Licensed Generative AI tools at UoB

When it comes to the use of AI in teaching and learning, we are still in the exploratory stage in higher education with best practice and emerging use cases. The applicability of AI is also heavily dependent on subject area/discipline and what licensed AI tools are available to an institution.

Introducing Microsoft Copilot

What was formerly known as Bing Chat or Bing AI has been renamed as Microsoft Copilot. To make things more confusing, there are multiple versions of Microsoft Copilot. Under our University of Brighton institutional license, we have the version of Copilot which refers to the AI chatbot. This is accessible to staff and students using their university logins.

The Copilot chatbot is powered by GPT4, the latest AI model, the same as the premium, paid version of ChatGPT. It is multimodal, allowing for AI image generation using DALL-E3. Additionally, it allows the Bing search index in real time to provide answers with citations to sources for extra transparency.

The main advantage of using our institutional Copilot generative AI tool rather than a 3^rd party tool such as ChatGPT is down to data protection. Use of Copilot is under commercial data protection regulations – this means that neither we as an institution or Microsoft have eyes-on access to anything that you input and your user account is protected. Additionally, chat data will not be used to train future AI language models. Due to this commercial data protection, chat history is currently not supported, meaning you will not be able to access past conversations once you have logged out of Copilot (or your log in session has timed out). Therefore, we would recommend copying or exporting any prompts or conversations before your next session.

Copilot Pro (for individuals) and Copilot 365 (for business) additionally includes Copilot (the AI assistant) in Word, Excel, PowerPoint, OneNote, Outlook and Teams. The latter is currently prohibitively expensive to roll-out institution-wide.

To log in:

Go to: http://copilot.microsoft.com and click ‘Sign in with a work or school account’ and then log in with your UoB credentials.

Once logged in you will be able to check that you are using your university account (rather than a personal Microsoft account) by ensuring your university email is displayed in the top-right of the screen.

Prompting

Copilot is similar to other AI models in that it is a conversational agent that works off prompts that you input. Good prompting is therefore an important skill that will help tailor and develop the outputs you receive. See some top tips below:

• Make sure to provide enough context to your prompts, for example include your role and the context or audience e.g., foundation level students. Give detailed instructions and specific requirements for the output.
• Don’t work off a singular prompt. You likely won’t get what you want on the first try and therefore part of working with Copilot is to refine and reframe prompts for the best output.
• Copilot will remember all prompts within a conversation. Therefore, create a new chat for a new concept or conversation to start ‘afresh’ <- NOTE that when starting a new chat your current chat history will be wiped so use the copy/export function before doing so.

Image generation

Alongside any form of text-based output, Copilot can generate images using the same chatbox. Copilot utilises the latest version of AI image chatbot, DALL-E 3 (developed by OpenAI) in order to create a series of suggested images based on a single prompt.

When designing a prompt for image creation in Copilot you will need to explicitly state the desired content using descriptive sentences. For example, you could ask for a “photorealistic image of a house” or an “architectural plan of a house”. CoPilot (using DALL-E) is capable of rendering different types of images, ranging from simple cartoons to photo-like quality. The more description you use in your prompting, the more control you will have over the output. Copilot will offer four alternative images as the output. You can download and use these images.

As a rule, text does not generate accurately within an image. You will likely find (if you ask, as part of your prompt, for text to display) that the text is distorted.

While OpenAI’s terms state that you own the copyright for any images created using DALL-E, and therefore Copilot, bear in mind that elements of copyrighted images and trademarks might appear. However, you can use these images for teaching purposes.

Other AI features in University of Brighton licensed tools

Aside from Microsoft’s new generative AI chatbot Copilot, more and more AI features are being added to our existing range of digital tools:

Padlet

‘I can’t draw’ is a new feature that allows you and your students to generate AI images on an existing Padlet board. Students simply click on the I can’t draw option, write a prompt and choose their generated image from multiple created. The tool moderates inappropriate content.

Use this tool to create visual art showcases or galleries. See Padlet’s example here: Watercolor art showcase

PowerPoint

PowerPoint has had several advanced AI features added to the tool. The in-house Designer function uses AI to make design suggestions for each existing slide to better fit the information. It recognises text and images and suggests many layout choices and allows you to adapt colour schemes and other visual elements.

Rehearse with Coach is a way to get feedback as you practice a presentations. The AI tool analyses pacing, your use of filler words and gives you tips to improve your delivery.

LinkedIn Learning

LinkedIn Learning is accessible for all UoB staff and students and contains a wealth of high-quality video tutorials on a range of skillsets and software and includes professional certificates. Now they have released AI-powered coaching. The chatbot answers your questions in real-time, asking follow up questions to provide personalised insights and linking to relevant courses and information.

Generative AI imagery powered by Firefly is now embedded in many of Adobe’s design tools including Illustrator, Photoshop and Adobe Express. Use LinkedIn Learning to find bespoke courses that train you to use these features:

3^rd party AI tools

There are a substantial number of 3^rd party (unlicensed) tools however great care and consideration needs to be undertaken for their responsible and ethical usage.

If you wanted to use a 3^rd party tool in your teaching you need to adhere to the following:

• Avoid any upload of university personal data (GDPR rules apply)​
• Student participation cannot require tool registration ​
• Make sure you have read a tool’s privacy policy (impact on intellectual property and copyright)​
• Consider inclusivity and accessibility limitations​

For University of Brighton staff, any new third-party AI tool will need to be requested via the IT Service Desk through their procurement process and checked against service terms and security etc.​ If a third-party tool requires personal data (ie., your UoB email etc.) which it most likely will, a DPIA (Data Protection Impact Assessment) will need to be raised with the Data Protection Team.

Challenges and limitations

Generally, any use of AI should be fair, transparent, lawful, accurate and accountable. However, this can be challenging due to the nature of the technology and the following should be acknowledged:

• Generated outputs may contain inaccuracies and errors as they are often not based on a specific source. Some sources may be fabricated, known as ‘hallucinated referencing’. Outputs need to be looked at critically and fact checked.
• AI outputs have data and algorithmic biases – they can be discriminatory and unethical. Think about how you can encourage the outputs to display a more inclusive view.
• Remember, AI use must adhere to data protection laws. You and your students must avoid the upload of university personal data as standard!
• We should also consider the impact on copyright and intellectual property – AI can create convincing replicas of copyrighted works such as text, art, music. It is difficult to enforce and protect intellectual property rights. There are also ethical debates surrounding exploitative labour used in its development and the ongoing negative environmental impact.

University of Brighton staff can visit the Artificial Intelligence SharePoint pages for more information and awareness.

Authors: Emilie Hayter and Nick Feather (Digital Learning Team).