An experiment using artificial intelligence has allowed social psychologists to uncover the all-too-human side of a racist robot.
Dr Dirk Van Rooy has found a way to make a computer algorithm racist.
Dr Van Rooy is a senior lecturer in social psychology at the ANU Research School of Psychology. As a social psychologist he investigates, among many other things, prejudice, discrimination, and social influence.
Unlike many other social psychologists, however, he uses techniques from artificial intelligence.
So how did they create a racist algorithm? By making it as much like us as possible.
“The brain at its most fundamental consists of large amounts of neurons, but it also has, if you look at the processes, what are called ‘self-organising’ algorithms.
“So if you throw information at the brain, it’ll organise it in a way which determines our behaviour; basically, to predict what will happen.”
Dr Van Rooy and his colleagues used neural networks, or software programs mimicking the way the brain works, to look at how information about social groups is processed.
This neural network was presented with equal amounts of positive and negative information describing two fictional social groups—one a minority (fewer group members), and the other a majority (more group members).
“It turns out that if you present a standard neural network with that type of information, at the end of the learning process, that network is actually racist. It’s prejudiced against the minority.
“And obviously there’s no type of motivation involved, it’s just the way our brain organises information.”
The researchers then went on to show that the same thing happens with people.
“It's because of this way that we organise information that we end up with stereotypes, especially negative stereotypes.”
One reason for this outcome, explains Dr Van Rooy, is that a typical brain response to complex information is to reduce what is called “dimensionality of information”, to limit the number of random variables under consideration . This can lead to our brains organising information into groups around positive/negative associations.
Most people, he says, are aware of the generalised nature of these associations and don’t express them or extrapolate from them.
However, when people are repeatedly exposed to certain minority groups in a predominantly negative light, it is very likely they will then develop illusory correlations between that group and the negative connotations.
These negative correlations will then become extremely accessible in the average person’s mind and memory. Every time individuals are asked to make a judgment about a part of that minority group, this negative stereotypical information is most likely to be retrieved.
“One of the implications is certainly that we should be careful how we depict minority groups. It might not be a good idea to relax the legislation around the Racial Discrimination Act, for example.”
Studies such as these are just the tip of the iceberg when it comes to the ability of social psychological research to explain real life processes, phenomena and complex global problems.
“If you look at some of the issues that we're dealing with in society at the moment—whether it's climate change or abuse in prisons, or banks that might be behaving slightly less than morally—those are all social psychological issues.”
It might take a robot to reveal the nature of our brains, but it’s the social psychologist who will reveal our very human nature.