI ools, he numbe o AI-gene a ed
ou pu s, o in e ac ion logs. Thi d, he s udy lacks mul iple da a sou ces o alida ion,
such as obse a ional da a o log analysis, which educes he obus ness o he
indings and may o e look impo an pa e ns. Fu u e esea ch could add ess his
limi a ion by inco po a ing mul iple da a sou ces o alida ion and de eloping objec i e
s anda ds o assessing AI con ibu ions. These imp o emen s would lead o mo e
eliable and comp ehensi e insigh s in o s uden s’ use o GenAI ools.
4 RESULTS
Th ough in e iews, his s udy iden i ies se e al key indings cen e ed a ound he
s a egic use o Cha GPT, i s e hical bounda ies, and i s implica ions o AI e hics
aining in enginee ing educa ion (Table 1). While some o hese indings align wi h
p io esea ch on s uden s pu suing enginee ing deg ees, compu e science s uden s
also exhibi ed unique cha ac e is ics in hei in e ac ions wi h Cha GPT.
Table 1. Key Findings on S uden Usage, Pe cep ions, and Sugges ions o Cha GPT
Ca ego y
Resea ch Findings
Ways o using
Cha GPT
S uden s use Cha GPT o ask analysis, b ains o ming, coding,
debugging, es ing, and epo w i ing.
Cha GPT p o oke con e sa ional lea ning, deepening unde s anding
and p o oking ideas.
Non-na i e speake s imp o e language exp ession o epo s.
S uden s p e e handling c ea i e and decision-making asks
independen ly, using Cha GPT only as a e e ence ool.
Balancing
independen
hinking and
dependency
S uden s limi Cha GPT’s con ibu ion (abou 30%) and e iew and
adjus i s ou pu s.
Only ew s udy and comp ehend AI-gene a ed codes line by line,
mos educe c i ical hinking when ou pu s seem co ec , neglec ing
in-dep h unde s anding.
S uden s a oid being limi ed by Cha GPT by discussing ideas in
eams o hinking independen ly be o e using i o e inemen .
E hics and isk
awa eness
S uden s s ongly oppose using Cha GPT ou pu s di ec ly,
conside ing i a iola ion o academic in eg i y.
S uden s emphasize acknowledging Cha GPT’s con ibu ions o
academic anspa ency.
Conce ns include p i acy isks, po en ial copy igh issues, and skill
deg ada ion om o e - eliance.
Sugges ions
De ine p ope Cha GPT usage, each e ec i e p omp design, and
encou age c i ical analysis.
Requi e clea acknowledgmen o Cha GPT’s in ol emen .
4.1 S a egic Use o Cha GPT
4.1.1 Cha GPT In eg a ion Ac oss he So wa e De elopmen Li ecycle
S uden s p ima ily used Cha GPT o asks such as cla i ying doub s abou
assignmen s, b ains o ming esea ch ideas, gene a ing code, p o iding explana ions,
es ing, debugging, and w i ing epo s. This pa e n aligns wi h he indings om o he
s udies (Champa e al., 2024; Akba e al., 2023).
S uden s indica ed ha he i s hing hey do upon lea ning abou an assignmen is o
use Cha GPT o help hem unde s and he ask equi emen s and p o ide ele an
backg ound knowledge. As in e iewee 4 s a ed: “I usually s a by discussing he
assignmen equi emen s om he ins uc o wi h he AI ool”. In e iewee 1 u he
explained ha i helped o quickly g asp co e concep s and his o ical con ex in a ield,
sa ing signi ican ime.
S uden s a e o ming a con e sa ional lea ning habi wi h Cha GPT, whe e in e ac ions
go beyond simple answe e ie al and inc easingly se e as a cogni i e ool ha
p o oked hei hinking. In e iewee 3 iewed Cha GPT as a hinking pa ne , no ing
ha i s explana ions o en o e ed new insigh s and deepened hei unde s anding o
he p oblem. In e iewee 5 u he s a ed: “This o en inspi es my hinking. Cha GPT
ends o conside wo s eps ahead, helping me b oaden my p oblem-sol ing
app oach.”
Non-na i e English-speaking s uden s also le e aged Cha GPT o language
op imiza ion, imp o ing he g amma o hei epo s. As in e iewee 2 no ed, i made
hei documen s look mo e p o essional and easie o unde s and.
Meanwhile, s uden s explained ha hey p e e o handle ce ain asks on hei own.
This includes c ea i e and decision-making ac i i ies such as so wa e a chi ec u e
design and complex algo i hm de elopmen , which equi e deep hinking and
o iginali y. In e iewee 2 highligh ed he impo ance o p oduc and a chi ec u e
design, no ing ha hese asks demand c ea i i y and hough ul conside a ion, wi h
Cha GPT only o e ing p elimina y sugges ions. In e iewee 6 explained: “C ea i e
and inno a i e asks—such as sol ing complex in e disciplina y p oblems o
p oposing no el scien i ic hypo heses— equi e lexible, human-s yle easoning and
in ui ion.” S uden s also emphasized he need o comple e o iginal wo k
independen ly, such as aking quizzes, o main ain academic in eg i y. In e iewee 3
emphasized ha academic in eg i y- ela ed asks such as o iginal pape w i ing,
e lec i e w i ing, and exams mus be comple ed independen ly by humans.
4.1.2 Balancing Cha GPT Assis ance and Independen Thinking
The in e iews e ealed ha s uden s we e mind ul o main aining bounda ies when
using Cha GPT, wi h i s con ibu ions o so wa e de elopmen p ojec s consis en ly
kep below 30% acco ding o sel - epo s. S uden s consciously and consis en ly
main ained c i ical hinking h oughou he p ocess.
S uden s did no ully us Cha GPT, belie ing i s gene a ed con en equi es u he
hough and e inemen , ea ing i as an ini ial p oposal o e e ence a he han a inal
solu ion. In e iewee 3 no ed: “A e ge ing i s ou pu , I always hink abou whe he i
eally i s my p ojec o my needs.” In e iewee 4 explained: “A single p oblem can
ha e many possible app oaches— he AI helps sugges hem, bu i ’s up o me o
choose and e ine one.” In e iewee 1 summa ized: “Cha GPT is mo e o an assis an ;
he inal code needs o be modi ied and e ined by me.”
Howe e , a no able conce n in he usage pa e ns is ha only less han hal o he
in e iewees epo ed ca e ully analysing and e iewing each line o AI-gene a ed
code o ensu e a ho ough unde s anding. Mos s uden s admi ed ha using Cha GPT
can make i ha d o main ain independen hinking. In e iewee 2 no ed ha i
Cha GPT’s esponses seem easonable and align wi h hei ideas, hey migh be
in luenced by i and s op hinking deeply abou ce ain issues. This eliance could lead
o o e looking he p ocess o c i ical hinking— o example, ocusing solely on sol ing
he p oblem wi hou unde s anding why he e o occu ed, and ailing o uly
in e nalize he p oblem-sol ing p ocess.
A he same ime, s uden s also ha e hei own s a egies o y o main ain
independen hinking. Fo example, in e iewee 5 desc ibed w i ing he own code i s ,
hen using Cha GPT o assess i s quali y and alignmen wi h ins uc ional goals.
In e iewee 1 emphasized ac i ely pa icipa ing in eam discussions and sha ing ideas
o e ine hei own pe spec i es and a oid being limi ed by Cha GPT’s iewpoin s.
4.2 E hical Bounda ies and Po en ial Risks in Cha GPT Usage
4.2.1 Awa eness o E hical Bounda ies
S uden s demons a ed a clea e hical s ance on using Cha GPT in academic se ings.
Fi s , s uden s explici ly opposed he di ec use o AI-gene a ed ou pu s. In e iewee
2 emphasized, "I you di ec ly use Cha GPT’s ou pu s, such as gene a ing a comple e
PPT o code wi h a single p omp , ha ’s de ini ely no app op ia e." Second, s uden s
s essed he impo ance o acknowledging Cha GPT’s con ibu ions. In e iewee 1
s a ed, "When p esen ing esul s, I clea ly explain ha Cha GPT p o ided ounda ional
ideas and e e ence in o ma ion, which I u he analysed, adjus ed, and inno a ed
upon o achie e my esea ch goals." This e lec s a commi men o academic in eg i y
and anspa ency in ecognizing ex e nal assis ance.
4.2.2 Po en ial Risks and Common Conce ns
Fo da a p i acy issues, s uden s exp essed conce ns abou Cha GPT’s cloud-based
ope a ion po en ially leading o sensi i e da a leaks. In e iewee 6 s a ed: “I’ e also
no iced ha AI e ains p e ious con e sa ions and may euse pe sonal de ails wi hou
explici consen .” S uden s exp essed ha hey a oided inpu ing pe sonal o
con iden ial da a in o Cha GPT o p e en po en ial da a leaks.
Rega ding code copy igh issues, hey no ed ha gene a ed code migh esemble
exis ing open o closed-sou ce code, posing isks o copy igh in ingemen (Lucchi,
2024). In e iewee 6 explained: “Relying en i ely on AI would no only iola e
academic s anda ds bu migh also unin en ionally in inge on o he s’ in ellec ual
p ope y, as AI-gene a ed con en is o en based on exis ing da a.” S uden s also
highligh ed he po en ial o skill deg ada ion, wa ning ha o e - eliance on Cha GPT
could e ode hei independen p oblem-sol ing and p og amming skills. As in e iewee
2 ema ked: “I Cha GPT suddenly wen o line, I would eally s uggle o comple e his
kind o assignmen om sc a ch.”
Finally, s uden s aised conce ns abou con ibu ion a ibu ion, exp essing con usion
o e how o quan i y AI’s indi ec con ibu ions. In e iewee 3 explained: “The solu ions
p o ided by Cha GPT inspi e my ideas, bu he inal implemen a ion is mine. I ’s ha d
o e alua e his indi ec con ibu ion.”
4.3 Implica ions o GenAI E hics T aining in Enginee ing Educa ion
S uden s sugges ed cla i ying Cha GPT’s ole and usage ules in educa ional se ings.
They p oposed de ining asks ha Cha GPT can assis and hose ha equi e
independen comple ion. In e iewee 5 poin ed ou : “Mos o my ins uc o s say AI can
be used as a suppo ool, bu he de ini ion o ‘suppo ’ is ague.” In e iewee 4 u he
s a ed: “I hink schools should clea ly de ine he ules o using AI: wha is and isn’
allowed.” Meanwhile, hey ecommended equi ing s uden s o speci y Cha GPT’s
in ol emen in epo s o uphold academic in eg i y. Fu he mo e, s uden s sugges ed
eaching how o c a e ec i e p omp s and c i ically analyse Cha GPT’s ou pu s.
DISCUSSION AND CONCLUSION
The analysis o in e iews e ealed how s uden s wi h CS backg ound a e s a egically
and e hically using Cha GPT o so wa e de elopmen assignmen s, le e aging i s
e iciency o epe i i e asks while main aining dominance o e c ea i e asks.
S uden s’ iews o e ed c i ical insigh s ha may con ibu e o cons uc ing a GenAI
e hics aining amewo k. This amewo k should cla i y usage bounda ies, p omo e
c i ical and independen hinking, s eng hen e hical awa eness, and p o ide echnical
aining o be e u ilize GenAI ools.
The indings o his s udy e eal a majo ans o ma ion in s uden s’ lea ning
app oaches due o he in eg a ion o Cha GPT. The adi ional "independen hinking-
coding p ac ice-debugging op imiza ion" lea ning model is g adually e ol ing in o an
"AI-assis ed hinking-in e ac i e p og amming-collabo a i e op imiza ion" model. This
ans o ma ion aligns wi h he indings o Rahman and Wa anobe (2023), who no ed
ha gene a i e AI is undamen ally eshaping adi ional lea ning ajec o ies.
S uden s applied Cha GPT ac oss mul iple s ages o so wa e de elopmen p ojec s.
Du ing he equi emen analysis s age, Cha GPT was used o supplemen in o ma ion
and expand esea ch pe spec i es. In he code a chi ec u e design s age, i p o ided
ini ial sugges ions, while co e design wo k emained a human esponsibili y. Du ing
code implemen a ion, Cha GPT se ed as a e e ence o code gene a ion. Fo code
debugging, i analysed e o s and sugges ed possible solu ions. Finally, in epo
w i ing, Cha GPT was used o enhance language cla i y and logical low.
Addi ionally, in e iewees gene ally ga e highe e alua ions o he e ec i eness o
Cha GPT and epo ed a highe equency o Cha GPT usage compa ed o he
a e age le els obse ed in p io esea ch (e.g., Dai, 2024). This inding aligns wi h he
pe spec i e ha su icien backg ound knowledge is necessa y o bene i om
Cha GPT (Dahe & Hussein, 2024). Rod igues e al. (2021) s a ed ha s uden s need
knowledge and skills o h i e in he digi al age, sugges ing ha he CS backg ound o
ou pa icipan s may ha e equipped hem wi h he echnical unde s anding needed o
le e age hese AI ools mo e e ec i ely. No ably, all he pa icipan s implemen ed sel -
egula ion s a egies o ensu e independen hinking when using Cha GPT,
consis en ly keeping i s con ibu ion below 30%, aligning wi h he c i ical awa eness o
AI ools desc ibed by Fa angi e al. (2024).
In e iewed s uden s exhibi ed e hical awa eness. S uden s demons a ed he abili y
o clea ly dis inguish be ween asks sui able and unsui able o AI ools and
spon aneously o med e hical guidelines conce ning da a p i acy and academic
in eg i y. This spon aneous e hical consciousness o e s a aluable e e ence poin o
building mo e sys ema ic AI e hics educa ion amewo ks.
Published esea ch iden i ies se e al e hical conce ns associa ed wi h Cha GPT.
These include hallucina ion, whe e i gene a es plausible bu inco ec in o ma ion
(S ahl & Eke, 2024); lack o o iginali y, as i s ou pu s a e ecombina ions o aining
da a; and bias, e lec ing imbalances in i s aining sou ces (Hua, Jin, & Jiang, 2024).
Addi ionally, p i acy isks a ise om la ge-scale da a use, while sus ainabili y
conce ns s em om he high inancial, en i onmen al, and human esou ce cos s o
AI sys ems (Mannu u e al., 2023). Compa ed o p io esea ch, s uden s in his s udy
displayed a highe le el o c i ical engagemen wi h Cha GPT’s ou pu s, e using o
accep hem unc i ically. As CS s uden s, hei echnical expe ise likely con ibu ed o
hei skep icism, allowing hem o ecognize he p obabilis ic na u e and easoning
limi a ions o la ge language models. While all pa icipan s men ioned issues such as
hallucina ion, o iginali y, and p i acy isks, he e we e no men ions o bias and
sus ainabili y. This sugges s he need o compu e science educa ion o expand
beyond echnical aspec s and place g ea e emphasis on sus ainabili y and
mac oe hical conside a ions, as sugges ed by G osz e al. (2019)
Se e al implica ions o enginee ing educa ion eme ge. The s udy d aws on s uden
iews and connec i o p io esea ch highligh ing he impo ance o e hical aining,
clea guidelines, and anspa ency o uphold academic in eg i y (Co on e al., 2023;
A las, 2023; Coope , 2023). Uni e si ies should p o ide all s uden s wi h ounda ional
knowledge o gene a i e AI o enable c i ical engagemen . Ta ge ed wo kshops on
p omp enginee ing, AI ou pu e alua ion, and p i acy conce ns could help s uden s
op imize AI usage while mi iga ing isks. This aining could be complemen ed by
in o ma ion on sus ainabili y issues pu po ing o he use o la ge language models
and how o e alua e and in e p e ou pu s as biased o non-biased. Clea academic
guidelines de ining Cha GPT’s ole in cou sewo k would p o ide s uc u ed
bounda ies o e hical use. Es ablishing ca ego ies o allowed, pa ially allowed, and
p ohibi ed AI usage in academic se ings, along wi h anspa ency equi emen s o
disclosing AI in ol emen , would enhance cla i y and in eg i y. Addi ionally, educa o s
could balance AI eliance wi h skill de elopmen by using AI-gene a ed con en as a
baseline and inco po a ing e lec ion epo s o encou age c i ical engagemen .
Toge he , hese insigh s poin owa d a mo e comp ehensi e and s uden -in o med
amewo k o suppo ing e hical and e ec i e use o AI in enginee ing educa ion.
