Artificial Intelligence - An Introduction Guide Ask Us

Definitions

Artificial Intelligence (AI): Technology that enables computers and machines to simulate human intelligence and problem-solving capabilities

Machine Learning: Algorithms that use neural networks with an input layer, one or two 'hidden' layers, and an output layer

Deep Learning: A subdiscipline of Machine learning, that involves more 'hidden' layers, that allow for unsupervised learning (i.e., without human intervention)

Generative Artificial Intelligence (GenAI): AI that utilizes deep learning and training data to generate original, statistically probable outputs based on prompts

Chatbot: A computer program that simulates human conversation with a user, chatbots do not always use AI, but applications like ChatGPT use natural language processing (NLP) to interpret complex queries and make interactions more conversational (IBM, n.d.)

Applications of AI

TED-Ed: How will AI change the world? | TED Talk

AI has been a part of our daily lives for years – for instance, the tools recommending TV shows or movies to watch based on your previous Netflix activity, or the suggested words that pop up as you type an email or word document are examples of AI. These AI tools learn how to complete a specific task based on training data and can very successfully operate within the constraints of the task it has learned. Generative AI, or GenAI is different, because it can create new information or outputs based on training data. Rather than just recommending you content from its existing library, GenAI could write you a script or create an image to entertain you. Much of the current discussion on AI, and this research guide, is focused on GenAI.

Ethical Considerations for AI Use

Beyond the technical limitations of AI, it is important to consider the broader social, and ethical impacts of utilizing AI.

Bias: The training data used to help AI 'learn' can, and often does, contain biases about women, racialized people, and other marginalized groups. Therefore it is critical to remember that AI is capable of perpetuating harmful biases or stereotypes

• 13 Principles for Using AI Responsibly (hbr.org)
• 3 Types of Bias in AI 
• Research shows AI is often biased. Here's how to make algorithms work for all of us | World Economic Forum (weforum.org)

Privacy: The inputs you feed into AI applications may be used for training data, or other purposes, ensure you read the terms of service, and are careful about inputting sensitive information. Companies may also be tracking your activity across different platforms, or your physical location, if you haven't explicitly opted out.

• Privacy in an AI Era: How Do We Protect Our Personal Information? (stanford.edu)
• Elon Musk claims Apple's new AI tools are a privacy risk. How much of a concern are they? | CBC News

Copyright: The intersection of copyright and Generative AI (GenAI) is an evolving field. Copyright considerations may arise when inputting (training) materials for GenAI, creating prompts, or sharing the output. Submitting copyrighted materials as prompts into GenAI applications may infringe upon the rights of the copyright holder or violate the terms of use. Additionally, the output from GenAI may lack originality, and combine both human and non-human creativity. Copyright laws in Canada are not clear on whether such works can be protected. The United States Intellectual Property Office does not consider works produced by GenAI protectable. For support in considering the copyright implications of GenAI, contact Joshua Dickison, the UNB Libraries Copyright Officer, at jjd@unb.ca or copyright@unb.ca

IBM (n.d.). What is AI? https://www.ibm.com/topics/artificial-intelligence

Before You Start

Navigating when to use AI can be confusing, as you must juggle considerations of academic integrity, reliability of outputs, and the copyright/intellectual property of materials in a constantly changing landscape of regulations and technology. This section of the guide is aimed to support you in making informed choices about using AI technology for educational purposes. Before utilizing AI for academic purposes, consider the following questions:

• Has my instructor specified any course policies around the use of AI or Generative AI more specifically? Have they specified how they would like use to be acknowledged (through a form, citation, or other method?) If unsure, talk to your instructor
• Which aspect of my assignment or study plan could AI support? What are the limitations of AI in this case or when would AI use not be appropriate?
• What kind of output am I looking for? Based on this, which type of tool should I use?
• Which AI tools are available to me, either as free tools, or through UNB subscriptions?
• Where can I go for help, if I have trouble navigating the tool, or if I'd like to verify information in an output?

Hallucinations 

GenAI texts can put together highly sophisticated sentences, and convincingly make arguments - however, it is important to remember these tools are not immune from mistakes. Hallucinations refer to instances when AI outputs contain non-sensical or incorrect information, due to issues with training data, or the model itself. Therefore, it is always key to verify information, and use your critical thinking skills to examine AI outputs. For instance, review the below considerations for using ChatGPT 3.5:

1. ChatGPT 3.5 (the free version) is not connected to the internet, and cannot search the web for you. When you put together a prompt, it is answering based on information that was in its training data. Therefore, if you are looking for information that would require a search engine or database (like finding research articles), ChatGPT is not the best place to do this. 
2. ChatGPT 3.5 was 'trained' with data up to January 2022, so it does not have access to information more recent than this. Therefore, it will have limited utility in answering questions that require more recent information. 
3. Remember that ChatGPT is stringing words together that it thinks it will address your prompt/question, based on its training data - it does not have a concept of 'truth'. Therefore, sometimes answers will be incorrect, or completely made up. 

Academic Integrity

Unless explicitly stated in assignment instructions, it is important that works you turn in for academic credit are original works. Review the Academic Integrity Guide tab on Generative Artificial Intelligence (AI) for support in navigating whether to utilize GenAI for assignments. If you're ever in doubt about whether your use of a tool is appropriate, connect with your course instructor. When you do use AI, cite it (if appropriate) or follow your instructor's procedures for acknowledgement.

Citing AI

When you use AI to support an academic assignment, it is important to properly acknowledge its use. Refer to your course syllabus to determine which citation style is appropriate. Then review the UNB Library Research Guide tab on Citing Generative AI

OpenAI (2024). ChatGPT (May 13 version) [Large language model]. https://chatgpt.com/?oai-dm=1

AI in the Classroom 

Here at UNB, the Centre for Enhanced Teaching and Learning regularly presents sessions for instructors, and has resources to support teaching. They can also provide tailored support to faculty who wish to either redesign their assignments to integrate AI or create AI-proof assessments.

• CETL Teaching Tips
• CETL Guide to Generative Artificial Intelligence in the Classroom

You can connect with CETL at 

• Teaching support: Teachinghelp@unb.ca
• D2L support: d2l@unb.ca

Informing Students About AI

If you're interested in discussing AI with your students, consider assigning one of the below readings to your class:

• Belicher, A. (2017, August 9). Demystifying the black box that is AI. Scientific American. https://www.scientificamerican.com/article/demystifying-the-black-box-that-is-ai/
• Crawford, K. (2021). Introduction. In Atlas of AI. Yale University Press. https://unb.on.worldcat.org/oclc/1237863151
• Heti, S. (2023, November 13). According to Alice. The New Yorker. https://www.newyorker.com/magazine/2023/11/20/according-to-alice-fiction-sheila-heti
• Mollick, E. & Mollick, L. (2023, September 25). Student guidelines for proper AI use. Harvard Business Publishing Education. https://hbsp.harvard.edu/inspiring-minds/student-use-cases-for-ai
• Perrigo, B. (2023, January 18). Exclusive: OpenAI used Kenyan workers on less than $2 per hour to make ChatGPT Less Toxic. Time. https://time.com/6247678/openai-chatgpt-kenya-workers/
• Turk, V. (2023, October 10). How AI reduces the world to stereotypes. Rest of World. https://restofworld.org/2023/ai-image-stereotypes/
• Weiser, B. & Schweber, N. (2023, June 8). The ChatGPT lawyer explains himself. The New York Times. https://www.nytimes.com/2023/06/08/nyregion/lawyer-chatgpt-sanctions.html
• Wolfram, S. (2023, February 14). What is ChatGPT doing… and why does it work? https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/

Incorporating AI Into Assignments

One method of building AI literacy in your students is having them complete assignments that incorporate AI. The following resources lay out best practices or approaches for potential assignments:

• Husain, A. (2024). Potentials of ChatGPT in computer programming: Insights from programing instructors. Journal of Information Technology Education: Research, 23, https://doi.org/10.28945/5240
• Mills, A. (2023, September 8). Artificial Intelligence and education: A reading list. JSTOR Daily. https://daily.jstor.org/artificial-intelligence-and-education-a-reading-list/
• Mollick, E. & Mollick, L. (2023, September 25). Part 4: AI as learner. Harvard Business Publishing Education. https://hbsp.harvard.edu/inspiring-minds/ai-as-learner
• Stanford Teaching Commons (n.d.). Integrating AI into assignments. https://teachingcommons.stanford.edu/teaching-guides/artificial-intelligence-teaching-guide/integrating-ai-assignments
• Tossell, C. C., Tenhundfeld, N.L., Momen, A., Cooley, K., de Visser, E.J. (2024). Student perceptions of ChatGPT use in a college essay assignment: Implications for learning, grading and trust in Artificial Intelligence. IEEE Transactions on Learning Technologies, 17, 10.1109/TLT.2024.3355015

Grant Applications

As of November 18, 2024, the Government of Canada has released a new guidance document on the use of Artificial Intelligence in the development and review of research grant proposals. The key aspects of the documents are:

• GenAI can be used in the preparation of grant applications if authors
  • Ensure information is true, accurate and complete
  • Properly acknowledge/reference information*
  • State if and how GenAI has been utilized in developing their application, and follow all instructions specific to each funding opportunity
• Publicly available GenAI cannot be used for evaluating grant applications
  • These tools could result in breaches of privacy and the loss of custody of intellectual property**

*Remember that some GenAI applications do not provide sources for their information, making proper attribution difficult

**Entering a piece of work that you do not hold the intellectual property rights to into GenAI applications (asking it to summarize an academic article) may be a breech of the owner's IP

Research Methods

If you are planning to use GenAI for any part of your data cleaning, processing or interpreting, keep in mind that oftentimes GenAI can be a black box. Essentially, inputs enter a closed system/application in which a process that we can't fully see, describe, or sometimes replicate occurs inside the black box, which results in an output. In research, this could be problematic, as not having an in-depth understanding of the methods used, and calculations undertaken would detract from the transparency and replicability of your work. Opt for applications and models that are transparent in their functions, and always double check your target journal or publication ahead of adopting the technology. 

Manuscript Preparation 

Again, before utilizing GenAI to contribute writing to a manuscript draft, ensure you review the specific journal/publisher policies around doing so. Some major publishers are listed below, and for assistance in finding/navigating publisher policies, contact the Library.

• Springer-Nature
• Elsevier
• Taylor & Francis
• Wiley
• Sage

In most cases, ChatGPT and other LLMs should not be listed as authors on a manuscript, as they usually do not meet requirements for authorship. Although there is no universal list of requirements for authorship, and different publishers may have varying regulations, authors generally:

• Contribute to the design, analysis or interpretations of the article
• Draft/review the article
• Give final approval for an article to be published
• Agree to be held accountable for the article (see ICMJE Authorship Definitions, or your target journal/publisher policies for more information)

These requirements are not all typically met by GenAI, especially because models like ChatGPT cannot be held accountable in the same way a person can. Human input or manipulation is required to generate the types of outputs that would be included in a manuscript, meaning that while ChatGPT can be used as a tool, it has not made a significant enough contribution to be considered an author. However, please review the individual policies of the journal/publisher you are interested in publishing with.

Peer Review

Peer Review has been identified as being an opportunity to harness AI to speed up lengthy processes, and reduce the workload of reviewers. However, while AI can assist reviewers in identifying issues like formatting or language (which are typically addressed in the copy editing stage anyways), it cannot necessarily replace or replicate the expert assessment conducted by researchers with in depth knowledge of their field, including patterns and methodologies. Additionally, publishers may explicitly ban the use of GenAI tools in the peer review process, or putting in any section of a manuscript into a GenAI application.

CIHR has banned the use of AI tools, including those that transcribe/summarize spoken words, in meetings focused on peer-reviewing funding applications. The rationale for this decision is based on the fact that some AI tools do store their transcripts and may use these to train their models, which does not adhere to CIHR guidelines on handling peer-review documents.

Further Reading

Conroy, G. (2023, September 8). Scientific sluths spot dishonest ChatGPT use in papers. Nature. https://doi.org/10.1038/d41586-023-02477-w

Crotty, D. (2024, March 20). The latest "crisis": Is the research literature overrun with ChatGPT - and LLM-generated articles? The Scholarly Kitchen. https://scholarlykitchen.sspnet.org/2024/03/20/the-latest-crisis-is-the-research-literature-overrun-with-chatgpt-and-llm-generated-articles/

Stokel-Walker, C. (2023, January 18). ChatGPT listed as author on research papers: many scientists disapprove. Nature. https://doi.org/10.1038/d41586-023-00107-z

Also, consider checking PubPeer (an online 'journal club' where community members can post comments about articles) or the ChatGPT Retraction Watch list, if you feel that a paper may have unacknowledged GenAI contributions.

Prompt Writing

An important skill to ensure you get the most relevant possible outputs is phrasing your query or prompt in a clear manner. Prompt writing is often referred to as 'prompt engineering' and it involves putting together a question or command that clearly outlines what you are looking for, and details about the request. Clearer prompts typically result in clearer outputs, so prompt writing is a key skill to practice. 

Leo S. Lo, a Librarian at the University of New Mexico developed the CLEAR path, which is a framework for developing effective prompts for use in generative AI interactions (2023). 

Concise

• Unlike humans, GenAI can't always identify the key or important aspects of requests (see the video on the introduction page for an example)
• Therefore, when your prompts are overly wordy or complicated, it is possible that the model you're using may misunderstand or misinterpret what you are looking for. 
  • For instance, what issues do you see with the prompt: "Could you please put together a blurb on what a student union does and why we should support them"
    • Could you please is good manners, but unnecessary here
    • What do you mean by blurb? How long is a blurb?
    • Who is we? What do you mean by support?
  • An alternative: "Write a paragraph on the functions and importance of student unions

Logical

• Again, remember that GenAI can't necessarily sort through your question and identify where to start, so laying out the beginning and the end of the process could result in a stronger output
• Explain how tasks are related or the progression from one to the next
  • For instance: "Review my list of assignments and deadlines, and using that information generate a schedule for completing these assignments"

Explicit

• You want to aim to be as specific as possible to ensure you get the desired output, oftentimes, without explicit instructions otherwise, ChatGPT will output your information in bullet points, or a list
• If you're looking for paragraphs, or a table, make sure you state that clearly
  • While another person might surmise what you mean by "What should I know about operating a power tool," a more helpful way of phrasing the prompt for the sake of GenAI might be "List the steps and safety measures a person should take when using a sander"

Adaptive

• It's okay not to get it right on the first try, if at first you don't succeed, try try again! 
• You can provide additional prompts, or clarify earlier prompts if the output is not what you first asked for, think of prompt writing as a refining process, rather than a single attempt
  • When I asked ChatGPT to make me a schedule, it generated a numbered list with times listed next to the numbers.
  • I found this visually confusing, as the specific times imply the order, making the numbered list unnecessary. Therefore, I rephrased to ask for the information to be in a table format, which I found to be more clear

Reflective

• Finally, it is important to reflect on the information you're getting back, if it met your needs, and if your prompt writing strategies worked
• GenAI is constantly evolving, so it's possible that approaches that were previously effective at generating the outputs you needed, won't be as effective anymore

Lo, L. S. (2023). The CLEAR path: A framework for enhancing information literacy through prompt engineering. The Journal of Academic Librarianship, 49(4). https://doi.org/10.1016/j.acalib.2023.102720