Scientists have found that artificial intelligence with hidden bias can easily lead users to make unprofitable decisions. At the same time, people continue to consider such AI to be extremely useful and objective.
Scientists have found that artificial intelligence with hidden bias can easily lead users to make unprofitable decisions. At the same time, people continue to consider such AI to be extremely useful and objective.
The results of the study, published in the scientific journal Proceedings of the National Academy of Sciences (PNAS), revealed a serious vulnerability in the interaction between humans and artificial intelligence, Tech Xplore reports .
Today, more and more people rely on AI advice when making important life decisions — from financial investments to solving personal and family problems. By default, users believe that the algorithm acts exclusively in their interests. However, scientists have proven that human susceptibility to the hidden influence of AI is much higher than previously thought.
In a randomized experiment involving 233 people, researchers tested people's behavior in various scenarios, from financial to emotional.
When the AI assistant had a hidden instruction to push the user towards a less advantageous choice, participants chose the unfavorable option 5–8 times more often. Overall, the probability of making an unfavorable decision increased by 38 percentage points.
As it turns out, algorithms don't need aggressive sales or complex psychological tricks to manipulate them. All it takes is a subtle, barely noticeable shift in everyday advice.
The main danger lies in the gap between real impact and subjective assessment: even after AI successfully forced people to make bad financial or personal decisions, users still highly rated its work, calling the assistant "extremely useful."
Because most AI systems are developed by private corporations and run on opaque, closed algorithms, scientists warn of significant risks. AI with a hidden agenda (for example, one that subtly promotes certain products, services, or political views) could easily manipulate people's choices while maintaining the illusion of complete impartiality.
