AI and GDPR
Use of discriminatory algorithms in connection to the processing of personal data
The use of discriminatory algorithms in connection to the processing of personal data is prohibited. This is important to keep in mind when developing AI models and using AI tools.
Principle of fairness: Discrimination is prohibited
The principle of fairness means, inter alia, that the processing of personal data must not be discriminatory. In other words, there must be fair, reasonable and proportionate processing of personal data in relation to data subjects.
It is not always easy to detect discriminatory algorithms and often the use of discriminatory algorithms occurs unconsciously.
Problems that may arise when using AI-based recruitment systems
A practical example of when discriminatory algorithms may occur when using an AI tool is if a company uses an AI-based recruitment system to analyse former employees in order to find new suitable candidates. Many workplaces are overrepresented by men, which may mean that the system favours male candidates.
What is important for companies to consider regarding discriminatory algorithms when developing and using AI models?
Companies developing AI models must take sufficient technical and organisational measures to ensure that the AI model complies with the rules of the GDPR. This includes ensuring that the use is in accordance with the principle of fairness, such as the use of non-discriminatory algorithms by the AI model. Therefore, it is important to make thorough assessments before launch, as well as regularly thereafter, to ensure that personal data is relevant and properly processed. In addition, it is good to constantly test the AI model so that, for example, it does not discriminate against groups of individuals or individuals.
Learn more
What legal bases may be appropriate to use for the development or use of AI models?
In the GDPR, there are a total of six (6) legal bases on which companies can support a processing. Each individual processing of personal data must be supported by a legal basis. Legitimate interest is the most flexible and common legal basis to use in the development or use of AI models. Consent as a legal basis can be difficult to use when developing AI models, but be more applicable when using AI tools that already exist. Contracts with data subjects may also be an appropriate legal basis in certain cases. The appropriate legal basis in each individual case depends on the circumstances of the specific case.