Is Openness in Decline? Da a
Sha ing Be ween AI Commons,
and P eda o y Cap u e
Ka ja Maye (Uni e si y o Vienna)
S e an Skupien (Be lin Uni e si y Alliance)
Jochen Knaus (Weizenbaum Ins i u )
New “old” challenges in he “AI e a” o Open Science?
AI Con ex
●ML and GenAI ely on publicly sha ed o sc aped knowledge
●Bu esul ing models and in as uc u es a e closed and co po a e-con olled
AI Commons unde P essu e
●Open laye s o he AI s ack (da ase s, models, amewo ks) depend on p eca ious, olun a y labou
●Few public o independen unde s sus ain hese in as uc u es
●Co po a e ac o s domina e main enance, di ec ion, and access
P eda o y Cap u e
●Open da a and code eused wi hou con ex o c edi
●Rein o ces bias, opaci y, and concen a ion o powe in p op ie a y sys ems
Resea che Pe spec i es
●G owing cau ion among long- ime ad oca es o openness
●Conce ns abou ci cula ion, misuse, and loss o con ol in comme cial AI en i onmen s
The 3 Ques ions o Today
1. You expe ience: Is openness becoming
ha de o pu sue in he age o AI?
2. Wha o ms o ecogni ion o ins i u ional
suppo would help?
3. How could new models o openness in he
age o AI look like?
This session:
●15-min in oduc ion aming issues
●3 ×10-min discussion ounds on guiding ques ions
●Collec i e insigh s documen ed ia li e polls & online boa d
→ Documen a ion ia collec i e pad, please add you name o he
au ho s lis zbw. o/osc25-pad03
C ea ing a space: “Yes, we’ e open!?” (Be lin, 25.3.2025)
- E en o b ing oge he
science, economy, ci il
socie y, and public
se ice, and poli ics
- 2 Panel discussion and
8 wo ld-ca e-sessions
- Aim was o explo e
expe iences and
needs o os e AI ha
includes and suppo s
Open Science ac oss
domains
C ea ing a space: “Yes, we’ e open!?”: Key indings
Resul s: Ou sha ed ision is a socially accep ed digi al cul u e in which AI echnologies a e designed o be
anspa en , ep oducible, sus ainable, and o ien ed owa d he common good. Key guiding p inciples a e based
on open science.
Simon B unel (Limo o Resea ch)
C ea ing a space: “Yes, we’ e open!?”: Key indings
Fou key s a egic a eas:
1. Sha ed knowledge spaces and s a egic
ne wo king,
2. Skills de elopmen and educa ion,
3. Sus ainable de elopmen h ough open
digi al in as uc u es,
4. Go e nance: Combining esponsibili y
and inno a ion.
Discussion Pape :
h ps://www.doi.o g/10.34669/WI.DP/51
Policy Pape :
h ps://www.doi.o g./10.34669/WI.PP/15
Backg ound II: P ojec Poli ics o Openness
Open Da a P ac ices in he Compu a ional Social Sciences
Funde : FWF Elise Rich e Fellowship
Du a ion: 2019–2025
PI: D . Ka ja Maye
Loca ion: Uni e si y o Vienna
Aim:
In es iga e how openness is en isioned, nego ia ed, and enac ed in he da a p ac ices o compu a ional social science.
Empi ical Fields:
●Ci izen science
●Da a in as uc u es
●Social media esea ch and da a science
Me hods: quali a i e in e iews, G oup discussions, pa icipa o y obse a ion
Quo es: Open Science unde new p essu e in he GenAI e a
•We designed ou eposi o y o human access and schola ly euse, no o bo s pulling e aby es o e nigh . The sc aping
by AI de elope s is o e whelming ou se e s and ou s a . (In as uc u e manage , esea ch da a eposi o y, 2024)
In as uc u es s ained by la ge-scale sc aping
Machine- eadabili y is supposed o make da a mo e ‘FAIR,’ bu in p ac ice i o en la ens hem. Con ex ual me ada a,
me hodological no es, e en unce ain y - all ge los in ansla ion o s anda dized o ma s (Social Scien is , 2023)
Machine- eadabili y imp o es in e ope abili y bu s ips con ex and quali a i e me ada a
•As soon as ou open da ase is used o model aining, i disappea s in o a black box. We don’ e en know i ou
a ibu ion s ays a ached, le alone how he da a a e in e p e ed (Open da a p ac i ione in he social sciences, 2023)
Once da a en e s p op ie a y AI pipelines: o e sigh and ecogni ion o en los
•AI esea ch exposes how agile ha idea has become. The complexi y o models, dependencies, and compu e
en i onmen s makes ull ep oducibili y p ac ically impossible —e en o hose who wan o do i igh . (Da a Scien is ,
2024)
See also: Hosseini e al. 2025
Limi s o openness o ep oducibili y: agile in AI due o hype pa ame e s, en i onmen s, hidden dependencies
Yes, openness is messy and some imes misused, bu closing o knowledge isn’ he answe . We need sma e o ms o
openness —ones ha build us , accoun abili y, and eal collabo a ion. (Da a s ewa d and Open Science ad oca e,
2023)
Balancing Openness and Responsibili y
Mohammad Hosseini, Se ge P. J. M. Ho bach, K is i Holmes, Tony
Ross-Hellaue ; Open Science a he gene a i e AI u n: An explo a o y
analysis o challenges and oppo uni ies. Quan i a i e Science S udies
2025; 6 22–45. doi: h ps://doi.o g/10.1162/qss_a_00337
Ques ion 1: Collec ing Expe iences
In you own esea ch o p ac ice, ha e you no iced ha
openness is becoming ha de o pu sue - o e en coming
unde a ack - in he age o AI? Fo example, ia
in as uc u al s ain, loss o con ex in pu suing be e
machine- eadabili y, lack o ecogni ion o opening
knowledge, challenges in go e ning da a euse, o he
agili y o ep oducibili y in AI/ML wo k lows?
Slido esul s
Slido esul s
Recap/W ap + Ou look
How will we use he ou comes o his session?
●Please add you names, no es and eedback o he sha ed Google documen
zbw. o/osc25-pad03
●We will publish a sho blog pos summa izing oday’s discussion
●The insigh s will in o m ou ongoing policy wo k on Open Science and AI
●We will in i e you o ake pa in implemen ing some o he p oposed ac i i ies
●To s ay in o med, please con ac us a open-science@be lin-uni e si y-alliance.de
