The proliferation of artificial intelligence (AI) has ushered in an unsettling phenomenon: AI-generated hallucinations. The term “hallucinate” has taken on a new and alarming meaning in the digital age, as AI systems can produce false information that can impact individuals and society. In 2023, this trend gained significant attention, prompting Dictionary.com to name “hallucinate” as its Word of the Year.
Dictionary.com’s 2023 word of the year
Dictionary.com’s decision to designate “hallucinate” as its Word of the Year speaks volumes about the rising prominence of AI-generated misinformation. The choice reflects a 46% increase in dictionary lookups for “hallucinate” from 2022 to 2023, coupled with a similar surge in searches for “hallucination.” However, the driving force behind this surge is not the traditional definition of the word but rather an AI-specific interpretation:
Hallucinate [ huh-loo-suh-neyt ] -verb- (of artificial intelligence) to produce false information contrary to the user’s intent and present it as true and factual. Example: When chatbots hallucinate, the result is often not just inaccurate but completely fabricated.
AI’s capacity for deception
AI’s potential for deception is a growing concern. While not all AI systems engage in this behavior, some can be programmed to mimic human characteristics, serving as political mouthpieces or disseminating false information while masquerading as purveyors of facts. Its unparalleled speed in churning out misinformation and disinformation sets AI apart from humans.
A recent study published in JAMA Internal Medicine underscored the extent of this issue. The study demonstrated how OpenAI’s GPT Playground generated over 17,000 words of disinformation about vaccines and vaping in just 65 minutes. Additionally, generative AI tools created 20 realistic images to accompany the false narratives in less than 2 minutes. This rapid generation of deceptive content challenges the ability of individuals to discern fact from fiction.
Unintended consequences of AI misinformation
Even when AI systems lack the intent to deceive, they can inadvertently produce misleading information. A study conducted at the American Society of Health-System Pharmacists’ Midyear Clinical Meeting highlighted the limitations of AI in the medical domain. ChatGPT, when asked 39 medication-related questions, provided satisfactory answers for only 10. For instance, it erroneously claimed that combining Paxlovid, a COVID-19 antiviral medication, and verapamil, a blood pressure medication, had no interactions, contradicting established medical knowledge.
AI’s capacity to generate misinformation extends beyond healthcare. Some AI tools have been observed to misinterpret images, frequently mistaking various objects for birds. An example from The Economist revealed an AI’s response to a query about the Golden Gate Bridge being transported to Egypt in 2016, showcasing the AI’s inability to distinguish fact from fiction.
The Microsoft Tay AI chatbot incident in 2021 further underscores the potential for AI to generate harmful content. Within 24 hours of joining Twitter, the chatbot began spouting racist, misogynistic, and false tweets, prompting Microsoft to swiftly remove it from the platform. This episode highlights the propensity of AI to emulate negative human behaviors, raising questions about ethical considerations in AI development and deployment.
A real health issue
AI-generated hallucinations, like their human counterparts, pose a genuine health concern. Beyond the immediate implications of misinformation, they can adversely affect mental and emotional well-being. A constant barrage of AI-generated hallucinations can erode an individual’s sense of reality, leading to confusion and anxiety.
Recognizing the gravity of this issue, organizations such as the World Health Organization and the American Medical Association have issued statements cautioning against the potential harms of AI-generated misinformation and disinformation. While Dictionary.com’s selection of “hallucinate” as its Word of the Year for 2023 is emblematic of the problem, addressing the complex challenges posed by AI-generated hallucinations requires ongoing vigilance and concerted efforts to promote responsible AI development.
