States must ensure that AI systems used in elections and referendums do not lead to unequal and discriminatory outcomes.

Article 14 ECHR specifically prohibits discrimination on grounds of “political or other opinions.” States should address the capability of AI to generate false information or lead to the exclusion of individuals or those who may be underrepresented or in a vulnerable situation from the democratic processes. States must also ensure that AI systems do not indirectly restrict or discourage the participation of specific groups from engaging in democratic processes.

Given the documented risks of algorithmic bias which may also include political affiliation, strong safeguards should be applied to ensure that such systems do not exclude, disadvantage or otherwise discriminate against particular groups thereby jeopardising the principles of impartiality and neutrality in the administration of the electoral process.[1] Accordingly, through the lifecycle of AI systems in electoral contexts, proactive measures, such as the use of inclusive and representative datasets when training AI systems, to identify and mitigate discriminatory outcomes should be implemented, ensuring that political affiliation or other protected characteristics do not affect enjoyment of the right to free elections or otherwise distort the democratic process.

 

[1] Sejdić and Finci v. Bosnia and Herzegovina [GC], No. 27996/06, 22 December 2009, § 44.