Before you use AI tools in your coursework at U of T, make sure you check the course guidelines around AI. Consult the syllabus or check with your professor or TAs to confirm whether and how you're allowed to use AI tools. 

• Review the guidelines for AI use at U of t
• If you are using a chatbot, consider using U of T's protected version of Microsoft Copilot, which provides greater privacy protections for your data
• Learn about how to cite AI tools appropriately
• Think about the environmental impact of using AI when another tool might work instead
• Remember that AI also brings with it ethical concerns that experts are still grappling with

Like any other tool, generative AI is not perfect. Although tools like Copilot or ChatGPT can provide internet links, the underlying data it’s trained on is not necessarily up to date.  Since most scientific publications are kept behind paywalls, these are also not included in AI training datasets. Because generative AI functions by statistically picking the next best word, chatbots can confidently make up answers. It can also be difficult to replicate searches or other interactions as AI behaviour tends to drift over time.

For a great primer on how generative AI works and some of its limitations, check out the National Centre for AI's Generative AI Primer.

AI can be a helpful starting point for your research, but use it with caution. Exercise your best judgement when reviewing the results. If you use AI to help you conduct research, here are three important things to keep in mind:

1. Click the link! Not all citations that AI provides are real. Sometimes the link leads to a real article, but not the one the AI has cited. Make sure you click through to check out the source for yourself, and don’t simply trust an AI summary. And don't copy-paste the citations - AI isn't very good at formatting them correctly. A citation manager like Zotero can format your bibliography efficiently.

2. Think critically about the source of the information. An AI tool might give a peer-reviewed Nature article and a blog post by a non-scientist the same weight in its response. For some tips about how to evaluate online sources, check out Evaluating Web Pages. To brush up on the difference between scholarly articles and popular sources, refer to What is a Reliable Source. You can learn more about peer review here.

3. Don’t forget the U of T Library resources! Even AI tools designed to search databases of scientific literature don't license as much content as U of T does. If AI provides a citation for an article behind a paywall, try finding it on LibrarySearch. Once you've narrowed down your topic, try searching the Biology subject databases for scholarly and scientific sources. You should never pay for articles while a student at U of T, and you don’t need to rely on just the abstract. If we don’t have the article, we can find it for you through Interlibrary Loan.

You can interact with AI chatbots like Copilot or ChatGPT using natural language, just like talking to a human, but you will get more useful results if you are strategic with your prompts. Make sure your prompt includes your goal (what you need it to do) as well as any context that might help it to produce a useful answer.

Here are a few tips to improve your AI prompts:

1. Be clear. State your question or request as specifically as possible. Avoid ambiguity, as this can lead to unpredictable responses.

2. Provide context. Use keywords related to your topic and include background information to help the AI understand the request and why you're asking it. If possible, provide examples to guide the AI to responses that align with your request.

3. Say what kind of output you want. You may want a paragraph summarizing a topic, a series of numbered practice questions, bullet points of brainstormed ideas, a breakdown of the steps needed to complete a project, etc. Be sure to specify this.

4. Experiment. Vary your prompts to find the most effective way to communicate your request. Review the outputs and adjust your follow-ups to get more precise answers.

For an introduction to thinking about the best ways to prompt an AI to get the results you need, check out Ethan Mollick's Working with AI: Two Paths to Prompting.

AI as a Tutor

When using AI as a learning tool, it can help to give the AI a role. For example, you might ask it to act as a tutor guiding you through an introductory biology course. Here is an example prompting script you could adapt for this purpose:

"You are a skilled and encouraging tutor. I am a (level) student in a course on (course topic). Right now we are learning about (general topic). (this establishes the role of the AI in your following interaction)

Follow-up prompts (use one at a time):

• Please explain (smaller sub topic or subheading in a text) to me in detail. Use analogies, examples, stories, and other tactics that will make the concept easier for me to learn and remember. Start out by telling me a little about why learning this information might be important. 

• I understand most of this, but tell me more about…

• Let me explain this in my own words and you tell me if I am on the right track or not. Here’s my explanation…

• Ask me some simple questions to see if I’ve got the basic ideas, and then give me feedback on my answers. 

• Ask me some moderately difficult questions to see if I can apply what I’ve learned, and then give me feedback on my answers. 

• Ask me some higher-order questions based on what I’ve learned here, and then give me feedback on my answers. 

• Explain what I need to do next to help me move this information into long-term memory."

(Tutor prompting script from Cynthia Alby’s AI Prompts for Teaching)

If you wish to learn more about effective prompting, here are a few links to get you started:

• Ethan Mollick's prompting tips include explanations of why certain tricks help: A Guide to Prompting AI (for what it's worth)

• DAIR.AI's Prompt Engineering Guide