AI Literacy & Ethics

Recent studies of AI use in higher education (e.g. Lund et al., 2025; Yusuf et al., 2024) report that students are increasingly interested in learning how to use generative AI tools responsibly and ethically, but often feel uncertain about appropriate practices. A 2024 Harvard undergraduate survey similarly highlights that students want explicit, consistent rules about AI use in their courses. Educators can help students navigate generative AI with confidence and integrity by providing clear guidelines and open communication.

AI Literacy For Students

A crucial aspect of communicating with students about AI is supporting their AI literacy. This means helping students understand what generative AI tools are, how they work, and where their strengths and limitations lie. Discussing how AI generates content, why it can sometimes produce errors or “hallucinations,” and how to responsibly use and cite these tools equips students to make informed decisions in their academic work. Some resources for this include:

• The AI Pedagogy Project by metaLab at Harvard
• The Harvard Libraries Artificial Intelligence for Research and Scholarship Guide, which includes information about citing AI. 

Building students’ AI literacy can start with a few key steps you take in your own course policies and classroom conversations. Here are some recommendations for how to communicate about AI use with your students:

• Include an AI policy on your syllabus. Be transparent with students about why you are asking them to complete a particular assignment and explain how using or not using AI tools will affect those goals. Syllabus statement advice is available through:
• The Office of Undergraduate Education website
• The Bok Center’s Illustrated Rubric for Syllabus Statements on Generative AI
• Having a clear AI policy on your syllabus is a good start, but you might also want to check in regularly with your students by posting the policy on your Canvas site, mentioning it in assignment prompts, and discussing it during class meetings and office hours. Regular reminders can help ensure that students understand and remember your approach to AI use.
• Require students to disclose their use of AI, whether for brainstorming, drafting, or other purposes.
• If students are not permitted to use AI, design assignments that minimize AI’s utility, such as personalized reflections, oral presentations, or in-class tasks, ensuring that students engage deeply with the material and demonstrate their own understanding.
• Model responsible AI use for your students by being transparent about how you incorporate AI in tasks such as preparing materials, generating content, or providing feedback. Demonstrating thoughtful use of AI can help students understand its potential as a tool for learning while reinforcing ethical and effective practices.

You can also encourage students to explore further resources for developing AI literacy, such as attending workshops, experimenting with new tools, or discussing their experiences with peers. Actively supporting students in building AI literacy can help them use these tools more thoughtfully and effectively. For example, students can use AI as a tool for deeper learning by generating their own study questions, analyzing content from multiple perspectives, or engaging in reflective critique of AI-generated outputs. Framing AI as both a partner in and an object of learning can help students build critical thinking skills and engage more meaningfully with course material.

Key Debates And Ethical Questions 

As AI becomes more integrated into academic life, it raises important questions around academic honesty, fairness, and responsible technology use. Addressing these topics with students is essential to help them understand the potential impacts of AI on their learning and future careers. 

Bias and Fairness in AI Systems

AI systems are trained on vast datasets– predominantly collected from the internet– and therefore incorporate the biases and stereotypes embedded in those datasets. AI-generated content reflects dominant cultural norms and can pose the risk of marginalizing or misrepresenting, reinforcing harmful stereotypes and perpetuating inequality. 

For the same reason, the use of AI might also undercut the learning goals of intellectual vitality. Intellectual vitality encourages students to question assumptions and resist arriving at premature conclusions— two areas where AI-generated output often falls short.

Another consideration is how certain AI tools can disadvantage certain student groups. For instance, AI systems that process language may struggle with non-standard dialects or multilingual speakers, leading to inaccuracies or misunderstandings. Vigilance in recognizing and addressing AI’s limitations is essential in diverse classroom settings.

Privacy and Data Security

Using AI in educational settings raises concerns about privacy and data security. Many AI tools require users to input personal information or academic work into platforms that collect and store data. In some cases, this data may be used for purposes beyond the immediate educational context, such as marketing or further training of AI models, often without the user's explicit consent. This raises ethical questions regarding the ownership and control over one's intellectual property and private information. 

To combat this, FAS members are encouraged to use Harvard-approved tools (Harvard’s AI Sandbox and ChatGPT Edu Workspace). These options allow for the upload of confidential materials (specifically, those materials Level Three and below).  

Environmental Impact

Training and running AI models uses substantial electricity and water. Data centers consumed about 1.5% of the world's electricity in 2024, a share the International Energy Agency expects to roughly double by 2030, with AI driving much of the growth. The strain is often local: data centers use more than a tenth of the electricity in several U.S. states, and many sit in dry regions where their water use competes with homes and farms.

The cost of any single request, by contrast, is small and often misunderstood. A few points that put common claims in perspective:

• One AI question costs about as much as a few seconds of watching TV. Asking a chatbot a typical question uses roughly a quarter of a watt-hour of electricity — comparable to running a microwave for about one second, and roughly level with or only modestly more than a Google search. It also uses somewhere between a few drops and a few tablespoons of water, depending on where the data center is and how you count.
• The famous "a bottle of water per conversation" claim is a misreading. The study behind it actually found a 500 mL bottle covers roughly 10–50 chatbot answers, including water used at power plants, not just the data center.
• Not all AI use is equally impactful. Generating an image, or asking a "reasoning" model to grind through a long problem, can cost tens of times more than a short text answer. The most expensive cases approach half a phone charge or a cup of water per request.
• Your individual usage is not where the environmental challenges reside. For personal footprints, one less burger or one skipped flight dwarfs years of chatbot use. The concern is the aggregate: AI is driving rapid data center construction, and about two-thirds of new U.S. data centers since 2022 are in places already short on water. Where the industry builds matters far more than whether any individual uses it.
• All these numbers are changing rapidly. Google reports its per-prompt energy dropped 33x in a single year. Figures more than a year or two old describe history.
• Growth in user demand for AI is a real concern. If usage growth significantly outpaces energy efficiency gains and data center location improvements, the environmental impacts could get worse.

Rather than framing AI as simply good or bad for the environment, we encourage weighing costs against benefits: match the amount of computation to the learning value of the task, and recognize that the larger challenges, such as growing demand, local impacts of data centers, and sustainable power, are collective ones that require a collective response.

Copyright and Intellectual Property

Generative AI tools rely on massive datasets that include copyrighted material. This can raise ethical and legal concerns around ownership, use, and attribution of content produced with AI.