Generative AI (GenAI) focuses on the creation of new content, based on existing data. Generative AI models, such as OpenAI's GPT, can generate text, code, images, and even music. This is by far one of the most popular applications among students. GPT uses vast amounts of information to produce outputs, mimicking human-like creativity. Despite the seemingly relevant and coherent output, however, tools such as ChatGPT or DALL-E also come with limitations.

Limitations of generative AI tools

Generative AI tools such as ChatGPT are valuable for a variety of tasks, including answering questions, providing summaries, and assisting with research. However, these systems are not without their flaws. Some of the main limitations include:

1. Untruths and hallucinations: AI models, while trained on large datasets, can sometimes generate false or misleading information. These inaccuracies arise because models do not "understand" the content but rather predict what comes next based on statistical patterns. Tools have generally been trained to provide an output, even if the veracity of the output is uncertain. The occurrence of hallucinations issue is especially concerning in academic contexts, in which factual precision is paramount. Researchers, students, and educators must double-check AI-generated outputs against credible sources to avoid the perpetuation of errors.
2. Bias: AI models learn from the data they are trained on, which often reflects the biases present in society. As a result, tools like ChatGPT may unintentionally reproduce stereotypes or skew perspectives based on gender, race, or culture. To address this bias, it is necessary both to improving training data and to foster a critical approach to using AI outputs in academic work.
3. Privacy, copyright, and security: Data privacy is another critical challenge. AI tools require vast amounts of data to function effectively, and sometimes sensitive, copyright-protected, or private information may be used during interactions. University libraries, which safeguard the privacy of both students and staff, need to be mindful of the risks associated with using AI tools. Ensuring that personal data is handled securely and responsibly should be a top priority.
4. Other challenges: There is an environmental cost for using AI, which must also be considered by the user. An AI query demands an enormous amount of energy and water. Estimates vary but what is clear is that an AI query uses significantly more energy than a search using a search engine like Google or Bing. This is owing to the much larger amount of data that is processed during an AI query. Users are advised to consider if AI is necessary to answer their questions or if a simple internet search would be equally effective.

AI and information literacy at universities

Information literacy forms the main focus of education taught by Leiden University Libraries. The ability to find, evaluate, and use information critically and effectively is important for all our researchers and students. As AI keeps evolving, it is crucial to know when you are reading AI-generated content or content written by a human. Several University Libraries in Europe have created webpages about this subject. Tilburg University launched tilburg.ai, a webpage with helpful tutorials on the use of AI technology especially for higher education. Newcastle University has several webpages on the use of AI in education, as well as an overview of GenAI tools.

Examples of AI tools

The range of AI tools is broad. Elicit for example, is a research tool which can assist with literature reviews and retrieving information from academic papers. Writing tools are popular as well. Flawlessly is a tool which does not only check the spelling or the grammar, it can also change the tone of a text to a more conservational or formal tone. Eightify is an example of a creative tool; it creates concise summaries of YouTube videos for quick insights.

Conclusion: The future of AI in university libraries

Generative AI presents both challenges and opportunities for university libraries. You are encouraged try them out yourself, but it is always crucially important to be conscious of the limitations and security and privacy issues of AI tools. There are numerous tools that can save you time while conducting research. Tools like Elicit, but also EvidenceHunt and Consensus are tools I personally use in orientation phases or for inspiration. If you want to know more about this, I recommend completing the University of Amsterdam's e-learning on tools; at the end of this tutorial, you can find a list of practical tools. You can also contact me on t.k.philippa@library.leidenuniv.nl.