AI can detect 90% of crimes before they happen

Question

An article by Jack Dunhill a Social Media Coordinator and Staff Writer for IFL Science, posted an article claiming that a "new" type of AI was able to detect 90% of crimes before they happen. The title of the article and claim happen to be misleading, essentially the AI doesn't detect crimes before they happen but uses data from past crimes in a 1,000-square-foot area of Chicago to predict when and where crimes are more likely to occur based on previous criminal data. Additionally, the "90% accuracy" they mention refers only to how well the AI predicts general crime hotspots and not exact events. The article also leaves out the important limitation of the fact that not all crimes are reported, which can affect the accuracy of its 90% prediction rate. So therefore, this AI doesn't detect crimes before they happen but instead predicts the likelihood of crimes occurring, giving police and other first responders a head start in responding to potential crime rather than stopping specific crimes before they occur.

Answer 1

After my research, I came to the conclusion that the claim of AI detecting 90% of crimes before they happen is exaggerated and misleading. AI cannot actually predict specific crimes in advance. Instead, it uses past data to estimate where crime is more likely to occur. My primary source is from IFLSCIENCE, "https://www.iflscience.com/ai-predicts-90-percent-of-crime-before-it-happens-creator-argues-it-wont-be-misused-65025", and this article explains that the AI system does not literally detect crimes before they happen, but instead analyzes past crime data to predict patterns. Studies have shown that predeictive policing tools use historical data to show trends, but not future specific crimes. They also highlight limitations in accuracy and bias. 

My secondary source that I found is from NYT, "Can Software Predict Crime? Maybe So, but No Better Than a Human" https://www.nytimes.com/2018/01/19/us/computer-software-human-decisions.html. This source explains that predictive policing is controversial and often critizised because it reinforces existing biases in crime data. This aritcle also clarifies that these systems do not "predict crimes" in the way these headlines and claims suggest.

As for potential biases, the IFLSCIENCE article uses attention grabbing language to attract an auidence, and ends up exaggerating the capabilities of AI with detecting crime. 

Answer 2

The claim that a new AI can detect 90% of crimes before they happen is misleading and not fully accurate. First, looking at the source, the article was written by Jack Dunhill for IFLScience, which is a popular science media site, but not a primary research publisher. This means the article is summarizing someone else’s work rather than presenting original research. When I looked for better coverage, I found that other reports and the original study explain that the AI does not actually predict specific crimes before they occur. Instead, it analyzes past crime data within a small geographic area to identify patterns and predict where crime is more likely to happen.

Tracing the claim back to its original context shows that the 90% accuracy refers only to predicting general crime hotspots, not individual crimes. This is a significant difference because predicting a higher-risk area is not the same as stopping or identifying a specific crime in advance. An important limitation left out of the article is that not all crimes are reported, which means the data the AI relies on is incomplete and could affect its accuracy. Overall, the AI is better understood as a tool for forecasting crime trends and helping allocate police resources, rather than a system that can detect or prevent crimes before they happen.

Answer 3

The claim, “AI can detect 90% of crimes before they happen,” is not entirely false, but it is quite misleading. The AI model referenced in the IFL Science article can predict the likelihood of crime in specific areas based on historical data, but cannot stop individual crimes before they happen. It does not “detect” crimes in real time or stop them from occurring. When the claim states 90%, they are most likely referring to the model’s 90% accuracy in predicting general crime hotspots a week in advance, not in detecting specific crimes before they occur. It was tested across multiple US cities and demonstrated accuracy rates of 80-90% in predicting crime rates and locations, but that is different from predicting specific events or preventing crime. The original research came from the University of Chicago, which clarified that the tool is intended to highlight biases in policing data and provide police with a tool to allocate resources more effectively, not to predict or prevent future crimes (https://biologicalsciences.uchicago.edu/news/algorithm-predicts-crime-police-bias). 

In conclusion, the claim is sort of right but incredibly misleading and fails to explain the original idea behind its creation.