5 AI Failures That Highlight the Importance of a Robust AI Strategy

Low-Quality Training Data

AI models need high-quality data to function properly. Low-quality data may compromise a service’s ability to differentiate between what’s true and what’s false. As a result, without high-quality data, the AI system may produce inaccurate and unreliable answers to user’s questions, and it may fail to interpret and respond to user inputs correctly.

Inherent Algorithmic Biases

An algorithmic bias occurs when an AI system produces an unfair or discriminatory outcome. This happens when a computer reinforces existing biases within its training data. Such biases may relate to racial, gender, or socioeconomic issues. As a result, an AI system training on unreliable data will also produce unreliable predictions.

Poor Contextual Understanding

If an AI system fails to interpret and respond to certain user inputs, it will produce inappropriate responses. This is usually the result of an issue with the natural language processing (NLP) component of the AI system.

The NLP is responsible for interpreting and responding to user inputs. A faulty NLP may struggle to understand who is talking, what they’re saying, how they feel, and why they feel a certain way. It may struggle to understand the true meaning behind the words that the user is transmitting to the system.

Air Canada Chatbot Gives Wrong Advice to Grieving Traveler

An AI chatbot is a powerful customer service tool. It can help enhance the customer experience, reduce waiting times, and ensure round-the-clock assistance. However, they must be deployed correctly – a lesson that Air Canada learned the hard way.

In November 2022, Air Canada passenger Jake Moffatt needed to book a flight to attend his grandmother’s funeral.

Before booking a ticket, Moffatt inquired about a bereavement fare to the company’s chatbot, which instructed him to book a regular fare and request a partial refund within 90 days. Unfortunately, he was not entitled to a refund under these circumstances, so the chatbot’s advice was inaccurate.

Air Canada denied Moffatt’s request for a refund, stating that it “cannot be held liable for information provided by one of its agents, servants, or representatives – including a chatbot.”

Moffatt took his case to the Civil Resolution Tribunal of British Columbia (CRT), which ruled in his favor. The CRT ordered Air Canada to honor the chatbot’s offer and reimburse Moffatt within 14 days of the ruling.

iTutorGroup’s Recruiting AI Discriminates Against Applicants Based on Age

There is no doubt that artificial intelligence in recruiting is a powerful tool. However, it must be used correctly to avoid discriminatory actions.

In 2023, iTutorGroup was forced to pay a $365,000 fine after being found guilty of discriminating against older applicants. When hiring tutors based in the United States, iTutorGroup programmed its tutor application software to reject applicants based on age.

The AI hiring software was programmed to reject male applicants aged 60 or older and female applicants aged 55 or older automatically. It rejected over 200 qualified applicants in the United States based on age.

Even though age-based discrimination was automated, the U.S. Equal Employment Opportunity Commission (EEOC) declared the employer was still responsible.

ChatGPT Creates AI “Hallucinations” That Falsely Describe People as Criminals

In 2024, German journalist Martin Bernklau made a shocking discovery: He was a convicted criminal. At least, that’s what Microsoft’s AI tool, Copilot, claimed he was.

When Bernklau typed his name into Copilot, he found the system was spreading misinformation about him. The system said that he was a 54-year-old criminal who had escaped from a psychiatric institution, dealt drugs, and preyed on widowers. None of this was true.

According to news reports, Copilot generated these “AI hallucinations” based on stories that Bernklau had written about real criminal trials. Thus, the system associated his name with those stories, creating a false narrative surrounding his real existence.

McDonald’s AI Drive-Thru Services Terminated Due to Poor Customer Response

In 2019, McDonald’s announced a trial drive-thru system. Developed by IBM and using voice recognition software to process orders, the AI-powered technology was used in more than 100 locations until June 2024.

Customers criticized the technology for its inability to interpret orders correctly. In one viral video posted to TikTok, a woman tries to order a caramel ice cream, only for the system to add stacks of butter to her order.

Despite the results of the trial drive-thru service, McDonald’s has since announced a new AI solution. Using AI and edge computing, the service helps perform predictive maintenance for kitchen equipment, improve order accuracy, and – once again – process drive-thru orders.

Sports Illustrated Accused of Publishing AI-Generated Articles

In 2023, Sports Illustrated was accused of publishing AI-generated articles and author biographies. Arena Group, the company that owns Sports Illustrated, outsourced its article-writing services to a third party, Advon Commerce.

Tech publisher Futurism ran a story after discovering issues with the articles provided by Advon Commerce. The most damning evidence was the author bios, which contained fake job titles, names, emails, and descriptions. Even the photos were found to be for sale through a website that sold AI-generated headshots.

Although Sports Illustrated denied the claims, the company took down the content and ended its partnership with Advon Commerce.