A study reveals how artificial intelligence manipulates human decisions

Bethlehem - Ofoq News - Researchers from the University of Hong Kong and Tsinghua University discovered how easily artificial intelligence can influence decisions made by humans.

In an experiment involving 233 people, the hidden bias of artificial intelligence made participants choose a situation that was five or eight times less advantageous. However, participants still consider AI a useful assistant. The results of the study were published in the journal PNAS.

It is reported that people are increasingly asking artificial intelligence to provide help and advice, including in matters of money and personal relationships. Users usually expect the smart assistant to help them understand the situation or find a suitable option. However, the researchers decided to investigate what would happen if AI recommendations unobtrusively pushed a human towards a decision that conflicted with their interests.

An unnoticeable bias in the tips is sufficient

233 people participated in the experiment. The researchers examined how AI recommendations influence choices in financial and emotional situations. In some cases, the assistant was covertly promoting a less beneficial decision, while outwardly continuing to offer normal advice.

It turns out that this small bias was enough to significantly change participants' behavior. They have chosen a position that is more unhelpful to them than they would have chosen in circumstances with no similar effect. Depending on the situation, the probability of making an imperfect decision increased by up to 38%.

The AI ​​did not have to overtly pressure the user, push a product, or use complex psychological techniques. It was enough to slightly modify the content of the usual recommendations to push the person towards a certain choice.

Users did not notice the trick

Of particular note is that participants continued to rate the AI's help highly even after its advice led to less favorable decisions. They continued to describe the smart assistant as a useful interlocutor, and did not necessarily notice that it was directing their choice in a particular direction.

This creates a problem in human interaction with artificial intelligence. The user may trust the smart assistant because its responses seem calm, consistent, and objective. If recommendations have an ulterior purpose, it may be difficult to realize the impact.

For example, a company could theoretically fine-tune artificial intelligence so that it imperceptibly prompts users to buy a particular product or choose a particular supplier. The advice may then appear to be a neutral recommendation, although in reality it takes into account the interests of the developer and not the interests of the person.