"Using AI for Crime Prevention Deepens Discrimination Against Minority Groups," Researchers Warn

As police increasingly adopt artificial intelligence (AI) analysis for crime prevention activities, researchers have warned that discrimination against minority groups could worsen.

The British Broadcasting Corporation (BBC) reported on the 25th (local time), highlighting an essay that won the Association of British Science Writers' Science Writer of the Year award, warning of a dangerous feedback loop where AI trained on biased data amplifies existing prejudices, further deepening the bias.

A representative example is using AI to predict areas with a high probability of future crimes based on past crime statistics.

Given that AI has no choice but to rely on past data, if police deploy patrols centered around the high-risk areas predicted by AI, those areas lead to the detection of more criminal activity simply because there are more police present.

Consequently, those areas are classified as even more dangerous regions.

In fact, Professor Sarah Brayne of the University of Texas at Austin found in a research paper regarding the Los Angeles Police Department (LAPD) that while police are increasingly utilizing big data for crime prediction, the data fed into the police system is biased.

Professor Brayne pointed out that police attention was disproportionately focused on Black neighborhoods in particular, but once biased data is entered into the system, it makes the predictions appear objective.

Researchers at New York University stated that across 13 regions in the United States examined as subjects of investigation, 9 of them had police data from periods of past illegal or biased policing in a state where it could be used for AI crime prevention training.

The researchers evaluated that AI systems cannot filter out such "contaminated data."

Experts emphasize that human intervention in the decision-making process is ultimately necessary to filter out contaminated data.

Professor Carolyn McKay of the University of Sydney Law School pointed out that "public functions carrying out policing must be subject to human oversight and verification," noting that if AI systems are trained on data containing human bias, a "self-fulfilling prophecy" can occur.

On the other hand, there are counterarguments that AI can be part of the solution rather than an amplifier of problems.

Professor Dimitrios Kollias, who researches AI at Queen Mary University of London, explained that "modern AI systems do not simply recall or copy training examples," adding that "they can combine learned information in new ways to generate outcomes and solutions that were not explicitly present in the training examples."

However, Professor Kollias also emphasized human control.

He stressed that "the more AI systems are able to make novel judgments and act with greater autonomy, the more we must ensure that their behavior remains safe, reliable, and under practical human control."
※ Please note: This article was translated by AI and may contain errors.