Key take-aways

Challenges of using GenAI for simulations

Limited outputs

In the early days of GenAI, a common challenge was the AI’s reluctance to generate hypothetical news stories or political statements. When prompted to create such content, these AI models would often push back, reminding the user that the requested information was not based on actual events. This limitation made it challenging to leverage GenAI for simulations, thought experiments, or creative writing exercises that relied on fabricated scenarios. However, as GenAI has evolved, developers have found ways to work around this issue. One effective approach is to explicitly define the nature of the project upfront, framing it as a simulation. By providing this context in the project header or introduction, users can help the AI understand that the subsequent prompts will involve hypothetical situations.

Conflict-avoiding bias

The AI setting is designed to avoid conflicts and lead to compromise, often favouring cooperative resolutions even in intense conflict situations. This meant that GenAI tended to avoid escalatory statements and instead steered interactions toward collaboration. As a result, careful prompt management and some output editing were necessary to ensure that the messages and information maintained an appropriate tone for the simulation’s context.

Limited image production

The AI tools used (ChatGTP, Copilot) could not produce highly realistic images. This limitation led to using existing media photos instead of AI-generated visuals to illustrate news stories. This could partly be due to my limited skills in image creation or the significant time required to achieve the desired outcome. Specialised AI tools likely focused on generating visuals would be more effective. However, creating realistic images remains an area of ongoing development.

Some recommendations for implementation

Enhance prompt writing

The most essential element is solid prompt writing. The AI tool should be provided with detailed background knowledge to assist the teacher in the simulation. Share your scenario objectives, define the tonality, planned outcome, the genre of the text, and the perspective on whose behalf the text is composed to ensure the most accurate and relevant outcome.

Regular reviewing

AI-generated outputs should be regularly reviewed and adjusted. The ChatGPT canvas function enables optimal improvement because it allows you to add your own text to the existing output or request adjustments to specific sections. This flexibility makes it easier to fine-tune the output according to your needs. Overall, it is essential to avoid quickly copying and pasting AI-generated content and instead critically assess the output to maintain the desired pedagogical focus.

Integration of various sources

If the AI does not have other input, the outcomes tend to follow similar templates. To adjust the style, genre, and authenticity of the output, it is essential to integrate different sources. For example, providing press releases from a government’s communication as examples can guide the AI in producing the desired style. Similarly, news stories of actual events can be referenced to contextualise and adapt existing events for a new purpose. These inputs help create outputs that are more nuanced and tailored to the specific needs of the simulation.