Generative AI
Almost everyone in higher education is talking about ChatGPT—but what is it and how does it work? In simple terms, ChatGPT uses generative AI technology, which is a form of artificial intelligence (AI) that specializes in producing new content. From images and text to music and videos, generative AI utilizes existing data to train models that recognize patterns and sequences. With this information tools are able to generate new content that mirrors the original data—in the case of ChatGPT this takes the form of text-based responses to user prompts.
Generative AI Tools - A Closer Look
Since generative AI services such as ChatGPT excel at mimicking human-generated content, there are opportunities and risks in the application of these emerging technologies. It can be difficult to distinguish the outputs of these tools from human-generated content, so concerns in higher education center on potential risks to academic integrity.
Generative AI relies on advanced techniques like deep learning and generative models such as Generative Adversarial Networks (GANs) or Variational Autoencoders (VAEs). GANs feature two critical components – the generator and the discriminator. The generator creates new content, while the discriminator evaluates its authenticity against real data. By constantly iterating and improving the model, the generator and discriminator produce increasingly realistic and high-quality generated content.
Since new AI tools and services are developed almost daily, it is difficult to maintain a comprehensive list of examples. However, the following link provides a curated list of examples, including ones that come up in our conversations with faculty.
AI Detection Tools
Advancements in generative AI technologies are outpacing development of tools designed to detect their usage. While numerous AI detection tools (e.g., Turnitin, GPTZero, Copyleaks, and Originality AI) are available, testing indicates a wide range in the accuracy and efficacy of these tools—including missed AI content detection and false positives.
Although all WSE assignments in Canvas can be processed with TurnItIn (TII), JHU and TII make no claims to the accuracy of this tool. Using TurnItIn may provide useful insights, but we encourage faculty to be aware of the challenges and limitations.
Also keep in mind that there are many strategies for avoiding detection, including tools designed for this purpose (e.g., Undetectable AI), as well as simply running AI generated content through a second generative AI tool.