When Machines Make Mistakes: The Growing Problem of AI Errors

Artificial intelligence (AI) is becoming a common part of our daily lives, but recent events show that this technology can be unreliable and sometimes harmful. As companies and governments rush to use these tools, they are discovering that machines can make serious mistakes. From false accusations in supermarkets to fake information in official reports, the consequences of relying on AI are becoming impossible to ignore.

One of the most visible examples of this problem occurred recently at a Sainsbury’s supermarket in London. The store has paused its use of facial recognition technology after a customer was wrongly identified as a shoplifter. Matt Arnold, a 46-year-old comedy promoter, was buying supplies for an event when he was suddenly approached by managers. According to The Guardian and BBC News, he was told he could not be served because of an earlier incident and was asked to leave the store. Mr. Arnold described the experience as a “terrifying glimpse of the future,” noting that the staff seemed to follow the machine’s alert without thinking for themselves.

Although Sainsbury’s apologized to Mr. Arnold and claimed the incident was caused by “human error” rather than a software failure, the event has sparked a wider debate. This is not the first time such a mistake has happened. Both The Guardian and BBC News reported that similar cases have occurred at other stores, including Home Bargains and B&M. Critics argue that even if the technology is mostly accurate, a small error rate still affects a large number of people when millions visit stores every week. Mr. Arnold suggested that staff are not being properly trained to use these tools, leading them to trust automated alerts over their own common sense.

While retailers struggle with surveillance, the problem of AI errors is also affecting government policy. In Australia, a report supporting a new law to ban teenagers from social media was found to contain significant mistakes. The report, which cost millions of dollars, included citations that did not exist or did not match the research they were supposed to support. According to The Guardian, these errors appear to be “AI hallucinations”—a term for when AI creates false information that sounds convincing.

Initially, the organization behind the report, the Age Check Certification Scheme (ACCS), denied using AI to create the document. However, after further investigation by The Guardian, the group admitted that they had used ChatGPT to rewrite parts of the report. The analysis showed that several links in the document contained hidden data identifying ChatGPT as the source. The report’s authors later provided corrected information, but even those updates contained new errors, such as citing articles that had not yet been published at the time the report was written.

These two situations—one in a supermarket and one in a government-backed study—share a common theme: a dangerous over-reliance on technology. In the case of the supermarket, staff were so focused on the machine’s alert that they ignored the fact that the customer was acting like a normal shopper. In the case of the research report, the authors apparently failed to properly check the work produced by the AI, leading to the inclusion of fake references in an important policy document.

Experts are now asking whether companies and governments are moving too fast. There is a growing concern that AI is being used to make decisions that should require human judgment. As Mr. Arnold noted, it is not just about the technology itself, but about the culture of allowing machines to make choices that humans then follow without question. He warned that if these systems continue to be used without better oversight, innocent people will continue to be humiliated or falsely accused.

Looking ahead, the pressure on companies to prove their systems are safe is increasing. While some organizations argue that AI is necessary for security or efficiency, the public is becoming more skeptical. If businesses and governments cannot guarantee that their AI tools are accurate and reliable, they may face stricter rules or be forced to stop using them altogether. For now, these incidents serve as a clear reminder that while technology can be powerful, it is far from perfect and requires constant human supervision to prevent serious harm.