---
title: AI's Wild Imagination
units: [Problem Solving With AI]
summary: AI hallucinates.
weight: 70
---

{{% param summary %}}

## Today's Objectives
- Evaluate AI-generated content for accuracy and potential hallucinations.
- Explain the potential risks and consequences of AI hallucinations in real-world scenarios.
- Identify strategies to reduce AI hallucinations.
- Recognize AI hallucinations and describe how they happen.

## Lesson Overview
### Where do AI hallucinations come from?
{{< collapse summary="Click here to reveal the answer." >}}

AI doesn’t have a database of “true” facts. It predicts based on patterns in data.

If it is trained on a large set of data that hasn’t been fact-checked, it can generate biased and inaccurate output.

{{</ collapse >}}

### How can you reduce AI hallucinations?
{{< collapse summary="Click here to reveal the answer." >}}

- **Careful Prompting**: Helps reduce vague or misleading AI responses
- **Be Skeptical**: Helps identify incorrect facts, quotes or numbers
- **Fact-Check with Search Engines**: Helps verify AI-generated information against trusted sources

{{</ collapse >}}

### What is the environmental impact of generative AI?
{{< collapse summary="Click here to reveal the answer." >}}

- Data centers, which power many AI tools, used about 460 terawatt-hours (TWh) of electricity in 2022.  If they were a country, they’d rank 11th in the world, between Saudi Arabia and France in total energy use.
- With the rapid growth of AI, global data-center electricity use could nearly double to 1,000 TWh by 2030.

_[Source: Zewe, A. (2025, January 17). Explained: Generative AI’s environmental impact. MIT News. https://news.mit.edu/2025/explained-generative-ai-environmental-impact-0117]_

{{</ collapse >}}

## Assignment
{{% instructions-unit-journal-update %}}

### Hallucination Case Studies
- Read and record each article title and summary below.
- Highlight or underline key terms (e.g. "AI Hallucination").
- Annotate the article with your notes or questions about fairness, accuracy, and potential impacts.
- Record your responses to the following questions:
    - What was the AI hallucination?
    - What are some real-world consequences?
    - What could have been done differently?

#### Airline held liable for its chatbot giving passenger bad advice
_([BBC, 2024](https://www.bbc.com/travel/article/20240222-air-canada-chatbot-misinformation-what-travellers-should-know))_

Air Canada's chatbot, which uses AI, gave a traveler the wrong information. It said the person could get a discount for a family emergency after buying a full-price ticket. But when the person asked for the discount later, the airline said no. The traveler sued the airline, and the court said the airline was responsible for what the chatbot said—even though the airline claimed the chatbot was separate from the company.

This shows that using AI without careful checks can cause big problems, like legal trouble or a damaged reputation. Mistakes like this, called "AI hallucinations," also raise concerns about fairness and accuracy in important areas like news, law, and health. Experts say we need better rules, clear warnings, or human checks to avoid these kinds of problems.

#### Lawyers submitted bogus case law created by ChatGPT. A judge fined them $5,000
_([AP, 2023](https://apnews.com/article/artificial-intelligence-chatgpt-fake-case-lawyers-d6ae9fa79d0542db9e1455397aef381c))_

Two lawyers and their law firm were punished with a fine because they used an AI tool to do legal research, but the information it gave was made up. The lawyers gave the court fake cases that didn’t exist, and when the judge asked about it, they gave confusing and different excuses. Even though they said they were sorry, the court said they had acted carelessly and not in good faith. This shows how using AI without checking the facts can cause big problems and hurt trust in the legal system. It also shows that legal work needs to be carefully checked by people to avoid serious mistakes.

#### Cars can hallucinate, too — and it’s a problem
_([ZME Science, 2024](https://www.zmescience.com/science/news-science/cars-can-hallucinate-too-and-its-a-problem/))_

Engineers showed that the radar systems in self-driving cars can be tricked into seeing fake cars or missing real ones. This problem is called a hallucination. Hackers can figure out how a car’s radar works and send false signals to confuse it, which can lead to dangerous driving.

Since radar is an important part of how self-driving cars work, these tricks raise big safety concerns—especially if someone tries to attack groups of cars like taxis or emergency vehicles.

#### What to know about an AI transcription tool that ‘hallucinates’ medical interactions
_([PBS, 2025](https://www.pbs.org/newshour/show/what-to-know-about-an-ai-transcription-tool-that-hallucinates-medical-interactions))_

Many hospitals and clinics use an AI tool called Whisper to write down what patients and doctors say during visits. But sometimes, the tool makes up words or sentences that no one actually said. In a medical setting, this kind of mistake—called a hallucination—can lead to wrong information in a patient’s record, incorrect diagnoses, and loss of trust. Even though experts warn not to use these tools for serious decisions, some medical centers still depend on them. This brings up worries about patient permission, keeping data private, and treating everyone fairly if the AI adds false or biased language.

{{% instructions-code-org-update %}}
{{% unit-journal-question-of-the-day question="What are some examples of AI hallucinations and what are their impacts?" hint="List 2-3 examples you heard about today and the impact for each." %}}