In 2024, despite many successes, the field of Artificial Intelligence (AI) faced several challenges. Over the past year, the AI industry has seen numerous product launches and even two Nobel Prizes in Chemistry and Physics. However, not everything went smoothly.
AI is an unpredictable technology, and the increasing availability of generative models has led people to test their limits in new, strange, and sometimes harmful ways. Here are some of the biggest AI failures of 2024:
1. AI Junk Litters the Internet
Generative AI makes it easy to create large amounts of text, images, videos, and other materials. In 2024, these often low-quality media were finally recognized as AI junk. This junk can be found everywhere online, from newsletters and books sold on Amazon to ads and articles on the web, and even in social media images. Such content is often shared widely, leading to high engagement and ad revenue for their creators.
The proliferation of AI junk poses a real problem for the future of AI models. As more websites become cluttered with AI junk, the performance and results of these models may deteriorate.
2. AI Art Distorts Expectations of Real Events
In 2024, surreal AI images began to affect real life. For example, “Willy’s Chocolate Experience,” inspired by Roald Dahl’s “Charlie and the Chocolate Factory,” made headlines when its AI-generated marketing materials led visitors to expect a more grandiose event than the actual sparsely decorated warehouse.
Similarly, hundreds of people lined the streets of Dublin for a Halloween parade that didn’t exist. An AI-generated event list was widely shared on social media, demonstrating how misplaced trust in AI-generated content can mislead the public.
3. Grok Allows Users to Create Images of Any Scenario
Most major AI image generators have rules to prevent the creation of violent, explicit, illegal, or otherwise harmful content. However, Grok, an assistant from Elon Musk’s AI company xAI, ignores these principles, creating images without restriction.
This disregard for rules undermines efforts by other companies to avoid problematic material.
4. Nude Deepfakes of Taylor Swift Circulated Online
In January, non-consensual nude deepfakes of singer Taylor Swift spread on social media. A Telegram community used Microsoft’s AI image generator to create these explicit images, highlighting the shortcomings of platform content control policies.
Although Microsoft quickly closed the loopholes, the incident showed how vulnerable we still are to non-consensual deepfakes.
5. Chatbots for Businesses Go Rogue
As AI becomes more widespread, companies are eager to use generative tools to save time and money. However, chatbots can invent information and are unreliable for providing accurate details.
Air Canada experienced this when its chatbot advised a customer to follow a non-existent refund policy. Despite the airline’s claims, a Canadian court ruled in favor of the customer.
6. AI Gadgets Fail to Ignite the Market
In 2024, the AI industry attempted to succeed with hardware assistants but failed. Devices like the Ai Pin and Rabbit R1 faced criticism for being slow and faulty, solving problems that didn’t exist.
7. AI Search Summaries Go Awry
Google’s AI overview feature made bizarre suggestions, like putting glue on pizza, due to its inability to distinguish between factual news and jokes. Such failures can spread misinformation and undermine trust in news organizations.
In conclusion, we hope the AI industry can learn from these failures.