AI competence at universities — mandatory from February 2, 2025

The EU AI Reg­u­la­tion also stip­u­lates a num­ber of oblig­a­tions for uni­ver­si­ties. First and fore­most is the acqui­si­tion of AI skills for uni­ver­sity staff and other per­sons.

This is the fourth and final post in our blog series “AI in higher edu­ca­tion”. After the first post was about AI detec­tors, we then looked at the legal aspects of AI and OER. The third post dealt with prompt tips for high-qual­ity OER. The series is now com­plete with the fol­low­ing arti­cle on manda­tory AI skills.

Obligations for universities under the AI Regulation

On August 1, 2024, the AI Reg­u­la­tion(AI Act (EU) 2024/1689) came into force and applies directly in the mem­ber states with­out the need for trans­po­si­tion into national law. The AI Reg­u­la­tion defines AI appli­ca­tion sce­nar­ios that are pro­hib­ited from the out­set and those that are likely to be asso­ci­ated with a high risk and there­fore require spe­cial care.

The KI-VO also cov­ers the GPAI sys­tems of providers such as Ope­nAI, which are acces­si­ble at many uni­ver­si­ties for teach­ing and study pur­poses via an inter­face of the uni­ver­sity net­work, as well as the LLM sys­tems oper­ated on their own servers. The uni­ver­sity assumes the role of oper­a­tor or provider within the mean­ing of the KI-VO. Art. 4, Art. 50, Art. 53 and Art. 55 of the AI Reg­u­la­tion impose oblig­a­tions on providers and oper­a­tors of GPAI mod­els. These are essen­tially infor­ma­tion and trans­parency oblig­a­tions, oblig­a­tions regard­ing risk and data pro­tec­tion man­age­ment, cyber­se­cu­rity, com­pli­ance with EU copy­right law and the pro­vi­sion of AI exper­tise.

Vio­la­tions of the oblig­a­tions imposed by the AI Reg­u­la­tion are sub­ject to heavy fines. Art. 113 of the AI Reg­u­la­tion pro­vides for a stag­gered start date for the oblig­a­tions to be com­plied with in order to facil­i­tate imple­men­ta­tion for the par­ties con­cerned.

The first applic­a­ble oblig­a­tion is the teach­ing of AI lit­er­acy. This must be imple­mented by Feb­ru­ary 2, 2025. What does this mean for uni­ver­si­ties? Uni­ver­si­ties must take com­pe­tence-build­ing mea­sures for the acqui­si­tion of AI com­pe­tence by staff and other per­sons involved in the oper­a­tion and use of AI sys­tems on their behalf. Mea­sures include work­shops or e‑learning courses. In addi­tion to uni­ver­sity employ­ees and mem­bers of the uni­ver­sity who are not employed by the uni­ver­sity, such as stu­dents, the per­sons to be trained will gen­er­ally also be lec­tur­ers.

What AI skills need to be taught?

Accord­ing to Art. 3 No. 56 of the AI Reg­u­la­tion, AI com­pe­tence com­prises the skills, knowl­edge and under­stand­ing that enable providers, oper­a­tors and data sub­jects, tak­ing into account their respec­tive rights and oblig­a­tions under this Reg­u­la­tion, to use AI sys­tems in an informed man­ner and to be aware of the oppor­tu­ni­ties and risks of AI and the poten­tial harm it can cause. Accord­ing to Art. 4 of the AI Reg­u­la­tion, the tech­ni­cal knowl­edge, expe­ri­ence, edu­ca­tion and train­ing, the con­text in which the AI sys­tems are to be used and the per­sons or groups of per­sons with whom the AI sys­tems are to be used must be taken into account.

Accord­ing to Wienrich/Carolus/Markus/Augustin, AI-com­pe­tent peo­ple should

• Rec­og­nize and iden­tify AI: They should dis­tin­guish between dif­fer­ent forms of AI and under­stand at least the basic prin­ci­ples of how they work.
• Know and fol­low eth­i­cal prin­ci­ples that enable the respon­si­ble inte­gra­tion of AI sys­tems: you should be able to reflect on the poten­tial dan­gers of AI and ini­ti­ate coun­ter­mea­sures if nec­es­sary.
• Be legally com­pe­tent: You should be famil­iar with the laws gov­ern­ing use with AI and have gen­eral data pro­tec­tion knowl­edge.

What implementation aids are available?

The EU Arti­fi­cial Intel­li­gence Office(Euro­pean AI Office) has announced that it will pub­lish prac­ti­cal guide­lines on the imple­men­ta­tion of oblig­a­tions for gen­eral pur­pose AI sys­tems (GPAI sys­tems) on the inter­net. The Euro­pean Data Pro­tec­tion Super­vi­sor (EDPS) has already pro­vided guid­ance on ensur­ing data pro­tec­tion com­pli­ance when using gen­er­a­tive AI sys­tems on the inter­net.

twillo already offers a work­shop on AI skills. The focus here is on the cre­ation of edu­ca­tional mate­ri­als, in par­tic­u­lar OER, with the help of AI. Didac­tic and legal aspects, such as copy­right, are dis­cussed in the work­shop. If you are inter­ested, please con­tact info@twillo.de to.