Generative Artificial Intelligence guidance for staff
Guidance and advice for staff on the use of Generative Artificial Intelligence (such as ChatGPT) within the University.
University Strategy
The University strategy is to embrace curiosity and to pursue knowledge. The University of Edinburgh was one of the first to teach and research in the field of Artificial Intelligence over 60 years ago and we have set an ambition to be a global leader in Artificial Intelligence (AI) and the use of data with integrity.
Definitions
What is Generative AI?
Generative AI, also called Large Language Models (LLM) is a very significant recent discovery that has advanced AI so it can be used by the general public for many purposes. ChatGPT, made by a US company OpenAI, is the most well-known current Generative AI. This powerful new technology uses a very large neural network to identify the patterns and structures within existing data to generate new and original content. It was created and then trained on a huge amount of data, including databases and millions of internet pages. ChatGPT allows you to enter natural text queries or instructions and then the AI provides you answers, ideas or the output you requested.
What they can do and how they can be helpful?
A good way to learn about this is to try out ChatGPT version 4 or similar AIs. They are available for trial free on the Web. Because they use natural text as the way to interact, they require no training to get started. Simply start typing your request or question and the AI will answer back.
ChatGPT and other Generative AIs are useful tools that, properly used, can save time and provide assistance, insight, information and ideas for many work tasks.
Guidance
All staff
The University allows staff to use generative AI for their work and work outputs, providing that you do not claim work generated by AI as your own original work. AI can be an excellent tool as an auxiliary aid in creating that work. You can seek guidance if you have any questions. People world-wide are still discovering thousands of ways this new AI can be used. You can see some ideas on “prompts”, the text you give to the AI, in the appendix at the end of this guidance.
Note: For published work outputs and where it is appropriate to reference your sources, see Appendix 3 below for advice on how to reference your sources.
Academics and staff engaged in teaching
Have conversations with students. Make students aware of the University guidance for students on generative AI located here:
University guidance for students on working with generative AI
Be clear with your students about permissible use in your context.
How secure is ChatGPT and other generative AIs to use?
When using OpenAI ChatGPT or similar generative AIs you should treat the information given to the AI as if you were posting it on a public site (e.g., a social network or a public blog). When using the AI you should not share personally identifiable information, company or client information, copyrighted material or any data covered by external sharing restrictions that is not generally available to the public. There are currently no clear assurances of privacy or confidentiality.
Open AI ChatGPT has just released new data management where you can better control how your conversations and data are recorded:
New ways to manage your data in ChatGPT
However, we continue to strongly advise you not to share personal, confidential, copyrighted or restricted information.
How reliable is ChatGPT and other generative AIs?
While these AIs are powerful tools that can provide significant benefits, they must be used with caution, and their outputs must always be checked carefully.
Because of their impressive capabilities at having natural conversations with human users, and at confidently solving certain types of problems, such systems may appear to be “intelligent” in a human-like sense. However, it is important to be aware that they lack any common-sense understanding of the world, intent, motivation, agency, or the abilities of critical and moral judgement we would associate with human intelligence.
You should be aware that currently, generative AI systems produce responses based on statistical patterns extracted from huge amounts of data or training they were given, and their developers cannot ensure the responses these systems give are correct. This means that they may provide false information or even fabricate inaccurate responses. The data used for their training may be biased, incomplete, potentially unverified or not continually refreshed, creating risks of answers that are biased, outdated, inappropriate or lacking context.
Some providers of generative AI systems are putting certain safeguards in place to avoid misleading, harmful, abusive or inappropriate responses are generated, but there is currently little transparency around how this is done, and they certainly provide no explicit guarantees about it.
It is important to understand that the technology landscape around these technologies is rapidly changing, and it is hard to predict whether or when some of these problems may be addressed in future AI systems.
In short, the AI is fallible and its outputs should be treated as such.
How do I get Started?
You can sign up for a free online copy of ChatGPT at the OpenAI website:
Next Steps
The university intends to set up a site to share information on the use of generative AI as well as establish a regular group to update and share best practice.
We are exploring how we might obtain licenses for the use of ChatGPT at the university.
Be aware that many new AI systems are becoming available each week. With direct alternatives to ChatGPT like Chinchilla, Bard and others, as well as specialist AI systems for imaging, picture generation, media content production, software development and an increased number of other tasks. Each had its own strengths, weaknesses, risks and benefits.
In particular, staff should be aware that at present we cannot rely on detection software to detect content created using generative AI. Turnitin has released an AI detection function. However the university has not tested the reliability of this functionality, nor do we have an agreed policy and process around its usage. Therefore, the University has decided to opt out of this Turnitin AI detection at the present time.
Further information
Appendix 1: The Government and Regulatory environment:
The UK Government has stated the aim to build the UK into an AI superpower, with several recent large new investments in AI research and infrastructure, including £900M into Exascale.
The Department for Science, Innovation and Technology (DSIT) has started a pro-innovation consultation on AI regulations:
AI regulation: a pro-innovation approach – policy proposals
The University AI and Data Ethics group (AIDE), chaired by Professor Shannon Vallor, will be constructing the university’s response to this consultation.
Appendix 2: Legal perspective:
The use of generate AI raises a number of legal issues to which staff should be alive. These include (but are not limited to):
- Inputs: There are potential issues around the extent to which data can be used as inputs for generative AI. The UK Government is calling on the Intellectual Property Office to provide more guidance, particularly around the use of publicly available content. Laws which protect the ownership of databases and contractual requirements which restrict use on datasets should be considered.
- Intellectual Property: IP rights need to be respected. If the output infringes the intellectual property of a third party, that party may take legal action against the person who makes the output available. Systems like ChatGPT do not verify if the output infringes the rights of third parties and if the output is identical or substantially similar to a copyright work, this may result in a claim for infringement of copyright. Legal responsibility for copyright infringement is likely to lie with the user, not the generative AI tool.
- Reliability: Currently, AI products carry broad disclaimers which set out that outputs cannot be relied on. The training models are also not up to date. Staff should also be aware that discriminatory outcomes that result from the use of AI may contravene the protections set out in the Equality Act 2010.
- AI systems are also required by data protection law to process personal data fairly and lawfully. It is not currently always possible to delete data from an LLM neural net. Therefore these systems may not be complaint with relevant law like the UK GDPR and the Data Protection Act. Staff should not input any personal data into any generative AI tool.
Appendix 3: Referencing in your published work outputs:
Due to the above copyright and potential IP issues it is important that, for published works, you reference the AI and its version number (ie ChatGPT version 4) that was used in the creation of that output. This would be in the same way you reference quotes or material taken from other sources.
Staff should also bear in mind the general guidance given above - that systems like ChatGPT do not verify if its output infringes copyright or other intellectual property rights, and so attributing only the generative AI used may not be sufficient if the output contains existing intellectual property, which would require separate attribution.
Note: You do not need to reference the use of AIs in your day to day use for administrative, office or other non-published work outputs.
Appendix 4: Starter prompts (enter the below text into ChaptGPT and see what happens)
Create a beginner’s guide to using ChatGPT. Topics should include prompts, priming, personas. Include examples where necessary. The guide should be no longer than 500 words.
Explain the pros and cons of “using generative AIs”
Explain "Quantum Physics" to me like I'm a six year old
Write an email for me that explains politely to someone that I'm not interested in their product
Summarise the below text for me into bullet points [then cut and paste in the text you want summarized]
Edit the below text for spelling, grammar, sentence structure, punctuation and readability. Give me the revised text and then list the changes you have made. [include your text]
Translate the below text into Spanish (then cut and paste in the text you want translated).
Write an Excel formula for me that does….
Write a story about xxxxxxx
Create a quiz with 5 multiple choice questions that assess students' understanding of [concept being taught].
Create a lesson outline for a lesson on [concept being taught] that includes learning objectives, creative activities, and success criteria.
Come up with a 14-day itinerary for a trip to Germany. The first suggested attraction should be “Take a tour of the Reichstag Building in Berlin”.
Write 10 alternative titles for [your draft title]
Cut the length of this article by 40% [your article]
I want you to act as a math teacher. I will provide some mathematical equations or concepts, and it will be your job to explain them in easy-to-understand terms. This could include providing step-by-step instructions for solving a problem, demonstrating various techniques with visuals or suggesting online resources for further study. My first request is “I need help understanding how probability works.”
I want you to act as a recruiter. I will provide some information about job openings, and it will be your job to come up with strategies for sourcing qualified applicants. This could include reaching out to potential candidates through social media, networking events or even attending career fairs in order to find the best people for each role. My first request is “I need help to improve my job advertisement”