Guidelines for Schools on use of Generative Artificial Intelligence
1. Introduction
As schools increasingly seek to deploy GenAI in day-to-day work, understanding the limitations of GenAI and establishing robust procedures and guidelines is critical. These guidelines are non-exhaustive and only provide a suggested and general framework as a starting point, covering both the initial procurement stage with AI vendors, as well as the daily use of GenAI by educators and staff. Schools are strongly encouraged to develop and maintain its own AI procurement and usage policies to ensure legal compliance and data protection.
2. Issues relating to AI Service Procurement
Many issues may arise during the various stages of AI service procurement, from the choice of vendor to contract negotiation, project implementation, performance monitoring and termination. Below are some non-exhaustive examples of key legal aspects that schools should consider:
- Legal and robust dataset: Understand how and from where the AI training data is obtained. Was the data legally obtained and cleared for intellectual property (IP) infringement risk? Do note that not all information publicly available on the internet can be used for AI training.
- Ownership and Use of input / output: Schools should check whether any AI training data and/or user prompts contain any third party work. Uploading third-party work into AI systems could constitute IP infringement. Similarly, if AI output is generated based on third-party work and/or are substantially similar, schools should evaluate whether its use violates any third-party rights.
- Warranty & indemnities: Does the vendor contract expressly warrant that the training data used, as well as any output generated, does not and will not infringe upon third-party IP rights? Would the AI vendor provide indemnity against (i.e. promise to compensate) losses suffered by the school arising from content that infringe third-party IP rights or is otherwise against the law?
- Transparency: Does the AI vendor provide clear documentation and information on how the AI system works, its capacities and limitations, to enable schools to explain to teachers, students, authorities and other stakeholders how the AI system works, as well as its capacities and limitations? For a better assessment of potential IP and regulatory risks, schools are suggested to require AI platforms to disclose the source of their underlying datasets and confirm that their datasets are not infringing.
- Data retention & termination: To what extent will the AI vendor retain any data provided by users to the system, including training data, input and output? For how long will such data be retained? Does the AI vendor warrant that such data will be erased upon service termination?
3. Risks Associated with Use of GenAI
Many GenAI platforms store and train from training datasets as well as its input and output data. AI training data may not be specific for the purpose, poisoned, or contain materials that infringe IP rights. As such, schools should ensure that human oversight should still be placed on day-to-day interactions with GenAI and the output generated.
- IP Infringement Risks
The use of GenAI carries inherent risks of IP rights infringement. The risk considerations apply to both inputs (the prompts/data you input/upload to the AI platform) and outputs (the materials generated by AI).
Common examples of IP infringement relating to use of GenAI include:
- The unauthorised use of GenAI materials that contain third party registered trade marks, when used in relation to goods/services under which third party trade marks are registered, may amount to trade mark infringement.
- The use of GenAI materials that resemble a known celebrity or cartoons, in circumstances where the public may be confused that the celebrity or the cartoon has authorised such materials, may amount to the tort of passing-off.
- The unauthorised use and uploading of third-party copyright work to train AI dataset may amount to copyright infringement.
- The use of GenAI materials that are substantially similar to third party copyright work may amount to copyright infringement.
Schools should pay particular attention to any copyright notice, licence terms or other rights statements accompanying copyright works. Where such notices expressly prohibit the use of the works for AI training, schools should not use those works for such purposes. For IP infringement risk examples and general suggestions, please see Appendix.
- Privacy & Confidentiality Risks
Personal data, i.e. information relating to a living individual that can be used to identify that individual, must be used according to the “Personal Data (Privacy) Ordinance” (PDPO). Personal data may include names, student number/identifier, contact details, assessment and exam results. Uploading personal data to an AI platform without prior notification or consent from the data subject, may violate the PDPO. Due to the nature of GenAI systems and that input/output are commonly used in further training AI models, schools are reminded to avoid uploading personal data onto GenAI systems, especially sensitive information relating to students. Before uploading any information on GenAI, ensure that personal data are redacted and metadata attached to the file are removed. For further guidance, you may refer to the “Guidance on the Ethical Development and Use of Artificial Intelligence” published by the Office of the Privacy Commissioner for Personal Data.
Appendix: IP Infringement Risk Examples and General Recommendations
|
|
Hypothetical examples & risks |
General suggestions |
|
Input |
Example 1: A teacher copies several chapters of a textbook upload them onto an AI platform to generate a quiz. |
Avoid uploading third party materials onto AI platforms, unless copyright has already expired, the terms and conditions relevant to the use of the materials specifically allows use for AI training, or an exemption under copyright law applies. HKRRLS is exploring how to introduce licenses for using licensed materials in AI training. Stay tuned for updates from HKRRLS. |
|
Example 2: A teacher extracts illustrations or photographs from textbooks, workbooks, reference books or publisher-provided teaching materials, such as diagrams of human organs, maps or plant structure diagrams, and uploads them to an AI tool / platform, asking the AI to redraw them as line drawings or coloured illustrations for use in worksheets, presentations or class activities. Risk(s): This practice may constitute unauthorised reproduction of copyright work. Even if the AI-generated image can differ from the original in some respects, the act of redrawing may still retain the substantive part and/or the overall impression of the original copyright work, and may still constitute the creation of a derivative work via substantial reproductions, thereby infringing copyright. If the original image is a work of a publisher or third-party creator, uploading it onto an AI platform without authorisation may also breach the applicable usage or licensing terms for the teaching materials. |
Avoid asking AI to redraw or alter original copyrighted images. Try to find openly licensed, royalty free or public domain images. For example, teachers may select images licensed via Creative Commons Zero (CC0) or public domain images as references or as the basis for modification, and should acknowledge the source and copyright owner in accordance with the applicable licence terms. Where only a general teaching illustration is required, use an original text prompt to generate an image instead of uploading the third-party copyrighted image for redrawing. Where school-specific teaching materials are required, schools should obtain formal authorisation from the publisher or relevant copyright owner before inputting such materials into AI systems, or use electronic teaching material platforms officially provided by publishers that expressly permit the use of AI functions. |
|
|
Example 3: A teacher uses textbook exercises, reading passages or a question bank as input to an AI tool / platform to generate interactive questions, such as those in Kahoot, Quizizz or an HTML5 game, and uploads the AI-generated materials to a third-party interactive platform where they are publicly accessible. Risk(s): The use of third-party copyright works as input may raise copyright infringement concerns. Uploading the AI-generated materials to a third-party platform and making them publicly accessible may further involve unauthorised reproduction, dissemination or communication of the copyright works, subject to the terms and conditions of the AI platform in question. |
Teachers may provide teacher-written learning objectives, key curriculum points or summaries of concepts and ask AI to design entirely new questions. If school-specific teaching materials are required for input into AI systems, schools should obtain formal authorisation from the publisher or relevant copyright owner, or use electronic teaching material platforms officially provided by publishers that expressly permit the use of AI functions. |
|
|
Example 4: A school or teacher uploads a PDF file of an entire textbook, e-book or the full text of a book to CustomGPT, a RAG system or another AI chatbot platform to create a dedicated question-and-answer chatbot, allowing students to enter questions and receive instant answers. Risk(s): This practice carries a very high level of risk. Importing an entire textbook or a substantial amount of teaching materials into an AI platform involves substantial copying of copyrighted works, which is very likely to constitute copyright infringement.Even if the chatbot only provides summaries or answers to students, its underlying knowledge base (e.g. the AI training dataset) may still contain substantial amounts of unauthorised copyrighted material, which in itself may amount to copyright infringement. Such use would generally go beyond the scope of the photocopying or scanning licences obtained from HKRRLS. Moreover, the copyright exemptions for educational establishments under the Copyright Ordinance are also unlikely to be available. |
Use publisher-authorised electronic teaching materials or AI platforms. If publishers provide AI tutoring, intelligent question-and-answer services or electronic teaching platforms, schools should give priority to using such authorised services. If a school creates its own chatbot, limit the knowledge base to teacher-created notes, school-original teaching materials, openly licensed resources, or materials for which explicit authorisation in AI use has been obtained. |
|
|
Example 5: A teacher purchases electronic teaching materials or digital learning resources using a personal account and uploads the materials to an AI platform, allowing multiple students or other teachers to access and use them. The materials are subsequently used for class activities or shared learning resources, resulting in the actual number of users exceeding the scope of the original licence. Risk(s): Allowing multiple persons to use materials licensed for a single user may breach the single user license, thereby constitute unauthorised reproduction and dissemination. The educational exception does not apply where the dissemination is being used as a substitute for purchasing a valid licence.Uploading copyrighted teaching materials to a public AI platform may also create risks of unauthorised copyright and storage or further reproduction of copyrighted materials. |
Select multi-user or school licences version where materials are to be used by multiple persons. In addition, give priority to enterprise/school versions of AI platforms with appropriate data protection features. Set access permissions to restrict access to teaching materials by authorised users only. |
|
|
Output |
Example 6: A school’s website is found to display an image generated by an AI platform that trained on the copyright owner’s work without consent. Risk(s): AI generated materials may contain unlicensed / infringing materials. |
Exercise human oversight by critically reviewing AI-generated outputs, verify the accuracy of AI outputs by cross-checking against reliable sources and conducting reverse-image or online searches before adopting or disseminating the information. |
|
Example 7: To engage students, a teacher requests AI to depict a short story in cartoon format, where the characters resemble an existing popular cartoon. Risk(s): Existing cartoon characters may be protected by copyright, and the AI generation of artwork substantially similar to those cartoon characters may infringe copyright. |
Avoid asking AI to generate any cartoons or artwork that are substantially similar to existing third-party cartoon characters or artwork. |
|
|
Example 8: A school uses AI-generated scientific charts, historical images, descriptions of individuals or data summaries in its website, publications, displays or class presentations. The AI-generated output contains errors, fabricated information or anachronisms, or the images contain suspected unauthorised watermarks, logos or other elements that may belong to third parties. Risk(s): AI-generated outputs may contain “hallucinations” such as inaccuracies or fabricated information, which may affect the quality of teaching and learning. Schools should not assume that content generated by AI is necessarily safe, accurate, or free to use. |
Exercise human oversight by critically reviewing AI-generated outputs and verifying their accuracy against reliable sources. For output involving science, history, law, medicine, data or factual information, verify the content against relevant academic works, textbooks, Education Bureau / government materials, or other reliable sources.For AI-generated images, check for watermarks, signatures, trademarks, logos or other elements that may belong to third parties, and conduct reverse-image or online searches where necessary before adopting or disseminating the outputs. |
|
|
Example 9: A teacher instructs an AI tool / platform to imitate the style of a well-known contemporary Hong Kong illustrator, living artist or particular contemporary designer, and uses the AI-generated output in teaching materials, exhibitions, publications or social media. Risk(s): Deliberately imitating the distinctive personal style of a living or contemporary creator may raise ethical, academic integrity, reputational concerns, as well as potential copyright, passing off, personal data, or other third-party rights concerns. The risk may be higher where distinctive characters, compositions or other recognizable elements are reproduced. |
Avoid asking AI tool / platforms to imitate the personal style of a named artist. More general descriptions may be used, such as “ink painting style”, “Impressionist style”, “geometric abstract style” or “children’s picture-book style”. Even when using a general style, avoid instructing AI to incorporate recognisable characters, compositions or other distinctive elements associated with a particular creator. Where appropriate, a work is used as a teaching example, in a school exhibition or in a school publication, schools are suggested to clearly indicate that the content is “AI-generated” or “AI-assisted creation” to maintain data sources’ transparency. |
|
|
Example 10: A teacher uses an AI tool / platform to produce teaching videos or handouts containing images, illustrations or text from third-party teaching materials. The completed materials are uploaded to the school’s learning management system (LMS), while some are made available on a personal website or YouTube. The materials remain accessible on the LMS for more than one year. Risk(s): The production of content containing images, illustration or text from third-party teaching materials may constitute substantial reproduction of third party copyrighted works, amounting to copyright infringement. Making the materials available on a public website without the copyright owner’s authorisation may also constitute infringement by communicating the copyrighted work to the public.In this scenario, the fair dealing defence under the Copyright Ordinance that is applicable to educational establishments may not apply, because (i) the materials are retained on the LMS for more than 12 months; and (ii) communicating the materials online via personal website or YouTube does not fall within “fair use” under the relevant defence. |
Unless otherwise agreed with copyright owner, remove or clear teaching materials containing third-party copyright works at the end of each academic year. Restrict online access of third party copyrighted materials to authorised users through appropriate login controls, and avoid making such materials available on public platforms. |
Contact Us
For enquiries concerning copyright compliance, reproduction of copyright works, or authorised licensing schemes, please contact the Hong Kong Reprographic Rights Licensing Society (HKRRLS) Secretariat at:
Email: Info@hkrrls.org.hk
WhatsApp: 852-62290879
Guidelines for Schools on use of Generative Artificial Intelligence
“Guidelines for Schools on use of Generative Artificial Intelligence” Poster
Copyright
- Guidelines for Schools on use of Generative Artificial Intelligence
- Useful Copyright Information
- Useful copyright information
- Reward Scheme to Combat Illegal Photocopying of Books and Periodicals (Funded By HKRRLS)
- Copyright Infringement
- Copyright Protection in the Hong Kong SAR
- The Copyright Classroom – HKU
- The Marrakesh Treaty on Copyright Exemptions for Persons with Disabilities
- Monitor Infringing Activities