Two researchers from the University of Deusto, in Spain, published a study revealing how artificial intelligence can manipulate decisions in the digital world. Psychology scholars Ujué Agudo and Helena Matute assessed how easily humans can be influenced by algorithms in their political positions and relationship preferences.
The researchers developed a battery of four tests, with results showing that manipulation by algorithms is possible. Furthermore, they concluded that “certain models of persuasion (explicit or indirect) are more efficient depending on the context (politics or dating)”.
To prepare participants, all four tests began with a fake personality quiz, which was supposed to guide the algorithms. In reality, the AI operated with the same user profile for everyone. The instructions the program received were purposefully vague, meaning they could fit anyone.
Political influence works more directly
The first two tests specifically assessed how AI can manipulate political decisions. In the first experiment, participants viewed images and profiles of fictional politicians and were told which candidates matched their personalities to a high degree. The results were compared with a control group, which received the same set of images but without algorithmic manipulation.
After the screening, users were asked which fictional politicians they would vote for, based solely on the information available. People in the first group were much more likely to vote for the candidate the AI suggested. The control group, on the other hand, had more mixed results.
The second experiment worked on manipulation indirectly: in it, the algorithm favored the display of four candidates, previously exposing their images to the participants to generate familiarity. As a result, the influence did not show a significant statistical difference.
Context of influence differs on dating apps
In the other two tests, the researchers verified whether artificial intelligence would have the same results when manipulating combination decisions in a dating app. As in previous studies, users viewed profiles of people who the app said would be ideal for their relationships. In this case, participants completely ignored the algorithms' suggestions.
On the other hand, when the AI was shown the same pre-selections in the tests for familiarity reasons, participants were more likely to choose the profiles they had seen before. The researchers’ conclusion was no longer whether we can be manipulated by algorithms, but rather which ones are most ideally used in each situation.
Through which channels you reach those people, classic and out of the box. The Next Web
Image: Cottonbro/Pexels/CC