SADF'S AI POLICY - BALANCING PRODUCTIVITY WITH COPYRIGHT INTEGRITY

Last reviewed 16.01.25 [1]

1. INTRODUCTION?

SADF is drafting its policy on the use of AI - an ongoing process of which this document is but the first step. We like to think of ourselves as forward looking; we certainly embrace increases in productivity. Because AI is a new phenomenon, clear guidelines are still being developed; this is especially true for actors such as think-tanks, which are not considered a priority – the EU AI Act considers think tanks as ‘limited risk’ for AI use and does not spend much energy them (Gopalasamy, 2023; EU Artificial Intelligence Act, 2024). Yet clarity is important. After a short survey of internet (and from Chat GPT itself!) sources, we are presenting our first draft of our policy on AI use.??

AI is a very good starting point to gather information so we can begin our inquiries. ChatGPT’s (OpenAI, 18.12.24 [2]) answer about the accepted and unaccepted uses of AI was as follows:?

All the accepted uses listed regard the automation of either ‘routine tasks’ that are labour intensive but not intellectual or various forms of wide-range data analysis that are also labour intensive (and prone to human error). Some were already being performed by computer programs such as Excel. AI can also assist in information gathering and tips on how to structure information for a given goal. The AI edge always lies in its capacity to compute enormous amounts of complex information in seconds.??

Missing from the list of acceptable uses, however, is the actual intellectual work of evaluating the data gathered by AI. According to Oxford University (University of Oxford, 2024), ‘AI tools cannot replace human critical thinking or the development of scholarly evidence-based arguments’. AI can be used to ‘support’ one’s inquiries; however, all its inputs must be revised for accuracy and other concerns. Finally, ‘Clear attribution’ and ‘clear differentiation’ of one’s own work from any text or material from AI tools is crucial to avoid plagiarism.??

***For instance, this policy’s subtitle, ‘Balancing Productivity Boosts With Copyright Concerns’, was drafted with the help of AI (for initial inspiration). Our goal is to balance productivity boosts with copyright concerns; consequently the prompt used was ‘I need a good slogan, or expression, for a think tank policy on AI that blends productivity boosts and copyright concerns’. Chat GPT’s (2025) answer was ‘"Empowering Innovation: Balancing Productivity with Copyright Integrity in the Age of AI." We borrowed the term ‘copyright integrity’, created our subtitle, and added a footnote referencing both the AI source and the way it was used (see footnote 1, p.1).***

According to the Organization for Economic Cooperation and Development (OECD.AI, 2024), AI use must follow five key principles: Transparency (clarity about how AI tools were used); Privacy (protect user’s privacy when handling personal data); Human Control (strict oversight); Fair Application (detecting and mitigating biases); Accountability (comprehensive documentation/ recording of AI operations that allow for future scrutiny). The Centre for Community Solutions (Carlyle, 2023) used AI to generate a few tips on AI limitations we must be aware of. These are:??

1. 'Bias and fairness’ (those of the humans who ‘designed the AI models’ in question and ‘provided the training data’; Systemanalyse Programmentwicklung [SAP], 2024);?
2. ‘Lack of context’ AI still lacks much of the subtle nuances of human understanding and must be provided with as much context information as possible in their prompts. Lack of context can lead to greatly inappropriate responses;?
3. ‘Limited understanding of the world’: AI ‘may not be able to generate responses to questions or prompts that fall outside of its training data.’?
4. Vulnerability to adversarial examples’: it is vulnerable to ‘attackers’ that input misleading or incorrect information into the model.?

These are the basis for SADF’s Policy on AI use.?

?

2. SADF’S POLICY ON AI USE?

SADF embraces the use of AI as a way to strategize the use of time and energy, automate tasks not related to a think tank’s intellectual pursuits, and compute great amounts of information to create starting points for human reflection. We do not see AI-generated information as a final product and do not condone the use of whole AI-generated texts. When we use AI, we will be fully transparent about it, providing not only in-text references and a final reference list but also various footnotes and any other additional information we deem relevant.??

The following table synthesizes SADF’s policy on the use of AI generated information:??

Crucially, SADF, like other such organisms, absolutely keeps the THINK in ‘THINK-tank’. A couple of points are important here:?

• AI-generated summaries are often grossly oversimplified. Each point in an AI-generated answer is but a starting point for more complex and nuanced enquiries.??
• AI-generated information merely gathers and processes information already present in the internet – that is, often enough, the status quo. SADF has a strong commitment to questioning, reflecting on, and often challenging said status quo.??

As a Europe-based think tank centred on South Asian issues, SADF faces the additional challenge of overcoming the ‘western echo-chamber’ on any internet research about the world; this includes AI-generated content. In order to deal with this:??

• Different AI chats will give different answers. So try Gemini, Microsoft Copilot, ChatGPT
• Try different keywords and prompts and see where it gets you. If you want to avoid eurocentrism, look for South Asian keywords.?

3. ON TRANSPARENCY?

Every time SADF uses any form of AI-generated outputs, we will make clear attributions?

• Collaboration with AI in the article’s subtitle?
• Acknowledgements of AI generated images and slogans??
• Proper reference of every AI input: in-text, in the final reference list; when necessary, additional information will be provided in notes and footnotes??
• A Standardised Disclaimer at the end of every article, along the following lines:??

AI was used in this article to create slogans, generate images, gather information, inspire questions. We accept texts edited through AI; however, we do not accept content generated by AI as a final product – only as initial food for thought. All AI generated material is properly referenced.?

In our REFERENCING STYLE, we follow the currently accepted norms. Norms on referencing AI are still being developed and adapted; an article last revised on November 6, 2024 (Caulfield, 2024) suggested the following methodology [full transcript and images]:

How to cite ChatGPT in APA Style?

• ·Create an APA reference entry that lists OpenAI as the author and ChatGPT as the title, adding the date of the version used (shown at the bottom of the page on the ChatGPT site), the descriptive text “Large language model” in square brackets, and the URL.?
• ·The in-text citation consists of “OpenAI” plus the year of the version you used. Add an in-text citation each time you quote or paraphrase text from the tool.?
• ·APA advises describing how you used the tool in your methodology section or introduction and including the prompt you used whenever you quote a ChatGPT response. You may also add an APA appendix that includes the full text of any longer ChatGPT responses you quote from.?

MLA advises that if you use an AI tool like ChatGPT or Bing AI to locate sources and then use those sources in your work (rather than using the AI-generated text itself), you only need to cite the sources you actually used, not the AI tool used to find them.?

MLA also states that if you used an AI tool to edit your writing or translate words, you should acknowledge this at an appropriate point in your text or in a note.

4. CONCLUSION

We understand AI could well become the greatest revolution since the computer itself. Because it gathers information so well and automatizes complex and time-consuming tasks so efficiently, it’s likely to propel human capacity for information processing to new levels. The bar for finished products involving intellectual analysis is therefore likely to be set much higher. In these early days, however, future projections are always risky. In this memo we focus on immediate concerns; namely, on the fact that SADF’s goal is to use AI only as a support and supplement to our own research.?

1. We use AI to make the gathering and initial processing of information infinitely faster, and to perform all the tedious tasks and social media-intended practices that are not in our intellectual scope.?

2. When using AI, we must be fully transparent and specific about when and how we did it?

3. We’re all in the initial phases of AI, and no policy is entirely clear cut on how to use AI; however, we do know what we don’t want to use it for. We are not to confound AI information processing with human insight and reinvention.?

REFERENCE LIST

Carlyle, P. (2023). Showing our work: does Community Solutions use ChatGPT? Open Community Sources. https://www.communitysolutions.com/resources/showing-our-work-does-community-solutions-use-chatgpt??

Caulfield, J. (2024). ChatGPT Citations | Formats & Examples. Scribbr. https://www.scribbr.com/ai-tools/chatgpt-citations/?

EU Artificial Intelligence Act. (2024). The AI Act Explorer. https://artificialintelligenceact.eu/ai-act-explorer/??

Gopalasamy, P. (2023). Navigating the EU AI Act. Onetrust. https://www.onetrust.com/blog/navigating-the-eu-ai-act/??

OECD. AI Policy Observatory. (2024). OECD AI Principles overview. https://oecd.ai/en/ai-principles??

OpenAI. (2024). ChatGPT [18.12.24]. [Large Language Model]. https://aichatbot.co/chat/1734513675553??

OpenAI. (2025). ChatGPT [16.01.25]. [Large Language Model]. https://aichatbot.co/chat/1737013327195?

Systemanalyse Programmentwicklung [SAP]. (2024). What is AI bias? https://www.sap.com/resources/what-is-ai-bias#:~:text=Bias%20in%20AI%20models%20typically,used%20to%20train%20the%20AI.?

University of Oxford. (2024). Use of generative AI tools to support learning. https://www.ox.ac.uk/students/academic/guidance/skills/ai-study??

[1] The slogan ‘Balancing Productivity Boosts with Copyright Integrity’ was inspired by an AI suggestion (OpenAI, 2025). The prompt used was ‘I need a good slogan, or expression, for a think tank policy on ai that blends productivity boosts and copyright concerns’. The answer was ‘"Empowering Innovation: Balancing Productivity with Copyright In the Age of AI"

[2] ?We used the following prompt: ‘What I need to know about how to use AI in my think tank is what specific policy should I follow. What is the information on the internet about the acceptable uses of AI in research, and which uses are unacceptable?’

By Madalena Borges Coutinho Casaca

#AIPolicy #AIandcopyright #copyrightconcerns #copyrightintegrity #AIandthinktanks #AIandproductivity