Produced by the IFLA AI SIG, https://www.ifla.org/units/ai/

In compiling this resource we are seeking to provide a useful non-technical resource for information library and information professionals. We try to point to authoritative sources which are open to all. 

If you have any comments or suggestions for improvement please email them to a.m.cox@sheffield.ac.uk

1. Introduction to generative AI

Generative AI are systems that can produce new text, images or other media.

They could be differentiated from descriptive AI which focuses on improving access to content such as text, images, audio and video by identifying features in them to enhance search.

Resources:

Ortiz, S. (2023). What is generative AI and why is it so popular? Here’s everything you need to know. ZDNet, https://www.zdnet.com/article/what-is-generative-ai-and-why-is-it-so-popular-heres-everything-you-need-to-know/

UNESCO quick start guide to ChatGPT in Higher Education, https://www.iesalc.unesco.org/wp-content/uploads/2023/04/ChatGPT-and-Artificial-Intelligence-in-higher-education-Quick-Start-guide_EN_FINAL.pdf 

1.1 Examples of generative AI

GPT has been around for a while. The launch of ChatGPT by OpenAI propelled this form of AI into the headlines probably because it made the use of GPT so user friendly through its chat interface. 

Image generators:

Other AI 

1.2 How generative AI works

Resources:

A guide to Generative AI terminology: https://blog.scottlogic.com/2023/06/01/generative-terminology.html

Clarke, S., Milmo, D. and Blight, G. How AI chatbots like ChatGPT or Bard work – visual explainer. The Guardian. https://www.theguardian.com/technology/ng-interactive/2023/nov/01/how-ai-chatbots-like-chatgpt-or-bard-work-visual-explainer

2 Ethical and information issues 

Generative AI such as ChatGPT has potential benefits (see uses below), but reflections on the ethics of AI should be considered prior to use.

The following have been raised as issues with some versions of generative AI, such as ChatGPT:

  • Makes biased statements because of biases in the training data and how the training data was curated, eg GPT has been shown to be biased about gender, race etc (Webb, 2023). Trained largely on open web data it is bound to under-represent regions that are under-represented there, and who are also under-represented in AI research (Komminoth, 2023).
  • “Hallucinates” information  which is inaccurate, feeding the flow of misinformation – fails to acknowledge sources and often fabricates citations – is not itself citable because it does not currently generate a consistent answer. 
  • Will accelerate the content creation explosion – leading to even more challenges of information overload.
  • Fails to be explainable because it is far from open about what data it is based on or how it works (Burruss, 2020).
  • Privacy is at risk if you share your data with it – many companies blocking use due to fear of loss of data. Students at many institutions are being advised to not put any personal data into bots. 
  • Violates copyright by using text and data from the open Internet as training data without permission and creates content heavily based on mined content (Mahari, Fjeld and Epstein, 2023).
  • Threatens jobs, eg of journalists, editors and those in marketing. Axel Springer has already announced that they will replace some journalists with bots. 
  • Is available to people with money to subscribe, so disadvantages those without
  • Was developed by exploiting very low paid Kenyan workers to detoxify content (Perrigo, 2023).
  • May not be environmentally sustainable: GPT models are known to use a lot of computing power (Ludvigsen, 2022).
  • Reveals the disruptive power in the hands of big Tech companies.

The balance of importance of these factors may vary between context, eg between higher education and universities (where the impact on academic integrity is central to debate) or corporate research (where it is the inaccuracy of information that is critical). In some contexts it may be possible to ban some forms of generative AI or procure a localised system. For example, it is possible to run some open source AI models locally via Python or R without uploading private documents to the cloud. 

Fundamentally, although generative artificial intelligence has enormous potential for innovation and undeniably has significantly more knowledge than any individual human, it lacks is the ability to reason, consciousness and some of the most advanced human qualities.

Concerns raised by ChatGPT, among other factors, have re-energised plans to regulate AI, notably the planned EU AI act (The Artificial Intelligence Act ). It is reported that (https://www.europarl.europa.eu/news/en/headlines/society/20230601STO93804/eu-ai-act-first-regulation-on-artificial-intelligence ) “Generative AI, like ChatGPT, would have to comply with transparency requirements:

  • Disclosing that the content was generated by AI
  • Designing the model to prevent it from generating illegal content
  • Publishing summaries of copyrighted data used for training”

Bommasani et al. (2003) weigh up if Foundation model providers comply with the draft EU AI Act.

Resources:

AIAAIC. 2023. “ChatGPT chatbot.” https://www.aiaaic.org/aiaaic-repository/ai-and-algorithmic-incidents-and-controversies/chatgpt-chatbot

AIID. 2023. “AI Incident Database”.  https://incidentdatabase.ai/?lang=en

Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021, March). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?. In Proceedings of the 2021 ACM conference on fairness, accountability, and transparency (pp. 610-623). https://doi.org/10.1145/3442188.3445922

Bommasani, R., Klyman, K., Zhang, D. and Liang, P. (2023) Do Foundation Model Providers Comply with the Draft EU AI Act? https://crfm.stanford.edu/2023/06/15/eu-ai-act.html

Burruss, M. 2020. “The (Un)ethical Story of GPT-3: OpenAI’s Million Dollar Model” Last updated 27 July, 2020. https://matthewpburruss.com/post/the-unethical-story-of-gpt-3-openais-million-dollar-model/

Floridi, L. (2023). AI as Agency Without Intelligence: On ChatGPT, Large Language Models, and Other Generative Models. Philosophy and Technology, 2023, Available at http://dx.doi.org/10.2139/ssrn.4358789

Komminoth, L.Chat GPT and the future of African AI. African Business, https://african.business/2023/01/technology-information/chat-gtp-and-the-future-of-african-ai

Ludvigsen, K. 2022. The carbon footprint of Chat GPT. Last updated December 21, 2022. https://towardsdatascience.com/the-carbon-footprint-of-chatgpt-66932314627d

Mahari, R., Fjeld, J. and Epstein, Z (2023). Generative AI is a minefield for copyright law. The Conversation, https://theconversation.com/generative-ai-is-a-minefield-for-copyright-law-207473

Perrigo, B. (2023). “Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic.” Time, January 18, 2023. https://time.com/6247678/openai-chatgpt-kenya-workers/

Webb, M. (2023) Exploring the potential for bias in Chat GPT JISC blog, https://nationalcentreforai.jiscinvolve.org/wp/2023/01/26/exploring-the-potential-for-bias-in-chatgpt/

3 Uses

3.1 Uses for information professionals

Large language models have immense potential for information work. Many of the problems of specific services such as ChatGPT are more about how it has been implemented than about the underlying technology.

Text based generative AI has many potential uses in library professional, such as:

  1. Summarisation of texts, eg lay summaries of academic papers
  2. Generating draft metadata to describe material
  3. General uses such as drafting documents and communications, eg policy documents, targeted marketing

3.2 Guiding end users to safe uses

This section summarizes how an information literate user should be trained to approach generative AI tools: to consciously evaluate how to use them effectively and to critically understand the wider context of how platforms work to shape information experience. It is presented in a form as a set of prompts.

  1. Learn how to use it effectively, by experiment and reading reviews
    • How should we conceive of this tool? Eg as a clever writing assistant or a single point of truth?
    • Understand what tasks it might help with, eg brainstorming, drafting, editing, writing in different styles, summarisation
    • Is it trustworthy as a source of information: is the information supplied accurate and sources given? 
    • Are there systematic inaccuracies in the material it produces, ie biases?
    • How can questions be formulated to get the best answer, eg
      • Define style/ audience
      • Repeat the request and synthesize answers 
      • Ask for sources that can be checked
    • Are there alternatives that might be better for certain tasks?
    • Keep on learning: the tools are evolving rapidly
  2. Use it to improve how you learn and be reflective about how you are using it
    • Is it helping to improve your learning or just making things too easy?
    • How does using the tool make you feel?
    • Is it making you feel less connected to people?
  3. Protect your own privacy
    • What types of information is it safe to share with it?
  4. Ask who owns, develops and profits from it and the wider related system of information discovery on the platforms you use
    • Is it owned by commercially driven corporations so that use feeds their power?
    • Is it open about how it works?
      • Is recommendation actually narrowing access to information as a form of filter bubble?
    • Was it created exploitatively eg by using low paid labour OR by mining information on the open web without permission?
    • Does it have a negative environmental impact?
    • Does everyone have equal access or is using it gaining an unfair advantage?
  5. Use it ethically, acknowledging how it is used:
    • Are you permitted to use it in this context, eg under what conditions (if any) does your institution permit its use?
    • Acknowledge it’s use appropriately for the context, eg there is APA citation guide https://apastyle.apa.org/blog/how-to-cite-chatgpt

3.3 Guiding researcher end users

It remains unclear what uses of generative AI will be determined to be legitimate. There are many open questions about how Generative AI could be used in science (Birhane et al. 2023).

Questions that will be important to researchers include:

  1. What uses of AI in the research process are permitted? Eg for transcription, simulation of data or writing papers
  2. Which journals/ publishers allow which sorts of uses?

Resources:

Birhane, A., Kasirzadeh, A., Leslie, D., & Wachter, S. (2023). Science in the age of large language models. Nature Reviews Physics, 1-4.

4. Wider AI resources

For our earlier listing of AI resources see 23 resources to get up to speed on AI in 2023, https://www.ifla.org/23-resources-to-get-up-to-speed-on-ai-in-2023/

Some additional resources:

About this document

Created by IFLA AI SIG

Version 04 01 2024