AI Literacy

This page provides an overview of the following things you should consider before using Generative AI tools. 

• Some Harm Considerations of Large Language Models (LLMs)
• Bias and Representational Harm in GenAI tools
• Data Privacy and Security Concerns
• Environmental Impact of GenAI tools

In addition to these considerations, you should refer to your professor's syllabus and consider whether the GenAI tool you have selected is relevant to your need.

“Some Harm Considerations of Large Language Models (LLMs)” by Rebecca Sweetman is licensed under CC BY-NC-SA 4.0 International.

Generative AI is only as accurate as the information that it is trained on. Most AI tools now have the capacity to ingest information online, so the tools are not limited to only the information fed to them by its engineers. Think of the different types of information available online that AI is now learning from: social media, opinion news articles, blog posts, and more. These types of information can be heavily skewed, but when AI is being fed this information it can't always distinguish between fact and opinion. That is a lot of opportunity for inaccuracies to be introduced, which can perpetuate biases in AI generated content. 

• Representational harms: AI can cause representational harm when it misrepresents a group of people in a negative manner. AI can be trained with information that has stereotypes embedded within it. For example, if the phrase "computer engineer" more frequently appears near the word "man" than it does near the word "woman" in AI's training data, it will more closely associate computer engineers with men, and therefore perpetuate the stereotype that computer engineers are men, not women. 
• Biases in image generators: In one study, an AI image generator was prompted to create 5,100 images of people. Image sets for higher paying jobs like CEOs, doctors and lawyers all had lighter skin tones on average, and image sets for lower paying jobs like janitor or fast-food worker were dominated by images of people with darker skin tones. These biases appeared for gender, with overrepresentation of men in image sets for higher paying jobs (Bloomberg).

Read or watch the following resources for more in depth information on how AI can perpetuate biases:

As AI becomes more integrated into our daily lives, the data we share—often unknowingly—can be at risk. AI tools may collect and distribute personal information, sometimes without our explicit consent, which can lead to privacy breaches and misuse of our data. Understanding how these tools handle your information and staying vigilant about privacy settings is crucial. By taking proactive steps like securing your accounts and keeping software up to date, you can better protect your personal data from unwanted exposure.

• Understand that AI algorithms may collect and share user data, including IP addresses, browsing activity, and personal details, often with third parties without explicit consent. Always review the terms and conditions of AI tools, paying particular attention to data sharing and permissions.
• Stay informed about the potential privacy risks associated with AI tools. Understanding these risks, as well as the basics of cybersecurity and privacy issues, is essential for minimizing exposure to threats.
• Protect your accounts by using strong, unique passwords and enabling multi-factor authentication. Weak passwords can be easily compromised, especially with AI tools potentially aiding in hacking attempts, so it’s crucial to secure your account access vigilantly.
• Regularly update your software and devices to benefit from the latest security enhancements designed to guard against data breaches and cyber threats.

Creating - and Recreating - Generative AI tools have an Environmental Cost

The creation, maintenance, and use of generative AI tools has an enormous carbon footprint. 

Generative AI tools will also likely need to have their knowledge base updated at some point. For example, ChatGPT was trained on information up to 2021. [But it can now access up-to-date information and browse the internet.] Aside from potential issues related to its knowledge base's currency, retraining these tools will also require another major up front cost in terms of natural resources (See Scientific American link below)

Using Generative AI Tools: Thinking Sustainably

Using these tools also has an environmental impact. While there's limited data on the environmental cost of a single question posed to one of these tools, it is estimated that it is 4-5x higher than searching that query in a search engine like Google. (See Wired Magazine link below)

Want to do a deep dive? Check out the sources below on some of these stats provided on the environmental costs of natural language processing tools like ChatGPT.