P ac ice Pape
Recommended ci a ion: Capde ila, I., Ui enho e, K. L., & Dehle Zu e ey, J.
(2025). Designing Ins i u ional Suppo o Gene a i e AI Adop ion: Building on
S em Teache s' Desi ed and Ac ual Use. 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.17631245.
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.
DESIGNING INSTITUTIONAL SUPPORT FOR GENERATIVE AI
ADOPTION: BUILDING ON STEM TEACHERS' DESIRED AND
ACTUAL USE
I. Capde ila a, K. Ui enho e b,1 , J. Dehle Zu e ey c
a EPFL, Lausanne, Swi ze land, h ps://o cid.o g/0000-0003-3997-9810
b EPFL, Lausanne, Swi ze land, h ps://o cid.o g/0000-0001-5450-3875
c EPFL, Lausanne, Swi ze land, h ps://o cid.o g/0000-0001-5163-807X
Au ho No e: Au ho s a e in alphabe ical o de and con ibu ed equally o he pape .
Con e ence Key A eas: Digi al ools and AI in enginee ing educa ion
Keywo ds: gene a i e AI, inno a ion adop ion
ABSTRACT
This p ac ice pape p esen s ou app oach o suppo ing acul y in he adop ion o
gene a i e a i icial in elligence (GenAI) in hei eaching a ou echnical uni e si y,
d awing on bo h empi ical e idence and heo e ical amewo ks. Fi s , we su eyed
acul y adop ion o GenAI, pa icula ly hei use o suppo ing s uden lea ning, along
wi h po en ial ac o s in luencing adop ion. Ou indings e ealed a subs an ial gap
be ween ac ual and desi ed use, wi h acul y exp essing signi ican ly g ea e in e es
in GenAI in eg a ion han hei cu en p ac ices e lec . Second, in o med by heo ies
on adop ion o inno a ion and change, we use his e idence o in o m a ge ed
suppo ini ia i es, designed no only o b idge he usage gap o acul y al eady
in e es ed in GenAI bu also o engage hose who cu en ly do no wish o use i , hus
aligning acul y needs wi h an e ec i e amewo k o inno a ion adop ion.
1 Co esponding Au ho
K. Ui enho e
kim.[email p o ec ed]
1 INTRODUCTION
In highe educa ion, GenAI is eme ging as a powe ul ool ha can assis wi h many
eaching asks and is apidly becoming widesp ead. A signi ican applica ion is he
use o GenAI o suppo s uden lea ning, allowing he c ea ion o new oppo uni ies
wi h well-documen ed lea ning bene i s. These include p o iding immedia e, scalable
eedback (e.g., Lee & Moo e, 2024), u o ing and answe ing s uden ques ions (e.g.,
Pa dos & Bhanda i, 2024), and s imula ing ac i e lea ning and me acogni i e
e lec ion (e.g., ElSaya y, 2024). Despi e he huge po en ial o GenAI o suppo
lea ning, s uden s using i spon aneously and wi hou pedagogical guidance a e a
isk o no bene i ing om his po en ial bu migh o load lea ning asks o GenAI
(e.g., Bas ani e al., 2024; Fan e al., 2024). Consequen ly, a key ques ion is how o
p omo e a pedagogically e ec i e and inno a i e use o GenAI ools by eache s.
1.1. Facul y Adop ion o GenAI Tools
A i udes and pe cep ions owa d GenAI’s ole in educa ion a e gene ally posi i e
(e.g., 2024; Kim e al., 2025). Fo example, in Kamoun e al. (2024), a majo i y saw
GenAI as an oppo uni y o pedagogical inno a ion, while a mino i y pe cei ed i as
a h ea o eache s’ jobs. Whe he GenAI in educa ion is pe cei ed as oppo uni y o
h ea is shaped by educa o s’ pedagogical o ien a ion (Cabellos e al., 2024; Choi e
al., 2023). Teache s wi h s uden -cen ed, cons uc i e o ien a ions end o iew
GenAI as an oppo uni y o en ich lea ning, while hose wi h con en -cen ed,
ep oduc i e app oaches a e mo e likely o see i as a h ea . Despi e gene ally
posi i e a i udes and pe cep ions, ac ual use is ypically low. Fo example, he
majo i y o acul y su eyed by Kim e al. (2025) epo ed in equen o no use o
GenAI ools.
Key ba ie s include insu icien knowledge and aining o implemen hese ools in
p ac ice (Choun a e al., 2022; An onenko & Ab amowi z, 2023; Al-Abdulla i , 2024),
and lack o clea ins i u ional guidelines (Kim e al., 2025). Fo example, he majo i y
o acul y su eyed by Kamoun e al. (2024) epo ed no ha ing he needed aining,
suppo , and esou ces o in eg a e Cha GPT in o hei eaching p ac ices. T aining
should ocus on p ac ical applica ions alongside heo e ical knowledge and AI
li e acy (e.g., Al-Abdulla i , 2024), as well as cons uc i e app oaches ha maximize
he po en ial o GenAI o suppo s uden lea ning (e.g., Cabellos e al., 2024). This
could inc ease AI li e acy and pe cei ed easibili y, associa ed wi h inc eased
adop ion o GenAI (Al-Abdulla i , 2024). In u n, mo e egula use has been
associa ed wi h eache s ecognizing GenAI’s educa ional alue, compa ed o
indi iduals wi h limi ed o no expe ience who see i mo e as a h ea (Cabellos e al.,
2024; Kaplan-Rakowski e al., 2023).
1.2. Theo ies on Adop ion o Inno a ion and Change
A ecen sys ema ic e iew by Belkina e al. (2025) examining 21 empi ical s udies
on he ac ual implemen a ion o GenAI in highe educa ion, ound ha deep
in eg a ion ede ining lea ning asks emains a e. Based on hei analysis, Belkina e
al. (2025) emphasize he impo ance o using s uc u ed pedagogical amewo ks o
suppo in eg a ion o GenAI ha is inno a i e as well as pedagogically sound.
Building on his, we a gue ha meaning ul in eg a ion needs o be suppo ed in a
holis ic app oach, ecognizing ha eache s, while cen al, a e no he only agen s
d i ing inno a ion and change. To be e explo e how GenAI can be implemen ed in
p ac ice, we d aw on wo well-es ablished heo ies ha concep ualize change no
only a he indi idual le el bu also in ela ion o social and sys emic dynamics. One
such amewo k, he Cul u al-His o ical Ac i i y Theo y (CHAT), concep ualizes
change as a esul o dynamic in e ac ions among six componen s o an ac i i y
sys em, iden i ying: he subjec ( eache s engaging wi h AI); he objec (s uden
lea ning), he ools (AI ools); he communi y (ins i u ional and pee in luences), he
ules (di ec i es and ecommenda ions), and he di ision o labo ( oles o he
di e en s akeholde s). The in e ac ion and con adic ions among hese componen s
c ea e oppo uni ies o lea ning and hus, change (Enges öm, 2001).
Complemen ing his pe spec i e, Di usion o Inno a ions heo y (DOI) p o ides a
lens o be e unde s and how inno a ion p og esses wi hin an ins i u ion. Roge s
(1983) de ines di usion as he p ocess by which an inno a ion is communica ed
h ough ce ain channels, o e ime among he membe s o a social sys em. Key
ac o s in luencing di usion include how he inno a ion is pe cei ed ( ela i e
ad an age, compa ibili y, complexi y, ialabili y and obse abili y), he di e en
adop e ca ego ies (inno a o s, ea ly adop e s, ea ly majo i y, la e majo i y, and
lagga ds), and he na u e o in o ma ion ans e , which is mo e e ec i e be ween
indi iduals who a e alike (pee s). DOI ou lines i e s ages in he inno a ion-decision
p ocess—knowledge, pe suasion, decision, implemen a ion, and con i ma ion—
which leads indi iduals o ei he adop ion o ejec ion o he inno a ion.
2 CONTEXT AND PRACTICAL WORK
As a echnical uni e si y, we a e discussing how o no me ely in eg a e, bu
mo eo e inno a e wi h GenAI in pedagogical p ac ice, all while emaining g ounded
in an e idence-based app oach. Making he b idge be ween collec ion o e idence,
in o ming ini ia i es, and implemen ing hose ini ia i es h ough ools and p ac ices
equi es he collabo a ion o di e en ins i u ional ac o s. The Cen e o Lea ning
Sciences (LEARN) collec s e idence on cu en educa ion p ac ices and in luencing
ac o s. The Teaching Suppo Cen e (CAPE) p omo es e idence-based e ec i e
pedagogies o lea ning among eache s. The Cen e o Digi al Educa ion (CEDE)
designs p ac ical ools and s udies hei e ec i eness. Meanwhile, mul iple in-house
esea ch labs in es iga e GenAI i sel and explo e i s inno a i e applica ions.
In his p ac ice pape , we aim o in o m ini ia i es ac oss ou ins i u ion wi h he oice
o a key ac o : he eache s. This wo k-in-p og ess pape p esen s he collec ed
e idence on he use o GenAI by ou acul y in hei eaching p ac ice, as well as
cha ac e is ics associa ed wi h he adop ion o GenAI. We discuss he implica ions
h ough he lens o adop ion and inno a ion change (CHAT and DOI), o sugges
a ge ed ini ia i es o suppo e ec i e and inno a i e use o GenAI by eache s.
2.2. The Su ey
We conduc ed a su ey du ing he au umn semes e o 2024 in i ing all eache s a
ou ins i u ion (N=1’473) o sha e hei pe cep ions, usage, and needs ega ding
GenAI in educa ion. I was adminis e ed in English and ecei ed a o al o 109
esponses om all 18 sec ions. We conside ed di e en eaching applica ions,
including using AI ools o designing eaching ma e ial (e.g., c ea ion o exe cises,
quizzes, slides, syllabi), suppo ing s uden lea ning (e.g., cha bo s ha p o ide
eedback du ing assignmen s, answe ques ions, o help s uden s moni o hei
lea ning), and e alua ing s uden pe o mance (e.g., iden i ying s eng hs and
weaknesses in essays, assessing accu acy on es s). We uniquely dis inguished
ac ual (“You al eady use gene a i e AI ools o hese asks.”) and desi ed (“You
would like o use gene a i e AI ools o hese asks.”) use o GenAI, epo ed on a 9-
poin a ing scale, whe e 0 ep esen ed "No A All" and 8 ep esen ed "Ve y Much".
Fu he mo e, we d ew upon es ablished echnology accep ance amewo ks (e.g.,
Da is, 1989) o iden i y ac o s ha migh in luence bo h ac ual and desi ed use and
o o mula e speci ic ques ions ele an o GenAI adop ion in eaching. We designed
speci ic unde s anding and skill ques ions o assess a ia ions in p o iciency among
a STEM eache coho ha is p esumed o ha e compa a i ely high GenAI
knowledge and compe ence. Responses we e collec ed on 5-poin Like scales
( om “No a all” o “Ve y much”, excep when s a ed o he wise):
● Unde s anding: How well do you unde s and GenAI? (Mode a ely: Key
p inciples and mechanisms o how and why GenAI wo ks. Ve y well:
Ad anced knowledge o neu al ne wo k a chi ec u e unde lying GenAI).
● Skill: How skilled a e you in using GenAI? (Mode a ely: Com o able using
mul iple GenAI ools ac oss di e en asks. Ve y skilled: P o icien in using
APIs, ine- uning models, and de eloping cus om applica ions).
● Use ulness: Can hese ools be use ul o suppo ing you eaching ac i i ies?
● Feasibili y: Is i easible o in eg a e hem in you eaching p ac ice? Conside
you ime and esou ces.
● Con ol: Can hey educe con ol and au onomy o eache s?
● Th ea : Can hey be h ea ening o eache s' jobs?
● Encou agemen : To wha ex en do you eel discou aged o encou aged a he
ins i u ion o inco po a e GenAI ools in o you eaching p ac ice?
Finally, we asked eache s wha ype o suppo hey need wi h GenAI in educa ion,
and ga e hem se e al op ions o choose om, including (1) dedica ed IT
in as uc u e; (2) policy and ules ("mus do"); (3) guidelines and bes p ac ices
("should do"); (4) in o ma ion and p ac ical examples ("could do"); (5) ou -o - he-box
solu ions; (6) wo kshops and aining; and (7) in-pe son consul a ions.
3 RESULTS AND INSIGHTS
3.1 Ac ual and Desi ed Use o AI Tools
Among he di e en eaching applica ions, he highes ac ual use o GenAI ools was
epo ed o designing eaching ma e ials (M = 2.03, SD = 2.38), ollowed by
suppo ing s uden lea ning (M = 1.32, SD = 2.16), while he lowes ac ual use was
obse ed o e alua ing s uden pe o mance (M = 0.52, SD = 1.23). Figu e 1 shows
a subs an ial gap be ween ac ual and desi ed use. Ac oss di e en applica ions,
86.3% o eache s exp essed a desi e o use GenAI ools o a g ea e ex en han
hey cu en ly do. In his p ac ice pape , we ocus speci ically on eache s' use o
GenAI o suppo s uden lea ning. In Figu e 2, we see ha e en hough 81.7% o
eache s wish o in eg a e GenAI o suppo s uden lea ning ( epo ed desi ed use
le el > 0 “No a all”), abou hal o hem do no ac ually use hese ools ye ( epo ed
ac ual use le el 0).
Fig. 1. Ac ual s. Desi ed Use (0 co esponding o
“No a all”, 8 co esponding o “Ve y much”).
Fig. 2. Teache G oups (“No”
co esponding o answe “No a
all”, “Yes” co esponding o all
o he esponses).
3.2 Fac o s In luencing GenAI Adop ion
To p o ide adap ed suppo , i is c ucial o iden i y hinde ing and acili a ing ac o s
associa ed wi h GenAI adop ion among eache s, conside ing bo h hei desi e o
use GenAI ools and hei ac ual usage. The e o e, we compa ed sel - epo ed
cha ac e is ics o eache s con as ing hem on wo le els: (i) be ween hose who
desi e and hose who do no desi e o use GenAI ools o suppo s uden lea ning,
and (ii) among hose who desi e o use GenAI, be ween hose who al eady use
GenAI o suppo s uden lea ning s hose who do no (see Table 1 o desc ip i e
s a is ics). Missing da a (M = 4.18% pe a iable) we e p edic ed and impu ed based
on o he a iables, p e en ing a 24.77% case loss.
We conduc ed a Bayesian mul i a ia e analysis in R (b ms) using ou Ma ko chains
wi h 4,000 i e a ions each (including 2,000 wa m-up i e a ions) o es ima e g oup
e ec s on s anda dized cha ac e is ics, modelled as Gaussian ou comes while
accoun ing o co ela ions be ween hem. As a esul , he pos e io mean
di e ences epo ed in he ollowing pa ag aph a e exp essed in s anda d de ia ion
uni s (unlike he able, which p esen s means in hei o iginal scales), and ep esen
he expec ed di e ences be ween eache s who desi e using GenAI ools and hose
who do no . S a is ically meaning ul di e ences we e iden i ied when he 95%
c edible in e als excluded ze o. Model diagnos ics con i med excellen con e gence
(Rha = 1.00 o all pa ame e s) and high e ec i e sample sizes, ensu ing obus
es ima es.
The pos e io mean di e ences indica e ha eache s who desi e using GenAI ools,
compa ed o hose who do no , pe cei ed hem as mo e use ul (1.11, 95% CI [0.66,
1.55]), had a mo e posi i e AI a i ude (0.99, 95% CI [0.52, 1.45]), and ound i
mo e easible o in eg a e (0.62, 95% CI [0.13, 1.09]). Smalle bu s ill meaning ul
posi i e di e ences we e ound o skill (0.57, 95% CI [0.09, 1.05]) and
encou agemen (0.55, 95% CI [0.05, 1.03]). They also pe cei ed GenAI as less
h ea ening (-0.59, 95% CI [-1.08, -0.09]) and less de imen al o hei con ol (-0.52,
95% CI [-1.01, -0.03]).
Taking he mo i a ional aspec ou , he u he analysis looks a only hose eache s
who desi e o use GenAI ools. Among hem, eache s who ac ually use GenAI ools
o s uden lea ning epo ed highe skill (0.55, 95% CI [0.14, 0.92]), g ea e
easibili y o in eg a ion (0.47, 95% CI [0.07, 0.87]), and mo e use ulness (0.41,
95% CI [0.04, 0.77]) han non-use s.
Table 1. Cha ac e is ics o Teache G oups compa ing desi ed and ac ual use o GenAI o
suppo s uden lea ning.
No Desi e
Desi e
Desi e, No
Use
Desi e, Use
m
sd
m
sd
m
sd
m
sd
AI a i ude
32.11
11.46
41.91
8.90
40.35
9.65
43.74
7.64
Unde s anding
2.58
1.02
2.23
0.96
2.02
1.04
2.47
0.80
Skill
1.16
1.26
1.74
0.97
1.50
0.96
2.03
0.91
Use ulness
1.26
1.24
2.58
1.02
2.36
0.99
2.84
1.01
Feasibili y
1.37
1.42
2.08
1.14
1.80
1.10
2.42
1.11
Con ol
2.38
1.31
1.79
1.00
1.88
1.10
1.68
0.88
Th ea
1.89
1.52
1.12
0.99
1.07
1.04
1.18
0.94
Encou agemen
-0.07
0.70
0.36
0.75
0.39
0.70
0.33
0.81
No e. Red highligh s s a is ically meaning ul di e ences.
3.3 Suppo Needs
As illus a ed in Figu e 3, he mos equen ly eques ed ype o suppo ac oss all
eache s was guidelines and bes p ac ices, indica ing a clea demand o s uc u ed
ecommenda ions. Addi ionally, eache s who had no ye begun using GenAI
exp essed a pa icula ly s ong need o in o ma ion and p ac ical examples. Teache s
who do no desi e o use GenAI show signi ican ly lowe demand o a ious suppo
o ma s.
Fig. 3. Suppo needs exp essed by eache s, g ouped acco ding o hei desi ed and ac ual
use o GenAI o suppo s uden lea ning.
4 IMPLICATIONS
Conside ing adop ion ac o s iden i ied in he eache su ey as well as key p inciples
om heo e ical amewo ks o inno a ion, we p opose a s uc u ed se o ini ia i es.
We aligned hem wi h he i e s ages o Roge s’ inno a ion-decision p ocess,
ensu ing suppo o educa o s a all le els o adop ion. While some o hese
ini ia i es a e al eady implemen ed a ou ins i u ion and could be u he e ined,
o he s a e s a egic ecommenda ions aimed a encou aging GenAI adop ion among
ou acul y.
S age 1, Knowledge: Ins i u ional and P ac ical Guidelines. This ini ia i e
in ol es he co-cons uc ion o ins i u ional guidelines b inging oge he mul iple
s akeholde s. These can be complemen ed by ideo-capsules o ac shee s
illus a ing conc e e examples, bes p ac ices and es imonials o EPFL eache s.
This app oach inc eases awa eness o he use ulness and easibili y o hese
p ac ices and in o ma ion is ans e ed ia pee s, a key aspec o adop ion in bo h
heo ies. Mo eo e , i ackles se e al componen s o he CHAT heo y os e ing he
mul i- oicedness o he ac i i y sys em. Finally, his ini ia i e esponds o he suppo
needs indica ed by all EPFL eache s, independen ly o hei desi ed and ac ual use.
S age 2, Pe suasion: In o mal Pee Exchanges. This ini ia i e aims o p o ide a
sa e space o eache s o egula ly mee and discuss, sha e di e en iews, ideas,
expe iences and conce ns abou GenAI. To p omo e meaning ul exchanges, he
sessions a e semi-s uc u ed wi h one eache p esen a ion, ollowed by a discussion
whe e di e se poin s o iew a e encou aged. This ini ia i e aligns wi h CHAT and
DOI o os e pee in e ac ions and con adic ions o lea n and inno a e. While all
ypes o adop e s can lea n om pee s in a low-s akes con ex , he mos scep ical
and conse a i e (la e majo i y and lagge s) can be pa icula ly pe suaded o y new
hings in hei p ac ice.
S age 3, Decision: Hands-On Wo kshops. This ini ia i e ocuses on he
de elopmen o s uc u ed, pedagogically in o med wo kshops o p o ide eache s
wi h he oppo uni y o explo e, expe imen wi h, and e alua e he in eg a ion o
GenAI in a low-s akes en i onmen . The wo kshop ca alogue is designed o
accommoda e he di e en ypes o adop e p o iles and also aligns wi h he di e en
s ages o he inno a ion-decision p ocess. To maximize e ec i eness, each
wo kshop should inco po a e h ee essen ial componen s: (i) hands-on ac i i ies
enabling di ec engagemen wi h GenAI ools; (ii) pee discussion, o os e
communi y, and (iii) guided e lec ion o suppo he ans e o insigh s in o
pa icipan s’ own eaching p ac ices. This ini ia i e enhances con idence and
compe ence and encou ages eache s ha desi e o use GenAI bu don’ cu en ly
do i , o ake he decision o adop i in hei cou ses.
S age 4, Implemen a ion: Facili a ing Access o GenAI Tools. Tools a e one o
he co e elemen s o he CHAT, and p o iding access o sa e, eliable ools ha mee
eache s' pedagogical needs is a p e equisi e o adop ion. By p o iding GenAI ools
(e.g., cha bo s) ha can be easily cus omized, h ough e ie al-augmen ed
gene a ion (RAG) and sys em ins uc ions ha de ine a speci ic pedagogical ole
(e.g., u o , ole-playing pa ne , men o ), we can di ec ly align hese ools wi h
eache s’ pedagogical needs, hus inc easing he pe cep ion o use ulness and
easibili y. Addi ionally, o e ing he op ion o build hese applica ions using locally
hos ed models ensu es ha da a ne e lea e he ins i u ion, hus add essing da a
p o ec ion conce ns ha cu en ly hinde o e en p e en hei implemen a ion and
use in pedagogical p ac ice.
S age 5, Con i ma ion: GenAI in Educa ion (AInEd) Labo a o ies. This ini ia i e
aims o c ea e a space and amewo k whe e inno a i e p ac i ione s and in-house
esea che s ac i ely wo king on GenAI inno a ion can collabo a i ely esea ch and
expe imen wi h he po en ial o his echnology in suppo ing s uden lea ning. I is
inspi ed by he CHAT Change labo a o ies (Vi kkunen & Newnham, 2013) which a e
concei ed o igge cycles o expansi e lea ning including simula ion and
conc e isa ion. In pa icula , acul y who al eady use GenAI can bene i om his
ini ia i e o con i m and u he de elop adop ion o GenAI in hei eaching.
By in eg a ing insigh s om CHAT and DOI heo ies wi h empi ical su ey da a,
hese ini ia i es o e a s uc u ed and ans e able amewo k o unde s anding and
suppo ing GenAI adop ion in STEM educa ion. Ou p esen ed e idence- o-p ac ice
wo k low, su ey ins umen (a ailable on OSF), and p oposed suppo ini ia i es,
can be di ec ly applied o STEM educa ion as well as o he highe educa ion
ins i u ions.
5 ACKNOWLEDGEMENTS
We acknowledge he aluable con ibu ions o Pa ick Je mann, Head o CEDE, o
his expe ad ice in designing he eache su ey, and Roland To mey, Head o
CAPE, o his insigh ul pe spec i es on he concep ualisa ion o his p ac ice pape .
