Generative AI in Assessment

What is Generative AI?

Artificial Intelligence (AI) is a rapidly developing field of technology with the potential to revolutionise the way we live, work and learn. A simple definition of AI is when machines mimic cognitive functions to perform tasks which usually require human intelligence.

Recent breakthroughs in Generative AI (GenAI) have implications for how we approach content creation. Popular tools such as Gemini, ChatGPT, Midjourney and DALL-E are able to:

• Generate text including essays, articles, emails, poems, song lyrics
• Generate images including artwork, diagrams, realistic images, combining images
• Translate text and enhance spelling, grammar and punctuation
• Generate, fix or rewrite programming code
• Summarise documents
• Ask and/or answer questions
• Interface with a user in a way which mimics human interaction

GenAI tools are trained on very large data sets, and can replicate different styles and produce convincing outputs. However, they are not always correct and can produce ‘hallucinations’. Incorrect information can be presented as fact, and images may show glitches such as a human hand with too many fingers.

For more information on how the University of Sheffield approaches the use of Generative AI, please visit the Generative AI Principles for Students. 

The University uses Google Apps, as part of this Gemini is the supported Gen AI tool at The University of Sheffield.

Why is Gen AI Important?

GenAI is here to stay. It is already having, and will continue to have, a far-reaching effect on what we need to learn and how we learn it. 

Responsible and ethical use of GenAI is a skill that has the potential to be transformative across all academic disciplines. It is also a skill that is increasingly valued by employers within a wide range of professional contexts. 

Our understanding of how GenAI can be used effectively as a tool for learning is still evolving, but we are committed to ensuring that you are supported to develop AI literacy without compromising the principles of academic rigour and integrity. 

The following guidance will help you to use GenAI tools appropriately and ethically in your studies. 

---

General Guidance

• The golden rule: always check your department’s guidance and the specific module assessment criteria as the use of GenAI may be specifically prohibited on certain modules or assessments. If you have any doubt at all about this, always ask the module tutor for clarification. 
• GenAI outputs should not be used as sources for assessment and you should never cite anything from a GenAI tool. This is because content generated by AI tools is not reliable and is usually non-recoverable and non-reproducible at a later date, so it cannot be retrieved from a link or citation.
• Do not copy and paste content from an output produced by a GenAI tool directly into your work. Instead, you should use an output as a way to inform your further research and thinking. 
• A full disclosure of any content produced by GenAI should always be acknowledged in your work. Attempts to pass off content as your own work is counted as unfair means (see below) and may lead to action being taken against you. 
• Just because you can use GenAI for an assignment, doesn’t mean that you should. In most cases your own ideas, knowledge and experience will provide the most valuable starting point. Using GenAI can be a way to shut down and narrow your own creativity, as well as a way to enrich it. 
• You should not provide any personal, private or confidential information in your prompts.

---

Using AI

Written Work

Academic writing is usually a formulaic process based on conventions or models that are widely replicated across published work. Literature reviews, scientific reports and academic arguments are organised around common structures that allow us to quickly and easily find the information we are looking for as readers. However, academic writing also requires a critical and cautious engagement with a wide range of sources, all of which should be properly integrated, engaged with and referenced. 

As you can see from the example provided, GenAI can be a helpful tool for some aspects of the academic writing process. It can quickly generate a basic overall structure and identify a number of relevant points. However, GenAI will usually produce a generic and basic output that lacks critical or creative insight. It will usually adopt a neutral or passive position, such as ‘Overall, the impact of generative AI on the job market is uncertain.’ It will be your job as a critical thinker to explore the evidence, formulate your own informed view and add nuance to this basic position. 

You may wish to consider using GenAI in the following ways to develop your written work:

• GenAI can help to identify an overall structure based on the ‘action’ word in an essay title (for example, analyse, evaluate, discuss, compare/contrast).
• GenAI may be able to help you to articulate your own ideas and formulate your thinking by acting in the capacity of a 'critical friend' (i.e. by providing counterpoints to your ideas or by asking you a series of salient questions related to a topic (for more on this, see Prompt Engineering below).
• You may be able to use GenAI to enhance the overall quality of your use of language, for example by providing useful linguistic structures (linking words and phrases) to help you to communicate your ideas or by providing feedback on your use of grammar.

The most important thing to remember is that GenAI will not always produce a reliable or accurate output. You will need adopt a cautious and critical approach to ensure that you are not using GenAI in a way that compromises the authenticity and originality of your own work.

In practical terms, you should always fact check any claims or sources that form part of an output generated by an AI tool and you should never replicate generated content directly in your own work. Instead, you should think about how it may inform your own thinking and use of language and refine your own writing accordingly. 

For more guidance on using Generative AI in academic writing, please see the Study Skills Online page on Using GenAI for Essay and Report Writing.

For more guidance on the implications of Generative AI for discovering information, see the Generative AI Literacy, University of Sheffield Library Guide.

Example prompt and output.

Image generation

AI-powered image generation has developed at speed and is capable of creating sophisticated images from text prompts. Real-world applications include medical imaging, marketing, and use in the entertainment industry, such as gaming and movies. 

As with AI-generated text, there are some limitations to using AI-generated images, both in their quality and in relation to the ethics of their use. 

Image outputs are dependent on the inputs they were trained with, so there is a lot of potential for bias and reproduction of  inequalities. Looking at the example, reflect on what biases may be being reproduced in this image. 

Assumptions about the age, ability and gender of students are present here. Image generators also tend to have difficulty creating certain details, in particular hands, eyes and teeth. 

The ability to create deepfakes for propaganda is a particular concern. It is extremely difficult to detect deepfake images, and content can spread very quickly. It is imperative to check the source of any image that you discover and make use of in your academic work. See the Library resource Discovering and Using Images, Graphs, Charts and Figures for more information.

One of the most difficult issues is around ownership of intellectual property. If an image is created by AI, who owns the intellectual property? Is it the person who wrote the prompt? The original artist that the AI was trained on, the AI developer, or the AI itself? There are no immediate answers to these questions and the answers may emerge through legal test cases. For the reasons above, it is worth considering whether an AI-generated image is right for your needs. Would stock images be more suitable? The library has a list of useful image banks here.  

However, there may be times when an AI-generated image is appropriate, particularly if you are working on something creative and want to bring an idea to life.

The image below was generated using the prompt UK 'university students'.

Source: stablediffusionweb.com, generated 25/8/2023.

Coding/Maths

Generative AI has a specific role in writing code and can be an extremely useful tool. There are a number of questions that should be considered when you are using it, as well as following the general guidance: What does the code produced do? Does it only give the desired output? Are there any undesired, unintended results? Is the code efficient? Does the code reflect the level of learning you are at? Can you explain how the code works?

The purpose of most assessed work relating to code is for the student to demonstrate their understanding of coding concepts. In this sense GenAI can be used as a supportive tool but cannot impart understanding and should not be relied upon too heavily. Always follow the departmental guidance and ensure you evidence any use of GenAI.

Another area of use of GenAI is to support and develop data analysis. However, GenAI should be used with significant caution, and concerns around the integrity and security of your data are paramount. Especially, but not exclusively, in research contexts it is vital that you understand that passing any data to a GenAI may well lead to security breaches in your data. AI tools store and learn from information submitted into them. Personal or sensitive information should never be entered. Entry of information will result in the loss of intellectual property, as it becomes open source and may also breach GDPR regulations.

There are many versions of GenAI that can do a very good job of producing written solutions to mathematical problems. These tools should only be used to enhance your comprehension, not as a shortcut to bypass problem solving. The understanding of mathematical concepts is inherently tied to how they can be applied. The long-term benefit of discussing a problem with a lecturer or tutor, or seeking additional support from services such as MASH far outweighs the short term benefit of handing in the correct answer with no development of your understanding. GenAI could be used to create a solution to try and follow and work out where your understanding fails to support these conversations, for example.

Spelling, Punctuation and Grammar

Generative AI can be a very useful tool to help you improve both your general and academic English. It can be used to check the correctness of your written or spoken work, give feedback on your English and as a tool to help you develop your understanding of grammar and improve your vocabulary range. You can also use it to help you prepare for different academic situations where language can be a challenge. 

Below are different ways it can be used to help develop your English skills with some example prompts:

• Checking if a phrase or sentence you wrote is correct in English. Possible prompts: a) ‘Is ‘.........’ correct in English?’, b) can you say ‘........’ in English?’ c) Does ‘.......’ sound natural in English?’.
• Getting linguistic feedback on your writing. Possible prompts: a) Look at the following text and identify the most common types of grammatical errors. Do not correct them, but give me explanations for those grammatical structures  b) Look at this extract from my essay and identify common grammatical and vocabulary errors. Do not correct them, but suggest general improvements I need to make my language clearer.
• Developing your vocabulary: a) What are the most common phrases used in academic writing in this text? [paste in text] b) What are some useful collocations in this text? c) Can you give me a synonym for '----------------------'? d) Explain the word/phrase '-------------------' and give me a variety of examples of when it can be used e) Explain the difference in meaning between xxx and xxx  and give me examples of each word in use.
• Revision and review of vocabulary or grammar points learned. Possible prompts: a) Give me a multiple choice test to check my understanding of the following words: assessment, analysis, research, divergence. Do not give me the answers until I have submitted my guesses b) create a gap fill activity testing my knowledge of the verb collocations that come before these nouns: analysis, argument, research, discussion, assessment. Do not give me the answers until I have submitted my guesses.
• Practise common situations at university where you will need to talk in English. This works particularly well if you can use the mobile app for the AI and use voice recognition to speak your parts. Possible prompts: a)  Imagine I’m a postgraduate engineering student at a university in the UK. I’m having problems starting my dissertation and I have a meeting with my supervisor to discuss this. Play the role of my dissertation supervisor and ask me one question at a time to help me get started. Wait for my response after each question before asking another one.

---

Prompt Engineering

GenAI platforms are often referred to as chatbots and it is tempting to interact with them in a conversational way. If you tend to use ‘please’ and ‘thank you’ in your interactions with GenAI, then you are not alone! However, to produce the most useful outputs, it can be more helpful to think of GenAI as a machine that you are programming by providing a set of detailed prompts. 

Elements that you may want to include in your prompts are as follows:

• Role: what register or style do you want GenAI to adopt in its interactions? For example: Act as an expert in xxx; Write in an academic register.
• Target audience: what sort of style or register do you want GenAI to adopt in its output? For example: Explain xxx to a school student studying xxx. 

  [Example chat here].

• Requirements: the more detail you can provide, the more accurate the output will be. For example, you might want to try listing the main points or perspectives you want addressed: Include arguments both for xxx and against xxx; Incorporate the following points: xxx
• Limitations: provide a guide to how much information you want or any other limitations that you have in mind. For example: Explain xxx in less than 1000 words.
• If you are including a quotation within your prompt, you are more likely to yield accurate results if you delineate the quoted text using speech marks. For example: Act as an expert in xxx. Generate a brief summary (250 words) of the following text. The summary should be accessible to someone with no specialist knowledge of xxx. Text: “xxx”.

Examples of ways to use prompts to help you to use GenAI in an ethical and productive way:

• Breaking down the question: ask GenAI to break down a large or complex question into a number of smaller sub-questions, then ask it to summarise your responses to those sub-questions. Example syntax: Act as an expert in xxx. Ask me a series of questions to explore the question 'xxx' Ask me one question at a time, wait for my response, then ask me the next question. After I have answered the questions, produce a summary for me. 

  [Example chat here].

• Explore counter arguments: provide an argument that you intend to explore and ask GenAI to provide one or more alternative points of view. Do you agree or disagree with these counter arguments? Example syntax: Provide alternative perspectives drawn from a range of cultural contexts to the argument that xxx. 

  [Example chat here].

• As a revision aid: GenAI can help you to revise by testing your knowledge of a given area. Be aware, however that it may not include all of the material covered by your course. Example syntax: Act as a tutor in a first year university xxx course. Ask me multiple choice questions that will test my application of xxx. After I respond, give me feedback on my answer.
• Providing formative feedback: GenAI can provide feedback on your work by identifying points that you may not have covered and prompts for further exploration. Example syntax: Act as an expert in xxx. Provide feedback on the following piece of writing focusing on key points that I may have missed. Text is as follows: “xxx”. 

  [Example chat here].

Using GenAI effectively is ultimately about experimenting with prompts to generate an output that is focused, appropriate and relevant. Often you will need to be patient and use trial and error to find an approach that works for you. And remember, the outputs that you generate should provide a starting point for the development of your own ideas and writing. 

Further reading: Danny Liu, Prompt engineering for students – making generative AI work for you, University of Sydney Teaching Tips, 27 April, 2023

---

Limitations

It is important to remember that the intelligence of GenAI is very different from human intelligence.  It can store and access a lot of data - much more than a human can - but it doesn't understand the information it generates.

GenAI is not like human Intelligence

• When GenAI provides outputs, it is easy to think that these have been confidently created using in-depth subject knowledge. This is not the case: GenAI is utilising the data it has been trained on, without any understanding of the quality or meaning of this data. AI tools cannot 'think': they cannot critically evaluate information to arrive at a conclusion, nor can they apply information to real-world contexts.
• AI does not have emotional intelligence. It cannot understand emotions and has no capacity for empathy.  It can also struggle to pick up on humour, sarcasm and nuance.

GenAI is dependent on the data it is trained on

• Any biases in the original data (or in practices of the data programmers) will be present in GenAI outputs, leading to the reproduction of social inequalities. Outputs may be racist, ableist and/or sexist, for example.
• Certain news organisations have blocked GenAI from scraping their data, which can result in information from particular perspectives being omitted from outputs. For example, the Guardian newspaper announced in September 2023 that they would block OpenAI from using its content. 
• Sources programmed into AI may be months or years out of date, so using GenAI responses may mean that recent key information or developments in the field are missing. For example, Chat GPT is trained on data that only goes up to September 2021, so any information generated since then will be omitted from its output.

GenAI is fallible

• GenAI is occasionally known to generate incorrect or false information, which is sometimes referred to as a ‘hallucination.’ 
• GenAI is not able to reliably distinguish between verified sources and other forms of information, so it may reproduce misleading or false information in its outputs.

Wider impacts

• AI has an environmental impact. It is energy-intensive and as datasets grow, its carbon footprint will get bigger.  

Sources and referencing

• GenAI does not currently have the ability to identify and reference the sources that underpin its assertions. With this limitation in mind, together with the above points around accuracy, it is imperative that any facts or information generated by AI are thoroughly checked using traditional research techniques, such as Library databases or Google Scholar, with references to sources included as appropriate. 

Please visit the 'GenAI and searching' section of the Generative AI Literacy, University of Sheffield Library Guide and the University Library Discovering pages for more information.

---

Unfair means

Unfair means refers to methods used by a student that give an unfair advantage over other students in assessments, including helping someone else to gain an unfair advantage. More information is available on the University of Sheffield’s Unfair Means pages. 

Even if you make a full disclosure of your use of GenAI, if this is not specifically permitted in your assessment criteria and guidance, or extends beyond what is permitted, there is a risk of this use being considered an unfair advantage, and therefore unfair means.

GenAI responses are not held to the same expectations of accuracy, integrity and ethics as students and staff in academia.  If you use responses from GenAI within your work, you could be using the following types of unfair means either knowingly or without even realising:

• Plagiarism - submitting work or ideas that are not your own, and not attributing the work of others through appropriate referencing.  For example, using a GenAI response as if it is your own work, or using an idea given to you by a response, without knowing or referencing where GenAI found the original idea.

• Buying or commissioning work - submitting work that you have requested a service to produce, and submitting it as if it is your own work.  Even if you have used a free GenAI tool, this is still unfair means if you claim it is your own work.

• Coursework sites - the University does not support the use of sites that share content such as lecture notes, essays, lab reports or exam questions. GenAI may be taking its responses directly from these prohibited sites without you knowing.

• Fabrication - submitting work where some information is made up, such as quotations, results of experiments, or survey findings that are not true.  GenAI does not have the capacity to judge whether its sources are objectively true or accurate and can give a convincing response to a question that it has found no answer for - this is sometimes called AI ‘hallucination’.

• Collusion - submitting work that has been created with the help or input of other people as if it is your own work in an assessment that does not permit group work.  If more than one student has been using GenAI responses within their work and these are similar, this could look like collusion.

If you are in any doubt about whether GenAI use is allowed, or if you feel reluctant to admit in your assessed work that you have used Gen AI, then seek clarification from staff in your department.

---

Acknowledge, Describe, Evidence

In assignments where you are sure that you are allowed to use GenAI, you may be asked to provide a full disclosure of how you have done so. This will ensure that your tutor can see how your thinking has intersected with and developed any content that you may have generated. 

You can provide this information by completing an Acknowledge, Describe, Evidence template which you may be asked to submit as an appendix to your assignment.  

1. Acknowledge

Select a statement as appropriate to acknowledge how you used the identified GenAI tool in your work. 

2. Describe

Provide a short summary to describe how you employed the tool and how you adapted or developed its outputs. 

3. Evidence

Provide evidence of how you have used GenAI by providing your prompts and copying and pasting outputs 

Adapted from: UCL, Engaging with AI in your education and assessment (February 2023)

---

Further Reading

Generative AI Literacy, University of Sheffield Library Guide.

Danny Liu, Prompt engineering for students – making generative AI work for you, University of Sydney Teaching Tips, 27 April, 2023

UCL, Engaging with AI in your education and assessment (February 2023)