Five phases from planning to implementation 5 Implementation and supervision Investigate whether biases in the system can be levelled out by compensatory 5 1 measures. Human control of the system must be ensured, and knowledge of the system made known to those affected, so that they can protect their rights. Figure 1 1 Planning Define formal and intended use of the AI system. The context in which the system is to operate is decisive for what significance it may have on an individual and societal level. 4 Testing Map whether the system works equally precisely for all groups. Even if personal data is removed, the system can find patterns in the data so that information that coincides with the grounds for dis­crimination is illegally emphasised. 2 4 3 3 Model development The system’s calculation must coincide with the overall formula. The variables the system’s calculations are based on must be relevant. Pavis factual formali­ty, necessity and proportionality for vari­ables that can be linked to a basis for discrimination. 2 Training data The data base model is trained to be representative based on the model’s functionality, and historical biases must be mapped and corrected. Source: New Norwegian Guide to Prevent AI Discrimination Launched With Minister of Digitalisation| by Alex Moltzau| Ethical AI Resources| Medium(Moltzau 2023) impact on important choices determining the way the AI-system is constructed, what data is used, and what im­plications the system may have for affected groups. Unfor­tunately, what too often happens is that systems are bought ready-made, and critical questions not asked before the system is already in operation, and sometimes not even then. This may be partly due to lack of knowledge and compe­tence at all levels in the demand-supply chain. The AI Literacy Gap The EU AI Act obliges providers and deployers of AI sys­tems to ensure that providers and deployers of AI sys­tems have a sufficient level of technical knowledge, ex­perience, education and training. AI literacy is defined as the“skills, knowledge, and understanding” needed to deploy AI systems responsibly(Gulley 2025). Research shows women tend to adopt AI tools at dra­matically lower rates than men, often due to concerns about ethics, judgment within the workplace, and a per­ceived lack of expertise. Ensuring that digital literacy ed­ucation is inclusive and responsive to such gender dy­namics is essential for raising awareness and empower­ing all groups(Blending 2025). A study involving 1,026 software engineers found that when women and men use AI to produce identical work, women are perceived as less competent than men. On average, engineers believed to have used AI were rated 9% less competent, but women who used AI faced a 13% reduction in perceived competence compared to a 6% reduction for men. The“competence penalty” against women using AI is particularly harsh when evaluated by male engineers who themselves do not use AI—they rate female AI us­ers 26% more harshly than male AI users for the same output(Travis 2025). Despite company efforts to encourage AI adoption in software development, only 31% of female engineers used the AI tool versus 41% of all engineers, likely due to concerns about this competence bias. Baseline knowl­edge and confidence levels varies considerably and there is a gender gap that needs to be bridged by building AI-literacy among women(Gulley 2025). Research also indicates that the differences in low-cost and premium options for AI are furthering inequality among white collar workers. Based on a survey of 4,000 knowledge workers across the UK, US, Germany, and Canada, the study reveals that higher earners have dis­proportionate access to the latest AI tools and training, 6 Friedrich-Ebert-Stiftung e.V.