Strengths and limitations of AI
There will be both opportunities and challenges arising from using generative AI within higher education. Please be aware that the limitations and strengths mentioned are generalised; however, exploration of new technologies is an important part of university learning, and the same can be said for AI, therefore it is important that we all can see how these tools can be effectively and safely embedded at the University.
Strengths
Generative AI can produce a wide range of content, including but not limited to images, text and media based on specific requirements.
AI can be used to help support the development of ideas and the construction of content.
Although we need to be careful with the data we input, AI can help analyse data to provide insights into patterns and areas it discovers.
AI can successfully summarise and feed back information in an easy to digest format.
Limitations
AI can create content quickly, but the quality and accuracy of the data is only as good as the information the AI uses. It can also produce ‘false data’ (also known as hallucinations) so that it can find an answer to the prompt used. The user will therefore need to check the accuracy of the information provided.
There are always ethical concerns around the use of AI, especially when it relates to academic integrity. The way the tool is used within teaching and learning needs to be considered from this standpoint.
AI models only learn from the models which they have available, meaning that existing biases could be perpetuated when returning results. Identifying the data sources and thinking about the wider implication and impact of that data is required.
Each response is only as good as the initial prompt. Writing prompts need to be considered carefully to ensure that the data returned is suitable for purpose and meets requirements.
Whilst some AI tools do specify where sources have come from, these sources still need to be checked and verified as AI tools can create false data sources to make data look more convincing. It should also be clarified that some AI will provide sources which can also be falsified.
Whilst AI models have developed to enable it to look at context in wider capacities, it needs to be acknowledged that the extent and detail to which AI can understand context is still limited and may require human intervention.