[隐私政策] How should educators respond to situations where students submit AI-generated content as their own work?

OpenAI recognizes that many school districts and higher education institutions currently do not consider generative AI in their academic integrity policies. OpenAI also understands that some students may have used these tools in their assignments without disclosing that they used AI. In addition to potentially violating schools’ honor codes, this situation may violate OpenAI’s terms of use: users must be at least 13 years old, and users between 13 and 18 must have permission from a parent or guardian to use the platform.

OpenAI will continue to provide resources and insights in this area, understanding that ultimately each institution will decide how and on what timeline to address these issues in a way that suits their educators and students.

Over the past year, different school districts and universities have developed new policies around AI-generated content. OpenAI encourages educators to do their own research to find the approach that best suits them.

Are AI detectors effective?

In short, based on OpenAI’s experience, no. OpenAI’s research on detectors shows that they are not reliable enough for educators to make judgments about students that may have long-term impacts. Although other developers have released detection tools, OpenAI cannot comment on their utility.

Furthermore, ChatGPT does not “know” what content may have been generated by AI. It sometimes fabricates answers to questions such as “Did you write this [essay]?” or “Could this have been written by AI?” These answers are random and have no factual basis.

To elaborate on OpenAI’s research on the inadequacy of detectors, one major finding by OpenAI is that these tools sometimes consider human-written content to be AI-generated.

When OpenAI attempted to train a detector for AI-generated content, OpenAI found that it labeled human-written texts such as Shakespeare and the Declaration of Independence as AI-generated.

There are also indications that it may disproportionately affect students who are learning or have learned English as a second language, as well as students whose writing is particularly formulaic or concise.

Even if these tools could accurately identify AI-generated content (which they currently cannot), students could make minor modifications to evade detection.

However, some people have found the following methods useful:

Some teachers have found it useful to encourage students to share specific conversations with ChatGPT (here). This approach has many benefits:

Showing their work and formative assessment:

Educators can analyze students’ interactions with ChatGPT and observe their critical thinking and problem-solving approaches.

Sharing links can enable students to peer-review each other’s work, fostering a collaborative environment.

By recording their conversations with AI, students can reflect on their own progress over time. They can see how their skills in asking questions, analyzing responses, and integrating information develop. Teachers can also use these records to provide personalized feedback and support individual growth.

Information and AI literacy:

Students can demonstrate their ability to interact with AI and their understanding of the shortcomings of AI systems. Educators can assess the quality of the questions asked, the relevance of the information obtained, and students’ ability to challenge that information, double-check it, and consider potential biases.

OpenAI foresees that in the future the use of AI tools such as ChatGPT will be commonplace. Encouraging responsible use helps prepare students for situations in the future where they may need to leverage AI in different environments.

Creating accountability:
Sharing interactions with the model ensures that students are accountable for how they use AI. Educators can verify whether students are using the tool responsibly and meaningfully, rather than simply copying answers.