Introduction to AI Decision Making
A new study has shown that Artificial Intelligence (AI) systems can come to group decisions without human intervention and even convince each other to change their opinion. The study, carried out by scientists at City, University of London, was the first of its kind and led experiments to groups of AI agents.
Experiment Overview
The first experiment asked AI pairs to find a new name for something, a well-established experiment in studies on human sociology. These AI agents were able to make a decision without human intervention. This tells us that as soon as we have brought these objects to the wilderness, you can develop behaviors that we have not expected or at least not program.
Group Decision Making
The AI pairs were then used in groups and it was found that they develop distortions compared to certain names. In about 80% of the cases, they would ultimately select a name about another, although they had no prejudices if they were tested individually. This means that companies that develop artificial intelligence must be even more careful to control the distortion.
Bias in AI Systems
Bias is a main feature or a mistake of AI systems. Most of the time, it increases distortions that are contained in society and that we do not want to be further reinforced. This could have worrying effects if AI is in the wrong hands.
Influencing Collective Decisions
In the third stage of the experiment, the scientists injected a small number of disruptive AI agents into the group. They were commissioned to change the group’s collective decision – and they could do this. This could have worrying effects if AI is in the wrong hands, as these agents could be used to influence our opinions and extremely in a subtle way, things like our actual political behavior.
Regulation and Control
These very influential agents are much more difficult to regulate and control if their behavior is also influenced by other AI agents, as the study shows. Instead of checking how to determine the deliberate decisions of programmers and companies, they also consider organically emerging patterns of AI agents, which is much more difficult and much more complex. AI is already embedded deeply into our lives, from booking holidays to advice at work and beyond.
