Using Generative AI

Example of bad AI generated image.

The main rule

We demand the same quality standards

Guidelines for using especially generative AI systems are based on the following:

• Universities' obligation to seek the truth
• Existing regulatory directives for privacy, consent, marketing, PR and journalistic practices
• Quality requirements for the finished products that are equivalent to the quality expected by the respective academic fields (author/graphic designer/photographer/illustrator) before generative AI systems. The quality of images should at least be on par with images and illustrations that can be purchased through online image archives.
• A human-centric approach, as defined in the EU's Artificial Intelligence Act (text adopted December 2023)
• Workshop with the communication organisation, 18 December 2023
• The University of Southern Denmark's Corporate Visual Identity (CVI)
• CO2 emissions. Some AI systems emit the equivalent of 1 smartphone charge per prompt.

Generative AI are systems that behave like humans and can autonomously produce a new image, text or combine images, text and audio into an autonomous output.

AI logic systems are implemented as assistants in various programmes today, including image and text software. These systems don’t generate new things – they optimise. They can and may be used for general editing, e.g. to create effects (toning, blurring), expanding images or removing backgrounds/isolating subjects.

Image: Inspired by AI-generated images: "Slumdog Millionaire"

Who is responsible?

Be aware of copyright

Since generative AI systems create images and text from pre-existing original images and text, special attention should be paid to the following:

• Original text or images may have been illegally collected through surveillance
• Original text or images may be protected by copyright
• Subjects or wording can be heavily biased, manipulative, fake or otherwise contain non-validated utterance
• Original texts, including research papers, contain knowledge the author does not necessarily want to feed into generative AI systems
• Generative AI systems can be hacked like any other system - and content protected by the GDPR regulation is particularly vulnerable
• Generative AI systems often reside in the cloud, so ask yourself: 'Who's really behind it?' - and don't share everything with everyone
• Dont't forget to read the fine print, too. For example, are you required to give up your own input (data) as payment to get something in return?

It is the responsibility of the individual employee to ensure that text, images and audio comply with copyright laws and do not discriminate, manipulate or distort the truth. In other words: If, as employees, we publish or disseminate

AI-generated content, the employee is liable for the outcome if the content violates legislation and directives or otherwise compromises SDU.

Generative AI systems as an assistant

Recommendations for good use

• For optimising automated processes
• As a creative assistant and feedback partner in the idea development phase
• For versions of the same basic concept, e.g. advertising copy
• For shorter generic texts such as advertising copy
• For illustrations where these comply with SDU’s CVI and prior concept work has taken place
• For photography, where these comply with SDU’s CVI and prior concept work has taken place
• To enrich already published web pages’ metadata (keywords, description) and teaser text(s) from a search engine optimisation perspective

Dont's

What Generative AI Systems are not used for

• We do not publish creative output (audio, images or text) if there is uncertainty about the copyright of the original images used or where people or context have been manipulated without consent.
• We do not publish photographs of people and locations that purport to be realistic documentation of events that took place but never happened.
• The communications organisation (journalists, communications consultants and marketing staff) does not publish text, including press releases, op-eds, profiles, longer opinion pieces on social media or entire campaigns without checking copyrights. Detected factual errors or possible biases. For long texts, a declaration will state if they are primarily composed using artificial intelligence.
• We will also continue to avoid images that go against our intention to be authentic in expression and that have the appearance of being a collage of stock photos. This guideline also attaches great importance to SDU's CVI, quality requirements for images and illustrations in general, and consideration for a market currently flooded with the same artificial AI-generated images.

Five tips for users of machine translation

1. Remenber that linguitic errors in the source text, such as missing words or syntax errors, may be reflected in the machine translation. You should therefore quality assure both the source text and the translation.
2. AI language models translate individual words, not the overall meaning, so be in the lookout for incorrect translations.
3. Always tell the recipient if your text has been machine translated. As a producer, you are responsible for your text - including when it is machine translated.
4. SDU has its own list of approved SDU terms. AI language models are not familiar with SDU's term list, so you need to find the list (see the Language Services website) and enter the terms yourself.
5. You must still comply with current legislation. Therefore, be careful when you feed text content into certain translation programmes. There may be confidential data, research results or other GDPR considerations you need to take into account.

Therefore: Machine translation is very useful for short text, such as emails and messages - but think twice about using it for long text, such as research articles and news items.

AI HUB

An AI HUB will be set up in SDU Communications to qualify in generating high-quality text and creatives in accordance with truth criteria and SDU's CVI. CVI is a design system for the organisation’s visual identity. The system regulates the use of colours, fonts, logos and images, and it defines the tone of voice, illustration style and image style. Our CVI is SDU’s visual representation of the brand – what we stand for and what we are known and remembered for.

The AI HUB is set up so that SDU can reap all the benefits of generating text, images and illustrations with generative AI systems without compromising truth or quality. Put differently, SDU Communications does not want to slow down development, but neither does it want to replace qualified professionals to develop texts and creatives or to provide advice and find images. Just because anyone with some knowledge of AI systems can visualise an idea doesn’t mean that the visualisation is good. It takes a long time and often a professional understanding to achieve valid expressions.